Contoh Jurnal Scopus mengenai Neural network models used by machine learning researchers

 

 

Deep Sparse Rectifier Neural Networks

Xavier Glorot, Antoine Bordes and Yoshua Bengio

Dept. IRO

University of Montreal

Montreal, QC

glorotxa,bordesa,bengioy@iro.umontreal.ca

Abstract

Rectifying neurons are more biologically plausible than logistic sigmoid neurons, which are themselves more biologically plausible than hyperbolic tangent neu¬rons. However, the latter work better for training multi-layer neural networks than logistic sigmoid neurons. This paper shows that networks of rectifying neu¬rons yield equal or better performance than hyperbolic tangent networks in spite of the hard non-linearity and non-differentiability at zero and create sparse rep¬resentations with true zeros which are remarkably suitable for naturally sparse data. Even though they can take advantage of semi-supervised setups with extra-unlabeled data, deep rectifier networks can reach their best performance without requiring any unsupervised pre-training on purely supervised tasks with large la¬beled datasets. Hence, these results can be seen as a new milestone in the attempts at understanding the difficulty in training deep but purely supervised neural net¬works, and closing the performance gap between neural networks learnt with and without unsupervised pre-training.

1 Introduction

Many differences exist between the neural network models used by machine learning researchers and those used by computational neuroscientists. This is in part because the objective of the former is to obtain computationally efficient learners, that generalize well to new examples, whereas the objective of the latter is to abstract out neuroscientific data while obtaining explanations of the prin¬ciples involved, providing predictions and guidance for future biological experiments. Areas where both objectives coincide are therefore particularly worthy of investigation, pointing towards compu¬tationally motivated principles of operation in the brain that can also enhance research in artificial intelligence. In this paper we show that two common gaps between computational neuroscience models and machine learning neural network models can be bridged by using the following linear by part activation: max(0, x), called the rectifier (or hinge) activation function. Experimental results will show engaging training behavior of this activation function, especially for deep architectures (see Bengio (2009) for a review), i.e., where the number of hidden layers is 3 or more.

Recent theoretical and empirical work in statistical machine learning has demonstrated the impor¬tance of learning algorithms for deep architectures. This is in part inspired by observations of the mammalian visual cortex, which consists of a chain of processing elements, each of which is asso¬ciated with a different representation of the raw visual input. This is particularly clear in the primate visual system (Serre et al., 2007), with its sequence of processing stages: detection of edges, primi¬tive shapes, and moving up to gradually more complex visual shapes. Interestingly, it was found that the features learned in deep architectures resemble those observed in the first two of these stages (in areas V1 and V2 of visual cortex) (Lee et al., 2008), and that they become increasingly invariant to factors of variation (such as camera movement) in higher layers (Goodfellow et al., 2009).

 

1

 

 

Figure 1: Left: common activation function motivated by biological data. Right: commonly used activation functions in neural networks literature: logistic sigmoid and hyperbolic tangent (tanh).

Regarding the training of deep networks, something that can be considered a breakthrough happened in 2006, with the introduction of Deep Belief Networks (Hinton et al., 2006), and more generally the idea of initializing each layer by unsupervised learning (Bengio et al., 2007; Ranzato et al., 2007). Some authors have tried to understand why this unsupervised procedure helps (Erhan et al., 2010) while others investigated why the original training procedure for deep neural networks failed (Ben-gio and Glorot, 2010). From the machine learning point of view, this paper brings additional results in these lines of investigation.

We propose to explore the use of rectifying non-linearities as alternatives to the hyperbolic tangent or sigmoids in deep artificial neural networks, in addition to using an L1 regularizer to promote spar¬sity and prevent potential numerical problems with unbounded activation. Nair and Hinton (2010) present promising results of the influence of such units in the context of Restricted Boltzmann Ma¬chines compared to logistic sigmoid activations. Our work extends this for the case of pre-training using denoising auto-encoders (Vincent et al., 2008) and provides an extensive empirical comparison of the rectifying activation function against the hyperbolic tangent on image classification bench¬marks as well as an original derivation for the text application of sentiment analysis.

Our experiments indicate that training proceeds better when the artificial neurons are either off or operating mostly in a linear regime. Surprisingly, rectifying activation allows deep networks to achieve their best performance even without unsupervised pre-training. Hence, our work proposes a new contribution to the trend of understanding and merging the performance gap between deep networks learnt with and without unsupervised pre-training (Erhan et al., 2010; Bengio and Glorot, 2010). Interestingly, rectifier networks can still benefit from unsupervised pre-training in the context of semi-supervised learning where large amounts of unlabeled data are provided. Furthermore, as rectifier units naturally lead to sparse networks and are closer to biological neurons’ responses in their main operating regime, this work also bridges (in part) a machine learning / neuroscience gap in terms of activation function and sparsity.

2 Background

2.1 Neuroscience Observations

For models of biological neurons, the activation function is the expected firing rate as a function of the total input currently arising out of incoming signals at synapses (Dayan and Abott, 2001). An activation function is termed, respectively antisymmetric or symmetric when its response to the opposite of a strongly excitatory input pattern is respectively a strongly inhibitory or excitatory one, and one-sided when this response is zero. The main gaps that we wish to consider between computational neuroscience models and machine learning models are the following.

Studies on brain energy expense suggest that neurons encode information in a sparse and distributed way (Attwell and Laughlin, 2001), estimating the percentage of neurons active at the same time to be between 1 and 4% (Lennie, 2003). This corresponds to a trade-off between richness of represen¬tation and small action potential energy expenditure. Without additional regularization, such as an L1 penalty, ordinary feedforward neural nets do not have this property. For example, the sigmoid activation has a steady state regime around 21, therefore, after initializing with small weights, all neu 

2

 

rons fire at half their saturation regime. This is biologically implausible and hurts gradient-based optimization (LeCun et al., 1998; Bengio and Glorot, 2010).

Important divergences between biological and machine learning models concern non-linear activa¬tion functions. The leaky integrate-and-fire (or LIF) (Dayan and Abott, 2001), gives the following relation between the firing rate and the input current, illustrated in Figure 1 (left):

 

 

where tref is the refractory period (minimal time between two action potentials), I the input current, Vr the resting potential and Vth the threshold potential (with Vth > Vr), and R, E, τ the membrane resistance, potential and time constant. The most commonly used activation functions in the deep learning and neural networks literature are the standard logistic sigmoid and the hyperbolic tangent (see Figure 1, right). The hyperbolic tangent has a steady state at 0, and is therefore preferred from the optimization standpoint (LeCun et al., 1998; Bengio and Glorot, 2010), but it forces an antisymmetry around 0 which is absent in biological neurons.

2.2 Advantages of Sparsity

Sparsity has become a concept of interest, not only in computational neuroscience and machine learning but also in statistics and signal processing (Candes and Tao, 2005). It was first introduced in computational neuroscience in the context of sparse coding in the visual system (Olshausen and Field, 1997). It has been a key element of deep convolutional networks exploiting a variant of auto-encoders (Ranzato et al., 2007, 2008; Mairal et al., 2009) with a sparse distributed representation, and has also become a key ingredient in Deep Belief Networks (Lee et al., 2008). A sparsity penalty has been used in several computational neuroscience (Olshausen and Field, 1997; Doi et al., 2006) and machine learning models (Lee et al., 2007; Mairal et al., 2009), in particular for deep architec¬tures (Lee et al., 2008; Ranzato et al., 2007, 2008). However, in the latter, the neurons end up taking small but non-zero activation or firing probability. We show here that using a rectifying non-linearity gives rise to real zeros of activations and thus truly sparse representations. From a computational point of view, such representations are appealing for the following reasons:

• Information disentangling. One of the claimed objectives of deep learning algorithms (Bengio, 2009) is to disentangle the factors explaining the variations in the data. A dense representation is highly entangled because almost any change in the input modifies most of the entries in the representation vector. Instead, if a representation is both sparse and robust to small input changes, the set of non-zero features is almost always roughly conserved by small changes of the input.

• Efficient variable-size representation. Different inputs may contain different amounts of infor¬mation and would be more conveniently represented using a variable-size data-structure, which is common in computer representations of information. Varying the number of active neurons allows a model to control the effective dimensionality of the representation for a given input and the required precision.

• Linear separability. Sparse representations are also more likely to be linearly separable, or more easily separable with less non-linear machinery, simply because the information is represented in a high-dimensional space. Besides, this can reflect the original data format. In text-related applications for instance, the original raw data is already very sparse (see Section 4.2).

• Distributed but sparse. Dense distributed representations are the richest representations, being potentially exponentially more efficient than purely local ones (Bengio, 2009). Sparse represen¬tations’ efficiency is still exponentially greater, with the power of the exponent being the number of non-zero features. They may represent a good trade-off with respect to the above criteria.

Nevertheless, forcing too much sparsity may hurt predictive performance for an equal number of neurons, because it reduces the effective capacity of the model.

3

 

 

Figure 2: Left: sparse propagation of activations and gradients in a network of rectifier units. The input selects a subset of active neurons and computation is linear in this subset. Right: The rectifier and the softplus activation functions, the second one is a smooth version of the first.

3 Deep Rectifier Networks

3.1 Rectifier Neurons

The neuroscience literature (Bush and Sejnowski, 1995; Douglas and al., 2003) indicates that corti¬cal neurons are rarely in their maximum saturation regime, and suggests that their activation func¬tion can be approximated by a rectifier. Most previous studies of neural networks involving a recti¬fying activation function concern recurrent networks (Salinas and Abbott, 1996; Hahnloser, 1998). The rectifier function rectifier(x) = max(0, x) is one-sided and therefore does not enforce a sign symmetry1 or antisymmetry1: instead, the response to the opposite of an excitatory input pattern is 0 (no response). However, we can obtain symmetry or antisymmetry by combining two rectifier units sharing parameters.

Advantages The rectifier activation function allows a network to easily obtain sparse represen¬tations. For example, after uniform initialization of the weights, around 50% of hidden units con¬tinuous output values are real zeros, and this fraction can easily increase with sparsity-inducing regularization. Apart from being more biologically plausible, sparsity also leads to mathematical advantages (see previous section).

As illustrated in Figure 2 (left), the only non-linearity in the network comes from the path selection associated with individual neurons being active or not. For a given input only a subset of neurons are active. Computation is linear on this subset: once this subset of neurons is selected, the output is a linear function of the input (although a large enough change can trigger a discrete change of the active set of neurons). The function computed by each neuron or by the network output in terms of the network input is thus linear by parts. We can see the model as an exponential number of linear models that share parameters Nair and Hinton (2010). Because of this linearity, gradients flow well on the active paths of neurons (there is no gradient vanishing effect due to activation non-linearities of deep networks of sigmoid or tanh units), and mathematical investigation is easier. Computations are also cheaper: there is no need for computing the exponential function in activations, and sparsity can be exploited.

Potential Problems One may hypothesize that the hard saturation at 0 may hurt optimization by blocking gradient back-propagation. To evaluate the potential impact of this effect we also investi¬gate the softplus activation: softplus(x) = log(1 + ex) (Dugas et al., 2001), a smooth version of the rectifying non-linearity. We lose the exact sparsity, but may hope to gain easier training. However, experimental results (see Section 4.1) tend to contradict that hypothesis, suggesting that hard zeros can actually help supervised training. We hypothesize that the hard non-linearities do not hurt so long as the gradient can propagate along some paths, i.e., that some of the hidden units in each layer are non-zero. With the credit and blame assigned to these ON units rather than distributed

1The hyperbolic tangent absolute value non-linearity  tanh(x) used by Jarrett et al. (2009) enforces sign symmetry. A tanh(x) non-linearity enforces sign antisymmetry.

 

4

 

more evenly, we hypothesize that optimization is easier. Another problem could arise due to the unbounded behavior of the activations; one may thus want to use a regularizer to prevent potential numerical problems. Therefore, we use the L1 penalty on the activation values, which also pro¬motes additional sparsity. Also recall that, in order to efficiently represent symmetric/antisymmetric behavior in the data, a rectifier network would need twice as many hidden units as a network of symmetric/antisymmetric activation functions.

Finally, rectifier networks are subject to ill-conditioning of the parametrization. Biases and weights can be scaled in different (and consistent) ways while preserving the same overall network function. More precisely, consider for each layer of depth i of the network a scalar αi, and scaling the parame-s

ters as W~i = Wi

αi and b~i = bi 

H

i . The output units values then change as follow: s' = ~n

j=1 αj j=1 αj .

Therefore, as long as jnj=1 αj is 1, the network function is identical.

3.2 Unsupervised Pre-training

This paper is particularly inspired by the sparse representations learned in the context of auto-encoder variants, as they have been found to be very useful in training deep architectures (Bengio, 2009), especially for unsupervised pre-training of neural networks (Erhan et al., 2010).

Nonetheless, certain difficulties arise when one wants to introduce rectifier activations into stacked denoising auto-encoders (Vincent et al., 2008). First, the hard saturation below the threshold of the rectifier function is not suited for the reconstruction units. Indeed, whenever the network happens to reconstruct a zero in place of a non-zero target, the reconstruction unit can not backpropagate any gradient.2 Second, the unbounded behavior of the rectifier activation also needs to be taken into account. We have experimented with several strategies, and propose the following two (the first strategy has proven to be the most efficient on image experiments and the second one on text ones):

1. Use a softplus activation function for the reconstruction layer, along with a quadratic re¬construction cost.

2. Scale the rectifier activation values coming from the previous encoding layer to bound them between 0 and 1, then use a sigmoid activation function for the reconstruction layer, along with a cross-entropy reconstruction cost.

4 Experimental Study

4.1 Image Recognition

Experimental setup We considered the image datasets detailed below. Each of them has a train¬ing set (for tuning parameters), a validation set (for tuning hyper-parameters) and a test set (for reporting generalization performance). They are presented according to their number of train-ing/validation/test examples, their respective image sizes, as well as their number of classes:

• MNIST (LeCun et al., 1998): 50k/10k/10k, 28 × 28 digit images, 10 classes.

• CIFAR10 (Krizhevsky and Hinton, 2009): 50k/5k/5k, 32 × 32 × 3 RGB images, 10 classes.

• NISTP: 81920k/80k/20k, 32 × 32 character images from the NIST database 19, with randomized distortions (Anonymous, 2010), 62 classes. This dataset is much larger and more difficult than the original NIST (Grother, 1995).

• NORB: 233172/58428/58320, taken from Jittered-Cluttered NORB (LeCun et al., 2004). Stereo-pair images of toys on a cluttered background, 6 classes. The data has been preprocessed similarly to (Nair and Hinton, 2010): we subsampled the original 2 × 108 × 108 stereo-pair images to 2 × 32 × 32 and scaled linearly the image in the range [1,1]. We followed the procedure used by Nair and Hinton (2010) to create the validation set.

For all experiments except on the NORB data (LeCun et al., 2004), the models we used are stacked denoising auto-encoders (Vincent et al., 2008) with three hidden layers and 1000 units per layer.

2Why is this not a problem for hidden layers too? we hypothesize that it is because gradients can still flow through the active (non-zero), possibly helping rather than hurting the assignment of credit.

 

5

 

Table 1: Test error on networks of depth 3. Bold results represent statistical equivalence between similar experiments, with and without pre-training, under the null hypothesis of the pairwise test with P = 0.05.

Neuron MNIST CIFAR10 NISTP NORB

With unsupervised pre-training

Rectifier 1.20% 49.96% 32.86% 16.46%

Tanh 1.16% 50.79% 35.89% 17.66%

Softplus 1.17% 49.52% 33.27% 19.19%

Without unsupervised pre-training

Rectifier 1.43% 50.86% 32.64% 16.40%

Tanh 1.57% 52.62% 36.46% 19.29%

Softplus 1.77% 53.20% 35.48% 17.68%

The architecture of Nair and Hinton (2010) has been used on NORB: two hidden layers with respec¬tively 4000 and 2000 units. We used a cross-entropy reconstruction cost for tanh networks and a quadratic cost over a softplus reconstruction layer for the rectifier and softplus networks. We chose masking noise as the corruption process: each pixel has a probability of 0.25 of being artificially set to 0. The unsupervised learning rate is constant, and the following values have been explored: .1, .01, .001, .0001. We select the model with the lowest reconstruction error. For the super¬vised fine-tuning we chose a constant learning rate in the same range as the unsupervised learning rate with respect to the supervised validation error. The training cost is the negative log likelihood  log P (correct classinput) where the probabilities are obtained from the output layer (which im¬plements a softmax logistic regression). We used stochastic gradient descent with mini-batches of size 10 for both unsupervised and supervised training phases.

To take into account the potential problem of rectifier units not being symmetric around 0, we use a variant of the activation function for which half of the units output values are multiplied by -1. This serves to cancel out the mean activation value for each layer and can be interpreted either as inhibitory neurons or simply as a way to equalize activations numerically. Additionally, an L1 penalty on the activations with a coefficient of 0.001 was used during pre-training and fine-tuning in order to increase the amount of sparsity in the learned representations.

Main results Table 1 summarizes the results on networks of 3 hidden layers of 1000 hidden units each, comparing all the neuron types on all the datasets, with or without unsupervised pre-training. In the latter case, the supervised training phase has been carried out using the same experimental setup as the one described above for fine-tuning. The main observations we make are the following:

• Despite the hard threshold at 0, networks trained with the rectifier activation function can find local minima of greater or equal quality than those obtained with its smooth counterpart, the softplus. Rectifiers are not only biologically plausible, they are also computationally efficient.

• There is almost no improvement when using unsupervised pre-training with rectifier activations, contrary to what is experienced using tanh or softplus. Purely supervised rectifier networks re¬main competitive on all 4 datasets, even against the pretrained tanh or softplus models.

• Rectifier networks are truly deep sparse networks. There is an average exact sparsity (fraction of zeros) of the hidden layers of 83.40% on MNIST and 72.00% on CIFAR10. Figure 3 (left) provides a better understanding of the influence of sparsity. It displays the MNIST test error of deep rectifier networks (without pre-training) according to different average sparsity obtained by varying the L1 penalty on the activations. Networks appear to be quite robust to it as models with 70% to almost 85% of true zeros can achieve similar performances.

With labeled data, deep rectifier networks appear to be attractive models. They are biologically cred¬ible, and, compared to their standard counterparts, do not seem to depend as much on unsupervised pre-training, while ultimately yielding sparse representations.

This last conclusion is slightly different from those reported in (Nair and Hinton, 2010) in which is demonstrated that unsupervised pre-training with Restricted Boltzmann Machines and using rectifier units is beneficial. In particular, the paper reports that pre-trained rectified Deep Belief Networks can achieve a test error on NORB below 16%. However, we believe that our results are compatible with those: we extend the experimental framework to a different kind of models (stacked denoising auto-encoders) and different datasets (on which conclusions seem to be different). Furthermore,

 

6

 

Figure 3: Left: influence of final sparsity on accuracy. 200 randomly initialized deep rectifier networks were trained on MNIST with various L1 penalties (from 0 to 0.001) to obtain different sparsity levels. Right: effect of unsupervised pre-training. On NORB, we compare hyperbolic tangent and rectifier networks, with or without unsupervised pre-training, and fine-tune only on subsets of increasing size of the training set.

note that our rectified model without pre-training on NORB is very competitive (16.4% error) and outperforms the 17.6% error of the non-pretrained model from Nair and Hinton (2010), which is basically what we find with the non-pretrained softplus units (17.68% error).

Semi-supervised setting Figure 3 (right) presents results of semi-supervised experiments con¬ducted on the NORB dataset. We vary the percentage of the original labeled training set which is used for the supervised training phase of the rectifier and hyperbolic tangent networks and evaluate the effect of the unsupervised pre-training (using the whole training set, unlabeled). Confirming conclusions of Erhan et al. (2010), the network with hyperbolic tangent activations improves with unsupervised pre-training for any labeled set size (even when all the training set is labeled). How¬ever, the picture changes with rectifying activations. In semi-supervised setups (with few labeled data), the pre-training is highly beneficial. But the more the labeled set grows, the closer the models with and without pre-training. Eventually, when all available data is labeled, the two models achieve identical performance. Rectifier networks can maximally exploit labeled and unlabeled information.

4.2 Sentiment Analysis

In text-related applications, data is usually very sparse. Deep rectifier networks, thanks to their naturally sparse behavior, could be an interesting match for this kind of learning task. To validate this intuition, we consider sentiment analysis, a text mining area which aims to determine the judgment of a writer with respect to a given topic (see (Pang and Lee, 2008) for a review). The basic task consists in classifying the polarity of reviews either by predicting whether the expressed opinions are positive or negative, or by assigning them star ratings on either 3, 4 or 5 star scales.

Following a task originally proposed by Snyder and Barzilay (2007), our data consists of restaurant reviews which have been extracted from the restaurant review site www.opentable.com. We have access to 10,000 labeled and 300,000 unlabeled training reviews, while the test set contains 10,000 examples. The goal is to predict the rating on a 5 star scale and performance is evaluated using Root Mean Squared Error (RMSE).3 The review text is treated as a bag of words and trans¬formed into binary vectors encoding the presence/absence of terms. For computational reasons, only the 5000 most frequent terms of the vocabulary are kept in the feature set.4 The resulting pre¬processed data is very sparse: 0.6% of non-zero features on average. Unsupervised pre-training of the networks employs both labeled and unlabeled training reviews while the supervised fine-tuning phase is carried out by 10-fold cross-validation on the labeled training examples.

The model are stacked denoising auto-encoders, with 1 or 3 hidden layers of 5000 hidden units and rectifier or tanh activation, which are trained in a greedy layer-wise fashion. Predicted ratings are defined by the expected star value computed using multiclass (multinomial, softmax) logistic

3Even though our tasks are identical, our database is larger than the one of (Snyder and Barzilay, 2007). 4Preliminary experiments suggested that larger vocabulary sizes did not markedly change results.

7

 

regression output probabilities. For rectifier networks, when a new layer is stacked, activation values of the previous layer are scaled within the interval [0,1] and a sigmoid reconstruction layer with a cross-entropy cost is used. We also add an L1 penalty during pre-training and fine-tuning. Because of the binary input, we use a “salt and pepper noise” (i.e. masking some inputs by zeros and others by ones) for unsupervised training of the first layer. A simple zero masking (as in (Vincent et al., 2008)) is used for the higher layers. We selected the noise level of each layer based on the classification performance, the other hyperparameters are selected according to the reconstruction error.

Table 2: Test RMSE and sparsity level obtained by 10-fold cross-validation on OpenTable data.

Network No hidden layer Rectifier (1-layer) Rectifier (3-layers) Tanh (3-layers)

RMSE 0.885 f 0.006 0.807 f 0.004 0.746 f 0.004 0.774 f 0.008

Sparsity 99.4% 28.9% f 0.2 53.9% f 0.7 00.0% f 0.0

Results are displayed in Table 2. Interestingly, the RMSE significantly decreases as we add hidden layers to the rectifier neural net. These experiments confirm that rectifier networks improve after an unsupervised pre-training phase in a semi-supervised setting: with no pre-training, the 3-layers model can not obtain a RMSE lower than 0.833. Additionally, although we can not replicate the original very high degree of sparsity of the training data, the 3-layers network can still attain an overall sparsity of more than 50%. Finally, on data with these particular properties (binary, high sparsity), the 3-layers network with tanh activation function (which has been learnt with the ex¬act same setup) is clearly outperformed. The sparse behavior of the deep rectifier network seems particularly suitable in this case, because the raw input is very sparse and varies in its number of non-zeros. The latter can also be achieved with sparse internal representations, not with dense ones.

5 Conclusion

Sparsity and neurons operating mostly in a linear regime can be brought together in more biologi¬cally plausible deep neural networks. Rectifier units help to bridge the gap between unsupervised pre-training and no pre-training, which suggests that they may help in finding better minima during training. This finding has been verified for four image classification datasets of different scales and all this in spite of their inherent problems, such as zeros in the gradient, or ill-conditioning of the parametrization. Rather sparse networks are obtained (from 50 to 80% sparsity for the best general¬izing models, whereas the brain is hypothesized to have 95% to 99% sparsity), which may explain some of the benefit of using rectifiers. Furthermore, rectifier activation functions have shown to be remarkably adapted to sentiment analysis, a text-based task with a very large degree of data sparsity. This promising result tends to indicate that deep sparse rectifier networks might yield powerful text mining tools in the future.

Acknowledgments

The authors are greatful for the support received from the following organizations: DARPA, NSERC, CBIE, CIFAR, the Canada Research Chairs, and MITACS.

References

Anonymous, A. (2010). Anonymous title. Technical report, Anonymous University.

Attwell, D. and Laughlin, S. (2001). An energy budget for signaling in the grey matter of the brain. Journal of Cerebral Blood Flow and Metabolism, 21(10), 1133–1145.

Bengio, Y. (2009). Learning deep architectures for AI. Foundations and Trends in Machine Learning, 2(1), 1–127. Also published as a book. Now Publishers, 2009.

Bengio, Y. and Glorot, X. (2010). Understanding the difficulty of training deep feedforward neural networks. In Proceedings of AISTATS 2010, volume 9, pages 249–256.

Bengio, Y., Lamblin, P., Popovici, D., and Larochelle, H. (2007). Greedy layer-wise training of deep networks. In NIPS 19, pages 153–160. MIT Press.

Bush, P. C. and Sejnowski, T. J. (1995). The cortical neuron. Oxford university press.

Candes, E. and Tao, T. (2005). Decoding by linear programming. IEEE Transactions on Information Theory, 51(12), 4203–4215.

 

8

 

Dayan, P. and Abott, L. (2001). Theoretical neuroscience. MIT press.

Doi, E., Balcan, D. C., and Lewicki, M. S. (2006). A theoretical analysis of robust coding over noisy overcom-plete channels. In NIPS’05, pages 307–314. MIT Press, Cambridge, MA.

Douglas, R. and al. (2003). Recurrent excitation in neocortical circuits. Science, 269(5226), 981–985.

Dugas, C., Bengio, Y., Belisle, F., Nadeau, C., and Garcia, R. (2001). Incorporating second-order functional knowledge for better option pricing. In NIPS 13. MIT Press.

Erhan, D., Bengio, Y., Courville, A., Manzagol, P.-A., Vincent, P., and Bengio, S. (2010). Why does unsuper¬vised pre-training help deep learning? JMLR, 11, 625–660.

Goodfellow, I., Le, Q., Saxe, A., and Ng, A. (2009). Measuring invariances in deep networks. In NIPS’09, pages 646–654.

Grother, P. (1995). Handprinted forms and character database, NIST special database 19. In National Institute of Standards and Technology (NIST) Intelligent Systems Division (NISTIR).

Hahnloser, R. L. T. (1998). On the piecewise analysis of networks of linear threshold neurons. Neural Netw., 11(4), 691–697.

Hinton, G. E., Osindero, S., and Teh, Y. (2006). A fast learning algorithm for deep belief nets. Neural Computation, 18, 1527–1554.

Jarrett, K., Kavukcuoglu, K., Ranzato, M., and LeCun, Y. (2009). What is the best multi-stage architecture for object recognition? In Proc. International Conference on Computer Vision (ICCV’09). IEEE.

Krizhevsky, A. and Hinton, G. (2009). Learning multiple layers of features from tiny images. Technical report, University of Toronto.

LeCun, Y., Bottou, L., Orr, G., and Muller, K. (1998). Efficient backprop.

LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P. (1998). Gradient-based learning applied to document recog¬nition. Proceedings of the IEEE, 86(11), 2278–2324.

LeCun, Y., Huang, F.-J., and Bottou, L. (2004). Learning methods for generic object recognition with invariance to pose and lighting. In Proc. CVPR’04, volume 2, pages 97–104, Los Alamitos, CA, USA. IEEE Computer Society.

Lee, H., Battle, A., Raina, R., and Ng, A. (2007). Efficient sparse coding algorithms. In NIPS’06, pages 801–808. MIT Press.

Lee, H., Ekanadham, C., and Ng, A. (2008). Sparse deep belief net model for visual area V2. In NIPS’07, pages 873–880. MIT Press, Cambridge, MA.

Lennie, P. (2003). The cost of cortical computation. Current Biology, 13, 493–497.

Mairal, J., Bach, F., Ponce, J., Sapiro, G., and Zisserman, A. (2009). Supervised dictionary learning. In NIPS’08, pages 1033–1040. NIPS Foundation.

Nair, V. and Hinton, G. E. (2010). Rectified linear units improve restricted boltzmann machines. In Proc. 27th International Conference on Machine Learning.

Olshausen, B. A. and Field, D. J. (1997). Sparse coding with an overcomplete basis set: a strategy employed by V1? Vision Research, 37, 3311–3325.

Pang, B. and Lee, L. (2008). Opinion mining and sentiment analysis. Foundations and Trends in Information Retrieval, 2(1-2), 1–135. Also published as a book. Now Publishers, 2008.

Ranzato, M., Poultney, C., Chopra, S., and LeCun, Y. (2007). Efficient learning of sparse representations with an energy-based model. In NIPS’06.

Ranzato, M., Boureau, Y.-L., and LeCun, Y. (2008). Sparse feature learning for deep belief networks. In NIPS’07, pages 1185–1192, Cambridge, MA. MIT Press.

Salinas, E. and Abbott, L. F. (1996). A model of multiplicative neural responses in parietal cortex. Neurobiol¬ogy, 93, 11956–11961.

Serre, T., Kreiman, G., Kouh, M., Cadieu, C., Knoblich, U., and Poggio, T. (2007). A quantitative theory of immediate visual recognition. Progress in Brain Research, Computational Neuroscience: Theoretical Insights into Brain Function, 165, 33–56.

Snyder, B. and Barzilay, R. (2007). Multiple aspect ranking using the Good Grief algorithm. In Proceedings of HLT-NAACL, pages 300–307.

Vincent, P., Larochelle, H., Bengio, Y., and Manzagol, P.-A. (2008). Extracting and composing robust features with denoising autoencoders. In ICML’08, pages 1096–1103. ACM.

 

9

 

 

 

2020/2021

 

 

1.0 Natural Resources Institute Overview 2

2.0 The Doctoral Thesis 3

3.0 The Doctoral Thesis Process 4

Stage One: Development and Approval of Thesis Research Proposal 5

• Thesis Advisory Committee 5

• Development and Approval of Thesis Research Proposal 6

• Ethics Review Process 8

Stage Two: Candidacy Examination 9

Stage Three: Field and/or Laboratory-Based Research 11

Stage Four: First Draft Meeting 12

• Thesis Format 12

Stage Five: Preparation of Final Thesis 13

Stage Six: Oral Examination 14

Stage Seven: Following Completion of the Oral Examination 19

4.0 Authorship of Publication 20

5.0 Appeal – Procedures and Guidelines 20

6.0 Credential to be Granted 20

 

Ph.D. Thesis Process 2020/2021 - 1 -

 

The degree of Doctor of Philosophy is granted only upon evidence of general proficiency and of distinctive attainment in a special field. In particular, a recognized ability for independent investigation must be demonstrated by the candidate in a thesis which embodies original research or creative scholarship and is presented with a high degree of literacy. It is a research degree and is never conferred solely as a result of coursework study.

The thesis is therefore one of the requirements for the Doctor of Philosophy degree, Natural Resources and Environmental Management. These thesis process guidelines are intended to familiarize Ph.D. students and committee members with the thesis research process. These thesis process guidelines are supplementary to the Faculty of Graduate Studies Thesis Guidelines. 

The Natural Resources Institute (NRI) at the University of Manitoba offers Master's and Ph.D. degrees in resource and environmental management. It was established in 1968 as a degree-granting, interdisciplinary unit with a threefold purpose, namely: (a) to teach management skills leading to a graduate degree of Master of Natural Resources Management (M.N.R.M.); (b) to conduct useful research on actual resource and environmental management problems; and (c) to provide a forum for examining problems in resource use, environmental sustainability, and management.

The Ph.D. program at the NRI was launched in 2001, as part of the Institute’s expanded mandate to contribute to advancing theoretical and applied knowledge in the interdisciplinary fields of resources and the environment. The Ph.D. program prepares students to conduct independent research aimed at solving the complex theoretical and applied issues facing the world environment using a variety of approaches, tools and methods.

 

Ph.D. Thesis Process 2020/2021 - 2 -

 

Ph.D. students in Natural Resources and Environmental Management (NREM) are required to:

• conduct original research;

• synthesize and communicate their findings and conclusions in a thesis; and

• demonstrate a contribution to the theory and practice of natural resources and environmental management.

Through student and faculty research, the Institute is involved in a wide range of natural resource and environmental management issues and problem areas. Research is conducted in conjunction with international agencies, national, provincial and local governments, business and private groups. Sources of the Institute's Ph.D. research support are wide ranging including national and international research funding agencies, client/interest groups, government departments, non-government organizations, and community-based organizations. Students are also financially supported by doctoral scholarships and fellowships from various sources, University of Manitoba Graduate Fellowships, and topical research projects of NRI faculty members.

The research process in general provides a valuable outreach function by bringing together university faculty and practising resource professionals. All thesis research conducted at the Institute is available to the public.

The Ph.D. thesis at the Natural Resources Institute is an independently written research document resulting from an interdisciplinary research project that involves theoretical analysis of a topic or issue, practical application of strategies to address a topic or issue, or a combination of these two, in the areas of natural resources and environmental management. The thesis should show that the student has mastery of the specific field of natural resources and environmental management under investigation and is fully conversant with the relevant literature. The thesis should also demonstrate that the candidate has made a contribution to knowledge in the field of interest. The thesis may entail cooperation with agencies in the community at large and with other faculties at the U of M.

In general, the overall goal of the thesis is to build or apply theory through disciplined and focused independent study. Therefore, the thesis should be based on scholarly study and research that encompasses both theoretical and empirical aspects in the natural resources and environmental management fields, as defined in the Mission  Statement of the Natural Resources Institute.

 

Ph.D. Thesis Process 2020/2021 - 3 -

 

The purpose of the thesis is to:

• develop the student's skills and abilities to conduct independent theoretical and empirical/applied research in the natural resources and environmental management field;

• make original and creative contributions to advancing knowledge; and

• prepare a dissertation demonstrating a high degree of literacy and analytical skills.

As part of the Ph.D. thesis research requirements, each student will:

• prepare a Ph.D. Thesis Research Proposal for committee approval;

• complete a Ph.D. Candidacy Examination;

• complete a first draft of their thesis and have it reviewed by their committee; and

• prepare and defend their Ph.D. Thesis.

The thesis research process has seven distinct stages. Please note that the timing of the seven stages may differ from student to student and that Stages 1 and 2 may occur in reverse order as decided on the advice the Advisor.

 

Each of the preceding stages is described more fully in the material that follows.

Ph.D. Thesis Process 2020/2021 - 4 -

 

Completion of the Ph.D. Thesis Research Seminar; development of the research proposal; Ph.D. Program of Study and Appointment of Advisory Committee; first committee meeting to review and approve the Ph.D. Thesis Proposal  (usually during the first 12-18 months); ethics and animal care  committee approvals; and preparation for the Candidacy Examination.

An Advisory Committee guides the student during the research process. The committee is formally appointed in writing by the NRI Director for each Ph.D. student. The Advisory Committee must consist of a minimum of three members of the Faculty of Graduate Studies, one of whom must hold a primary appointment from within NRI and one of whom must hold a primary appointment outside NRI. Committees may also include one guest member who has expertise in a related discipline/field but is not a member of the Faculty of Graduate Studies. The membership of the Committee, including the advisor, as well as any subsequent changes to it, must be approved by the Dean of the Faculty of Graduate Studies. The Advisor is the Chair of the Advisory Committee.

When the Advisory Committee is formed, the Faculty of Graduate Studies Ph.D.  Program of Study and Appointment of Advisory Committee form and the NRI  Thesis Committee Data form must be completed.

Responsibilities of the committee are to approve the program of study and thesis proposal and to exercise general supervision over the student's work throughout the Ph.D. program. The committee should meet with the student periodically (no less than once a year) to review the student's progress and to report this progress to the Faculty of Graduate Studies on the annual Progress Report form.

The ultimate goal of the committee is to assist the student during the research process. The Institute faculty advisor is formally appointed by the Director based upon the student's stated research interest prior to his/her admission into the program. The faculty advisor provides on-going guidance to the student, directs the work of the committee, and calls and chairs selected meetings of the committee leading to the final oral examination of the thesis. At each stage of the research process, the student must submit draft material to their faculty advisor for review and make necessary changes prior to submission to the committee as a whole. In addition, the faculty advisor and the student are responsible for ensuring that all forms required by the Natural Resources Institute and the Faculty of Graduate Studies are duly and accurately completed.

 

Ph.D. Thesis Process 2020/2021 - 5 -

 

Each student, working in close consultation with their faculty advisor, is responsible for co-ordinating committee meetings and for ensuring that committee members are informed by the NRI, in writing, of meeting date, time and location.

. Students should take into account committee members' schedules in arranging meeting dates. Once the student has identified an appropriate date and time, the faculty advisor and General Office should be notified two weeks in advance using the appropriate form so that proper arrangements for the meeting can be made (confirmation letters and seminar room booking).

If at any time during the research process the student feels a committee member is not fulfilling his/her role, the Director should be approached with the concern. If a student wishes to change faculty advisor at any time, written justification for such a change must be provided to the Director, or Associate Head, if the Director is the faculty advisor.

at the 7000 Level or above,

beyond the Master’s degree (or its equivalent),

A minimum of 6 credit hours of courses must be completed within the Natural Resources Institute and must include NRI 7310 Ph.D. Thesis Research Seminar. The Ph.D. Thesis Research Seminar is offered to assist students with the thesis research and proposal development process.

Changes in program of study, including withdrawal from individual courses or changes of course category, are permitted but only with the approval of the advisor and the Director of the Natural Resources Institute.

Concurrent with the coursework, each student will develop a thesis research proposal in consultation with their faculty advisor and Ph.D. Advisory Committee as the first step in the thesis research process at the NRI.

The proposal should include the following:

• Introduction and Literature Review

• Required subsections:

• Background or Rationale

• Relevant theories

• Management/conservation and/or policy implications

• All of the above must be supported through logical argument and appropriate references to relevant academic literature

• Additional potentially important subsections (to be determined with committee):

 

Ph.D. Thesis Process 2020/2021 - 6 -

 

• Conceptual framework

• Purpose/Goal, Objectives, questions, and/or hypotheses

• Methods/Methodology

• Required subsections:

• Study design

• Sampling methods or Coverage of study subjects

• Analytical methods

• Timeline

• Budget

• Additional potentially important subsections (to be determined with committee):

• Philosophical worldview

• Study area

• Literature/References cited

The advisor will call an Advisory Committee meeting to consider the adequacy of the proposal, at which time a student will give a 20 minute oral presentation and address questions from the committee.

Written notice of the meeting must be provided to each member of the Advisory Committee.

The purpose of the proposal meeting is to:

• enable the Advisory Committee to assess the student’s preparation for undertaking independent Ph.D. research;

• consider the feasibility of the proposed research; and

• provide necessary feedback for further advancement of the research.

The proposal meeting will be chaired by the advisor, and will have the following steps:

1. Introduction by advisor;

2. Presentation of highlights of the research proposal by the student;

3. Review of the proposal by the Advisory Committee; and

4. Determination of approval of the proposal by the Advisory Committee.

The decision of the Advisory Committee must be unanimous. The student is expected to work with the advisor to address any concerns raised in the proposal meeting. The feedback received must be incorporated into the document for circulation among the committee members and the advisor when requested by the committee. The Advisory Committee and NRI Director must approve the oral examination of the thesis proposal on the Faculty of Graduate Studies Ph.D.  Thesis Proposal form, after which the student may proceed to the Candidacy Examination.

 

Ph.D. Thesis Process 2020/2021 - 7 -

 

Completed and approved thesis proposal; development of research design and instruments; and procurement of clearance from the Human Ethics or Animal Care Review Committees.

Ethical conduct related to “human research” activities at the University is governed by The Ethics of Research Involving Human Subjects Policy. All students whose research will involve collecting data from human participants must consult this policy, complete the Course On Research Ethics (CORE), and complete an Ethics Protocol Submission Form. The policy and forms can be obtained from the Office of Research Services, Human Ethics Coordinator, 208-194 Dafoe Road, or by visiting http://umanitoba.ca/research/orec/ethics/humanethicsindex.html.

Submission forms should indicate that the protocol is to be submitted to the Joint-Faculty REB.

The Policy defines as: “any project that involves the collection of specimens, data or information from persons, through intervention or otherwise. Included are procedures that have a low degree of invasiveness (e.g. surveys, interviews, naturalistic observation, exercise or psychometric testing, examination of patient records), as well as more invasive procedures (e.g. blood sampling, insertion of a cannula, administration of a substance).” A

in human research is defined as “a person, who by virtue of his/her involvement in a data-gathering situation or activity, is a source of primary data or information.”

The NRI adheres to the University of Manitoba Policy and Procedures on  Animal Care and Use for research and teaching. All research involving animals must be reviewed and approved by a Protocol Management and  Review Committee prior to being undertaken. It is the responsibility of those needing to use animals to provide sufficient information to the Committee in order that an informed ethical decision can be made regarding their request to use animals. Approval is granted only after the proposed usage and care of the animals, i.e., the protocol of procedures, has been carefully examined. Animal Use Protocol Forms are available online at : http://umanitoba.ca/research/orec/animalcare/animalcareforms.html 

 

Ph.D. Thesis Process 2020/2021 - 8 -

 

The candidacy examination will explore the student’s theoretical and applied knowledge in the broader field of study relating to the dissertation. The focus of the examination will be on the selected thematic areas or subfields (up to four thematic areas), and their synthesis, relating to natural resources and environment and the literature. The candidacy exam is comprised of a written and an oral component. The written examination will normally be completed over a period of three to four weeks. After receiving the Candidacy assignment, the student may seek clarification from members of the examining committee for three days.

(*taken directly from FGS website)

The Candidacy Examination is an absolute requirement of the Faculty of Graduate Studies and, as such, cannot be waived under any circumstances. However, the format and content of the candidacy exam will vary from unit to unit. The purposes of the candidacy exam in doctoral programs is to determine the student's competence in the discipline with respect to understanding and absorbing a broad spectrum of material, and then researching, identifying, analysing, synthesizing, and communicating ideas about that material in depth.

At the time specified by the advisory committee--normally within the first year after the completion of the Ph.D. program coursework but in no case later than one year prior to expected graduation--the student must successfully complete the formal Candidacy Examination.

The examination is conducted according to a procedure established by the unit and approved by the Academic Guide Committee of the Faculty of Graduate Studies. Please see the unit supplemental regulations for the format and composition of the examination committee for the Candidacy Examination. The Candidacy Examination must be held at The University of Manitoba.

This examination, which is independent from the thesis proposal, may be oral, written, or both and may cover subjects relevant to the general area of the candidate's research. These must be made known to the students.

A pass decision of the examiners must be unanimous. Students must be provided with feedback on their performance and access to the reasons for the pass/fail.

 

Ph.D. Thesis Process 2020/2021 - 9 -

 

The Dean of the Faculty of Graduate Studies must be informed whether the candidate has passed or failed the Candidacy Examination on the “Report on Ph.D. Candidacy Examination” form (http://umanitoba.ca/faculties/graduatestudies/forms/index.html).

Any student who fails the Candidacy Examination twice will be required to withdraw from the Faculty of Graduate Studies.

On successful completion of this examination, the student will be considered a candidate for the Ph.D. degree.

Up to four thematic areas or sub-fields relating to the thesis and the broader fields of natural resources and environment will be jointly identified by the thesis Advisory Committee and the student. These thematic areas will serve as the topics of the Candidacy Examination and will encompass the “student’s broader field of study.” The identification of these thematic areas or sub-fields can occur after the Proposal Meeting or if the student and Advisor decide that the Candidacy will occur before the Proposal, the committee and Advisor will need to work together to identify the thematic areas.

The oral examination will normally proceed within one month of the completion of the written component. The oral Candidacy Examination will be conducted by the advisory committee, open to the public, chaired by the Director (or designate), and entail the following:

1. Introduction and explanation of procedures by Chair;

2. Formal presentation by the student for about 25 minutes.

3. Questions by the advisory committee on the selected thematic areas and the student’s broader field of study;

4. In camera determination by advisory committee of pass/fail of student’s Candidacy Examination, both written and oral components; and

5. Completion of Faculty of Graduate Studies form.

Both the written and oral examinations will be evaluated together at the completion of the oral examination. The Chair of the examining committee will call upon each member of the examining committee to determine whether the student has adequate knowledge in the field of study as demonstrated in the written and oral examinations. Considering both the written and oral components of the examination, the committee will vote with regard to the following outcomes:

If this is the first attempt at the Candidacy Examination, the following are potential outcomes of the exam:

• Pass – no further work required – Requires a unanimous decision of the

 

Ph.D. Thesis Process 2020/2021 - 10 -

 

committee.

• Fail - allowed to repeat – Major weaknesses have been identified. After further formal work as determined by the committee, the student will be allowed to re-take the Candidacy Examination. The second attempt of the Candidacy Examinations will normally be completed within six months of the failed attempt.

Note: If allowed to repeat with failure after the first attempt, both the written and oral must be repeated.

If this is the second attempt at the Candidacy Examination, the following are the potential outcomes of the exam:

• Pass – on successful completion of the Candidacy Examination, the student will be considered a candidate for the Ph.D. degree.

• Fail – the Faculty of Graduate Studies will require the student to withdraw from the program.

Collection of primary and secondary data through field and/or laboratory-based research and compilation and processing of the obtained data.

Upon completion of the Candidacy Examination and approval of the research design and instruments (including ethics and animal care approvals), each student is expected to complete field and/or laboratory research work to generate, collect, compile and process primary and secondary data. The nature and duration of field and laboratory work will vary from student to student. Normally, NRI faculty advisors will supervise the student in conducting field research. However, in case of advisor's unavailability, proxy field supervisor should be arranged by the advisor.

Latitude in format to accommodate the subject matter of each thesis is anticipated. Empirical fieldwork and other data collection techniques are critical components of the research process. Accurate records of data sources are essential. Students should ensure that duplicate copies of reports (electronic or hard copy) are safely kept in alternate locations to prevent loss of irreplaceable material. Care should also be taken to maintain confidentiality of material entrusted to the student by agencies or individuals.

 

Ph.D. Thesis Process 2020/2021 - 11 -

 

Presentation and discussion of a first draft of the thesis at an Advisory Committee meeting. The first draft includes the data collected, analysis and discussion of findings in the context of the literature reviewed, and inferences and conclusions drawn.

Prior to the final oral examination of the thesis, the student and the advisor must organize a first draft meeting once a draft of the thesis is complete and deemed ready for presentation to the Advisory Committee for feedback. After data collection and analysis have taken place, writing the first draft of the thesis requires careful presentation of data, analysis and conclusions, and attention to composition, grammar and to the logical organization of the thesis. An appropriate style manual should be consulted prior to thesis writing. An outline is useful in the writing process and care must be taken in the preparation of maps, figures, and tables. The Faculty of Graduate Studies Thesis Guidelines and Information  website outlines the publication requirements of the University.

The student must submit a copy of the first draft of the thesis to each member of the Advisory Committee at least two weeks prior to the meeting.

The Advisory Committee will discuss and recommend modifications necessary for progressing to the final thesis draft and oral examination.

Students should be made aware that the thesis requirement may be met using different formats. Students should discuss possible formats with their faculty advisor and thesis committee members at the proposal meeting.

A typical format often used in the preparation of the thesis is as follows:

Title Page

Signature Page

Abstract

Acknowledgements

Dedication

Table of Contents

List of Tables

List of Figures

Chapter One: Introduction and Statement of the Problem

Chapter Two: Review of Literature

Chapter Three: Methods

Chapter Four: Analysis and Findings

Chapter Five: Discussion and Inferences

 

Ph.D. Thesis Process 2020/2021 - 12 -

 

Chapter Six: Summary, Conclusions, and Recommendations

Footnotes, References, and Appendices

Figures, Illustrations, and Photographs

Additional Materials (i.e. Consent and Access Forms, Copyright Page)

In cases where the student, advisor, and Advisory Committee deem it appropriate, the format for the thesis may take, in part, the form of a manuscript, or manuscripts, ready for submittal to a recognized, peer-reviewed, journal in the appropriate field. The Advisory Committee will ensure that all NRI thesis requirements (i.e., problem or issue statement, objectives, literature review, conclusions, recommendations) are met in the manuscript or chapters as required. The Advisory Committee at the proposal meeting must agree to the journal(s) and manuscript(s) referred to above. The manuscript cannot be in the form of a journal 'note', 'observation', 'opinion', or other 'minor' publication.

Upon completion of the first draft of the thesis, a meeting of the Advisory Committee will be arranged by the advisor to review the thesis for completeness and adequacy to proceed on to the oral defense. The Advisory Committee will advise on the modifications necessary for progressing to the oral defense.

Upon reviewing the first draft document, the Advisory Committee will suggest modifications prior to the final oral examination. Each student is responsible for completing necessary revisions as suggested by committee members. Particular care should be taken to ensure that the linkage between the objectives, hypotheses and/or expected results, findings, conclusions, and recommendations are maintained. In addition, all supporting evidence should be clearly provided throughout the document as appropriate. Figures, tables and maps should be presented in their final form. An abstract and acknowledgements should be included in the final submission. Proper bibliographic referencing should be used according to the appropriate style guide. Detailed publication requirements as described in the Graduate Studies Thesis Guidelines and Information must be followed. The onus is on each student to complete a detailed final editing (grammar, punctuation, spelling and other mechanics) prior to submission to his or her faculty advisor.

Once the thesis is prepared for defence, it is formally submitted by the student to the Faculty of Graduate Studies electronically via the university’s JUMP portal. Detailed instructions for this are provided in a document available at https://umanitoba.ca/faculties/graduatestudies/media/ThesisDistributionPortal.  pdf. Graduate Studies then circulates the thesis to the Advisory Committee members (at this stage called the Examination Committee) for their review. The Examination Committee members must rank the thesis as Class 1 or 2 (see below) in their report to the Dean of Graduate Studies before it is sent to the

Ph.D. Thesis Process 2020/2021 - 13 -

 

External Examiner for their review. The External Examiner must also rank the thesis Class 1 or 2 in order to proceed to the oral examination.

1 The thesis represents a distinct contribution to the candidate’s filed of research and it is acceptable as it stands (or with minor revisions to either content, structure, or writing style) • Thesis has not

received final approval

• Candidate may proceed to their oral examination

2 The thesis has merit since it makes a contribution to the candidate’s field; however, there are research-related concerns that have the potential to be dispelled in the oral examination. The structure and writing are acceptable or require only minor revisions. • Thesis has not

received final approval

• Candidate may proceed to their oral examination

3 The thesis has some merit, but it is not acceptable in the current state since it requires major revisions to one or more of its core components, viz., research content, structure and writing style. • Candidate has

attempt and

to oral

examination

4 The thesis is unacceptable with respect to its core components, viz., research content, structure, and writing style. • Candidate has

attempt and

to oral

examination

The information in this section is adapted from:

http://umanitoba.ca/faculties/graduatestudies/thesis/Ph.D.oegi.html.

Doctoral students must pass an oral examination of the subject of their thesis and matters relating thereto before they may obtain the Ph.D. degree. The examination will be conducted by the Dean of Graduate Studies or the Dean's designate chosen from among scholars of the University. The examining committee will include at least three members of the University of Manitoba academic staff, normally the same members as that of the candidate's advisory committee, and an examiner external to the University of Manitoba. The doctoral candidate, advisor, and all internal members of the examining committee must be present. The presence of the external examiner is strongly encouraged but not mandatory.

 

Ph.D. Thesis Process 2020/2021 - 14 -

 

The following must be assured prior to scheduling the oral examination:

It is required that all internal members of the examining committee be present at the defence, unless exceptional circumstances prevent this and prior approval of the Dean of the Faculty of Graduate Studies has been given. If for any reason an internal examiner is unable to be present in person, participation via video-conferencing may be approved by FGS depending on the constitution of the committee and the presence of other examiners. Oral examinations will be cancelled and deferred if members of the examining committee are not present at the scheduled time and place.

At least two weeks prior to the proposed date the following information must be received by the Faculty of Graduate Studies:

1. Thesis examining committee reports (both internal and external)

2. Student’s examination program information (submitted in a Word document):

a. Biographical data if desired (i.e. birthplace)

b. List of degrees obtained--institution and date

c. List of the candidate's awards, publications, and joint publications

d. An abstract of the thesis (not more than 350 words)

Notes:

1. It is the responsibility of the Department to reserve the room for the oral examination and notify FGS of the date, time and location.

2. The examination must be held at either the University of Manitoba Fort Garry or Bannatyne campus, or the St. Boniface Research Centre.

Once the oral examination has been scheduled, the Faculty of Graduate Studies must be notified. A minimum of two weeks' notice is required. The Faculty of Graduate Studies will then ensure that memoranda announcing the examination are sent to the Department/Unit and the event is posted on the University webpage.

Notes:

1. The oral examination is open to all members of the University community. Graduate students should especially be encouraged to attend.

2. Notices of the examination, including the invitation to graduate students, should be posted within the Department/Unit.

 

Ph.D. Thesis Process 2020/2021 - 15 -

 

Upon receipt of the External Examiner’s report, FGS will forward the report and the “Guidelines for the Scheduling and Conduct of Ph.D. Oral Examinations” to the Department/Unit Head, Advisor, and Internal Examining Committee members.

Approximately one week prior to the oral examination, FGS will provide the Chair of the examination with a Chair package including the following:

• copies of all of the examiner's reports, both internal and external,

• the questions and expected answers, if provided, from the external examiner if not attending,

• the examination programs,

• release and copyright forms for the student,

• "Guidelines for the Scheduling and Conduct of the Oral Examination".

Note:

Questions submitted by the external examiner will not be forwarded to the advisor or to the candidate prior to the examination, unless this is explicitly requested by the external.

The individual chairing the oral examination bears a very important responsibility. A thorough examination and defence of the thesis serves as the University's guarantee that the standards of the Ph.D. are not only maintained, but strengthened. To this end, the procedures for the conduct of the examination must be consistent and adhered to strictly.

The Chair ensures that:

• the location, facilities and equipment are considered appropriate for a Ph.D. defence,

• copies of the examination program are distributed to the candidate, the Examining Committee, and members of the audience, and

• the examining committee is aware of and understands the examination procedures.

The Chair welcomes all of those in attendance and impresses upon the candidate, the examiners, and the audience that this occasion is the senior academic event in the Faculty of Graduate Studies and is a formal examination process prior to awarding the highest academic degree of the University of Manitoba.

 

Ph.D. Thesis Process 2020/2021 - 16 -

 

Referring to the Examination Program, the Chair:

• introduces the candidate, noting previous degrees obtained, awards, special achievements, and publications, and

• introduces the Examining Committee, beginning with a special welcome to the external examiner (if present) and proceeding with introductions of the non-departmental examiner, departmental examiners, and finally the advisor.

Note:

The Chair should note that all members of the examining committee have read the thesis and submitted written reports. The advisor may or may not have included a report.

The Chair outlines the procedures for the conduct of the examination for the candidate, examining committee, and audience as follows:

1. The candidate will be asked to give a 25 to 30 minute presentation of the thesis to include the major objectives, content, results, and conclusions. The Chair also notes that the Candidate's ability to present the thesis clearly and concisely in a 25 to 30 minute period is an essential part of the examination process.

2. At the conclusion of the candidate's presentation, the examiners will be invited to pose questions. The Chair reminds the examiners that they will each be initially allowed approximately 20 minutes to pose their questions and that, because this is an examination on which candidates may, and do, pass or fail, their comments and discussions on the subject matter of the thesis, apart from those essential to the questions, should be made only at the post-examination meeting of the examiners.

3. The order of questioning from the examiners is: external examiner, non-departmental examiner, departmental examiners, and the advisor. The Chair notes that ample time will be given to ensure that the examiners are able to complete their questions and that they may raise additional matters after hearing the initial round of questions. The Chair also notes that, if the external examiner is not present, the Chair (or advisor, as decided) will pose the questions from the external examiner to the candidate.

4. If time permits, and after the questioning by members of the Examining Committee is completed, the Guest member (if applicable and in attendance) may be given an opportunity to ask questions, followed by other members of the FGS and members of the audience.

Ph.D. Thesis Process 2020/2021 - 17 -

 

Notes:

1. The Chair is not permitted to pose any questions of his/her own to the candidate. The Chair must remain neutral at all times.

2. At the conclusion of the candidate's presentation and before the questioning period, the Chair should allow sufficient time for those members of the audience who wish to leave the session to do so.

3. One and one-half hours should normally be sufficient for the conduct of the oral examination by the committee, but in no case should the questioning exceed two hours.

The Chair conducts the remainder of the Oral Examination as follows:

• the audience (and Guest member, if applicable) is thanked for their attendance and support of the candidate and, along with the candidate, is asked to leave so that a confidential meeting of the Examining Committee may be held.

• each examiner is asked, in sequence, to comment on the candidate's performance with respect to the answers to the questions and on whether the candidate should pass or fail the oral examination*.

• the candidate is invited back into the room and informed of the results of the examination, the nature of revisions, and the method of approval of the revisions.

• the candidate is reminded of the procedures for submission of the thesis to MSpace, and of the meaning and use of the copyright and partial release of copyright forms.

• the candidate is congratulated (if appropriate) and members of the Examining Committee thanked.

Notes:

1. *If the external examiner or two or more internal examiners indicate a fail, the candidate fails the examination.

2. The decision of the Examination Committee is reported on the “Final Oral Examination of the Ph.D. Thesis” form, submitted to FGS.

3. A second attempt at the oral defence of the thesis may be scheduled in accordance with the regulations of FGS.

4. The nature of required revisions, prior to submission of the final copy of the thesis to MSpace as well as the procedures for their completion is determined. This is done by the Chair in discussion with the committee.

5. The members of the Examining Committee are asked to sign the approval form, and indicate whether they assign the candidate a vote of “Pass” or “Fail” on their oral defense.

 

Ph.D. Thesis Process 2020/2021 - 18 -

 

• communicate to the Dean of Graduate Studies, through the submission of the “Report for Final Examination of the Ph.D.” form, the nature of the revisions and the procedures to ensure that the revisions are satisfactorily made.

• ensure that the External Examiner is thanked for his/her services.

• approve the revisions to the thesis according to the agreed upon directions from the Examination Committee.

• obtain the committee members’ signatures on the "Final Oral Examination of the Ph.D. Thesis" form, including any members who attended the defence remotely.

• submit the "Final Oral Examination of the Ph.D. Thesis" form to the Graduate Studies Office by the appropriate deadline.

• submit all required forms to the Faculty of Graduate Studies, room 500 University Centre, including thesis release forms and, if applicable, copyright permissions.

• submit a digital version of the final thesis as an e-thesis at the MSpace website(http://mspace.lib.umanitoba.ca).

In addition, the student is recommended for graduation after submitting all the

final copies to the NRI General Office prior to the Faculty of Graduate Studies deadline for submission, required final copies are as follows:

• 1 digital copy – NRI General Office

• 1 digital copy – Advisor

• 1 digital copy – Each Committee member

Students are required to complete all the necessary documentation prior to graduation. Costs for duplicating and binding are the student’s responsibility.

 

Ph.D. Thesis Process 2020/2021 - 19 -

 

Students are encouraged to publish the results of their research work in addition to the preparation of a thesis. The right of first authorship rests with the student for a period of one year from the date of the oral examination of the thesis. Secondary authors should also be agreed upon prior to publication.

After the expiration of the one-year period identified above, or in cases where a thesis is not completed, the faculty advisor and/or other committee members may initiate discussion with the student respecting waiver of publication rights. Any party involved shall refer any dispute respecting publication rights, in the first instance, to the NRI Director or the Associate Head.

Students who disagree with a decision have access to appeal routes and should consult with the Student Advocacy Office, Room 519, University Centre. The Faculty of Graduate Studies Discipline Appeal Form outlines the appeal process and Student Discipline By-Law Appeals Process.

The degree of Doctor of Philosophy, with a specialization in Natural Resources and Environmental Management, is granted upon completion of the requirements of the Natural Resources Institute and the Faculty of Graduate Studies.

 

Ph.D. Thesis Process 2020/2021 - 20 -

 

Twelfth International Multi-Conference on Information Processing-2016 (IMCIP-2016)

Role of Text Pre-Processing in Twitter Sentiment Analysis

Tajinder Singh and Madhu Kumari

National Institute of Technology, Hamirpur 177 005, India

Abstract

Ubiquitous nature of online social media and ever expending usage of short text messages becomes a potential source of crowd wisdom extraction especially in terms of sentiments therefore sentiment classification and analysis is a significant task of current research purview. Major challenge in this area is to tame the data in terms of noise, relevance, emoticons, folksonomies and slangs. This works is an effort to see the effect of pre-processing on twitter data for the fortification of sentiment classification especially in terms of slang word. The proposed method of pre-processing relies on the bindings of slang words on other coexisting words to check the significance and sentiment translation of the slang word. We have used n-gram to find the bindings and conditional random fields to check the significance of slang word. Experiments were carried out to observe the effect of proposed method on sentiment classification which clearly indicates the improvements in accuracy of classification.

© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of organizing committee of the Organizing Committee of IMCIP-2016

Keywords: Classification, CRF, n-Gram, Sentiment, Text Pre-Processing.

1. Introduction

Since the early 1990s the use of internet has increased in different forms. People are communicating with each other using various appearances. In the past era the traffic has become almost the double on internet3. With this growth of internet traffic different online social networks such as Facebook, Twitter, LinkedIn, etc are also becoming famous. This in the digital world, things are changing in a very small time and become popular and trendy over OSN (Online Social Network). Different practices of sharing and communicating are not based the content but also on the basis of repetition of the content4. In the recent era micro-blogging has become very common21 and popular platform for all online users. Millions/Billions of users are sharing their opinion on various aspects on very popular and trendy websites such as twitter, Facebook, tumbler, flicker, LinkedIn etc.5 Twitter is a famous micro-blogging and social networking service which provides the facility to users to share, deliver and interpret 140 words’ post known as tweet3, 6. Twitter have 320 M monthly active user. Twitter is accessible through website interface, SMS, or mobile devices. 80% users are active through mobiles7. In the micro-blogging services users make spelling mistakes, and use emoticons for expressing their views and emotions13. Natural language processing is also playing a big role and can be used according to the opinions expressed17.

Corresponding author. Tel.: +91-9882551893.

E-mail address: madhu.jaglan@gmail.com

1877-0509 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of organizing committee of the Organizing Committee of IMCIP-2016

doi:10.1016/j.procs.2016.06.095

 

550 Tajinder Singh and Madhu Kumari / Procedia Computer Science 89 ( 2016 ) 549 – 554

Table 1. Twitter’s User Distribution.

Twitter Distribution Total

Monthly Active users 320 M

Active users on mobile 80%

Language Supported 35

Unique visits monthly to sites with embedded Tweets 1 B

Table 2. Social Text Quality Challenges.

Challenge Description

Stop List Common words frequency of occurrence

Lemmatization Similarity detection of text/words

Text Cleaning Removal of unwanted from the data

Clarity of Words To clear the meaning in text

Tagging Predicting data annotation and its characteristics

Syntax/Grammar Scope of ambiguity, data dependency

Tokenization Various methods to tokenize words or phrases

Representation of Text Various methods and techniques to represent text

Automated Learning Similarity measures and use of characterization

2. Related Work

Due to irregular, short form of text (hlo, whtsgoin etc.), short length and slang text of tweets it is challenging to predict polarity of sentiment text. In sentiment a mixture of applications are needed to study and these all demands large number of sentiments from sentiment holder. A summary of sentiment is needed, as in polarity disambiguation and analysis; a single sentiment is not adequate for decision. A common form of sentiment analysis is aspect based e.g. phone, quality, voice, battery etc.

Rafael Michal Karampatsis8 et al. described the twitter sentiment analysis for specifying the polarity of messages. They used the two stage pipeline approach for analysis. Authors used the sum classifier at each stage and several features like morphological, POS tagging, lexicon etc are identified.

Joao Leal et al.11 worked to classify polarity of messages by using machine learning approaches. Joachim Wagner et al. described work on aspect based polarity classification by using supervised machine learning with Lucie Flekova et al.10 also worked on sentiment polarity prediction in twitter text.

Nathon Aston et al.3 worked on sentiment analysis on OSN. They used a stream algorithm using modified balanced for sentiment analysis. Lifna C.S.4 puts forward a novel approach where the various topics are grouped together into classes and then assign weight age for each class by using sliding window processing model upon twitter streams. In the similar way Emma Haddi et al.12 discussed the role of text pre-processing for sentiment analysis.

Efthymios Kouloumpis14 defined and explained three way sentiment analysis in twitter for identify positive, negative and neutral sentiments. Efstratios Kontopoulos16 proposed a novel approach for analysis of sentiment. The approach is ontology based and it simply find out the sentiment score as well as grade for each distinct notion in the post.

3. Challenges of Social Text Quality

In most of the social media, language used by the users is very informal15. Users create their own words and spelling shortcuts and punctuation, misspellings, slang, new words, URLs, and genre specific terminology and abbreviations. Thus such kind of text demands to be corrected. Thus for analysing the text HTML characters, slang words, emoticons19, stop words, punctuations, urlsetc are needed to be removed. Splitting of attached words are also be noticed for cleansing. Fangxi Zhang et al.9 used Stanford Parser Tools1 for POS tagging and for parsing while the Natural Language Toolkit2 was used for removing stop words and lemmatization.Users who are also rating the product, services and facilities provided by various websites are needed to be addressed. Various systems for analysing users behaviour, views, attitude are needs to be analysed and demands to be normalized. Various shopping and

 

Tajinder Singh and Madhu Kumari / Procedia Computer Science 89 ( 2016 ) 549 – 554 551

customer services supporting websites used various scales like star scale system18 where the highest rating has 5 stars and the lowest rating has only 1 star, binary rating system where 0 and 1 etc. are used which demands to be normalized.

3.1 Text normalization

 

Fig. 1. Text Normalization as Process.

4. Proposed Scheme

The algorithm to deal with slang and identified words of short text massages of twitter used the coexistence of these words with different entities then decides the significance of slang words based on the sentiment strength and probability of co-occurrence of binding words with slang and unidentified words. Various steps involved in the proposed scheme is given below.

 

552 Tajinder Singh and Madhu Kumari / Procedia Computer Science 89 ( 2016 ) 549 – 554 Assumption

Two unidentified words cannot be consecutive in a tweet; binding of this word is spread up to maximum two neighboring words.

Input

Tweet which is having unidentified word, slang word (Ws) and Folksonomies (except emoticons). Output

Insignificance/significance of slang word and if slang is found insignificant then it weeded out from tweet else it is replaced with positive or negative score with respect to hash tag of the tweet.

4.1 Procedure

Step 1: Find the close binding of the slang word with different senses (coexisting) present in collected tweets so far

based on bigram and trigrams language models.

Let Ws be unidentified word and Wxis word sequence which coexist in collected tweets.

Bigram: If we consider bigram language model then mod(Wx) = 1, then this word can occur. Wx Can occur before or after Ws. Collecting prospective sense binding vector C(w), where w is a ordered pair of words and its associated probabilistic weight

C(w) = fall(Wx, P(Wx, P(Wx, Ws) where P(Wx/Ws) or P(Ws /Wx) > 01

P(Wx, Ws) = P(Wx)*P(Ws /Wx) if Wx occurs before Ws.

P(Ws, Wx) = P(Ws)*P(Wx/Ws) if Ws occurs before Wx.

• If WX occurs before and after Ws, then we can use the following equations to resolve this situation: Max(P(Wx, Ws), P(Ws, Wx)

Trigram: If we consider trigram sense binding vector C(w) where w is a ordered triplet of ordered pair of words and their associated probabilistic weight with Ws.

C(w) = fall((Wx 1, Wx2), P(Wx1, Wx2, Ws)) Where P(Wx 1/Ws Wx2), P(Ws /Wx1Wx2) or P(Wx 2/Ws Wx2) > 01

• If Ws occurs within Wx 1 and Wx2 the following combination:

Ws Wx 1Wx2, WsWx2Wx1, Wx 1Ws Wx2, Wx2Ws Wx1, Wx 1Wx2Ws and Wx2Wx 1Ws,

• Then

P(Wx1, Wx2, Ws) = max(P(Ws Wx 1Wx2), P(WsWx2Wx 1), P(Wx 1WsWx2), P(Wx 2WsWx 1), P(Wx 1Wx 2Ws), P(Wx 2Wx 1Ws)).

At this stage no filtering is done. We try to collect possible bindings.

Step 2: Analysis of these binding of slang word based on fields associated with coexisting words using Conditional Random Fields (CRF) is done at this stage to decide the significance of Ws. Using CRF Part of Speech (POS) tagging of the tweet which contains Ws is done then significance of Ws the measured using following rules:

 

Tajinder Singh and Madhu Kumari / Procedia Computer Science 89 ( 2016 ) 549 – 554 553

 

 

 

Fig. 2. Distribution of Sentiments over Dataset before and after the Pre-processing.

 

Fig. 3. Accuracy of Proposed Scheme under Different Variants of Preprocessing.

 

a. If Ws occurs before and after a proper noun then it is significant.

b. If is coexisting with collective noun and has reference to a proper noun then Ws is less significant.

c. Else Ws is insignificant and Ws can be weeded out of tweet.

Step 3: After collecting the significant all Ws these words are replaced by positive/negative sentiment scores with respect to the concept present in the binding set computed in step1. Following procedure is used to compute the sentiment score (SentiScore) of Ws.

SentiScore(Ws)  max (P(Ws, Wx)  senti(Wx))

P(Ws, Wx) is computed in first step and Wx is a vector, where senti(Wx) max[t (xi)]. t (xi) is sentiment of the tweet in which xi word which is component of Wx is present sentiment of tweet which has Ws is updated as follows:

t(Ws)told  SentiScore(Ws)

told is earlier sentiment value of tweet which was holding Ws.

5. Experiments and Results

For experimentation we have used twitter corpus data. More description of data can be found in2, this data comprises of six fields, first field is sentiment class of the tweet which are negative, neutral and positive, represented by 0, 2 and 4 respectively, rest of field are the id of the tweet, the date of the tweet, the query, the user that tweeted, the text of the tweet. In order to evaluate and measure the impact of proposed scheme on the sentiment classification task we have used Support Vector Machine (SVM) based classifier. We carried out experiment in to two phases, in the first phase we applied the proposed scheme of normalization to the tweets’ text by ignoring their sentiment class. After the normalization process we consider the sentiment class and class 2 i.e. is resolved in to new classes as: 1 (less negative) and 3 (less positive) based on the sentiments of unidentified (slang) words.

Results of experiments clearly suggest that proposed scheme not only robust to size of data but also perform better in terms of accuracy of sentiment classification.

 

554 Tajinder Singh and Madhu Kumari / Procedia Computer Science 89 ( 2016 ) 549 – 554

6. Conclusions and Future Scope

This work is to analyse the impact of pre-processing and normalization on short massages like tweets which are full of information, noise, symbols, abbreviations, folksonomy and unidentified words. Looking at the interestingness to interpret the slang and unidentified words in tweets towards the sentiment, this paper focuses to identify the importance of slang words and to measure their impact on sentiment of the tweet. The proposed scheme used in this paper first gathers the coexisting words with the slang and then exploits characteristics of these binding words to define the significance and sentiment strength of slang word used in the tweet which not only facilitate the better sentiment classification but also ensure the sturdiness of classier as shown in the results. It is yet to be seen the how well the proposed scheme will perform with different classifiers on text streams.

References

[1] http://cs.stanford.edu/people/alecmgo/trainingandtestdata.zip

[2] Twitter Sentiment Classification using Distant Supervision

[3] N. Aston, T. Munson, J. Liddle, G. Hartshaw, D. Livingston and W. Hu, Sentiment Analysis on the Social Networks Using Stream Algorithms, Journal of Data Analysis and Information Processing, vol. 2, pp. 60–66, (2014).

[4] C. S. Lifna and M. Vijayalakshmi, Identifying Concept-Drift in Twitter Streams, ICACTA-2015, Elsevier, (2015).

[5] Ayushi Dalmia, Manwitter Sentiment Analysis Thish Gupta, Vasudeva Varma, The Good, the Bad, and the Neutral, Sem Eval (2015).

[6] Santhi Chinthala, Ramesh Mande, Suneetha Manne and Sindhura Vemuri, Sentiment Analysis on Twitter Streaming Data, Springer International Publishing Switzerland, (2015).

[7] http://twittercommunity.com

[8] Rafeal Mcheal Karampatsis, John Pavlopoulos and Prodromos Malakasiotis, Sentiment Analysis Two Stage Sentiment Analysis of Social Network Messages, SemEval, (2014).

[9] Fangxi Zhang, Zhihua Zhang and Man Lan, ECNU: A Combination Method and Multiple Features for Aspect Extraction and Sentiment Polarity Classification, Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), Dublin, Ireland, pp. 252–258, 23–24 August (2014).

[10] Lucie Flekov, Oliver Ferschk and Iryna Gurevych, UKPDIPF: A Lexical Semantic Approach to Sentiment Polarity Prediction in Twitter Data, Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), Dublin, Ireland, pp. 704–710, 23–24 August (2014).

[11] Joao Leal, Sara Pinto, Ana Bento and Hugo Gonc¸alo Oliveira, Paulo Gomes, CISUC-KIS: Tackling Message Polarity Classification with a Large and Diverse set of Features, Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), Dublin, Ireland, pp. 166–170, 23–24 August (2014).

[12] Emma Haddi, Xiaohui Liu and Yong Shi, The Role of Text Pre-processing in Sentiment Analysis, First International Conference on Information Technology and Quantitative Management, Elsevier, (2013).

[13] Apoorv Agarwal, Boyi Xie, Ilia Vovsha, Owen Rambow and Rebecca Passonneau, Sentiment Analysis of Twitter Data, Department of Computer Science Columbia University New York, NY 10027 USA.

[14] Efthymios Kouloumpis, Theresa Wilson and Johanna Moore, Twitter Sentiment Analysis: The Good the Bad and the OMG!, Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media, (2011).

[15] Sara Rosenthal, Alan Ritter, Preslav Nakov and Veselin Stoyanov, SemEval-2014 Task 9: Sentiment Analysis in Twitter, Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), Dublin, Ireland, pp. 73–80, 23–24 August (2014).

[16] Efstratios Kontopoulos, Christos Berberidis, Theologos Dergiades and Nick Bassiliades, Ontology Based Sentiment Analysis of Twitter Posts, Expert Systems with Applications, vol. 40, issue 10, pp. 4065–4074, August (2013).

[17] Chetashri Bhadane, Hardi Dalal and Heenal Doshi, Sentiment Analysis-Measuring Opinions, International Conference on Advanced Computing Technologies and Applications (ICACTA), vol. 45, pp. 808–814, (2015).

[18] Xing Fang and Justin Zhan, Sentiment Analysis Using Product Review Data, Journal of Big Data, 2015 Springer, (2015).

[19] Xia Hu, Jiliang Tang, Huiji Gao and Huan, Liu, Unsupervised Sentiment Analysis with Emotional Signals, Proceedings of the 22nd International Conference on World Wide Web, WWW’13, ACM, (2013).

[20] Ana Mihanovi´c, Hrvoje Gabelica and ˇZivko Krsti, Big Data and Sentiment Analysis using KNIME: Online Reviews vs. Social Media, MIPRO 2014, 26–30 May 2014, Opatija, Croatia, pp. 1463–1468, (2014).

[21] Lowri Williams, Christian Bannister, Michael Arribas-Ayllon, Alun Preece and Irena Spasic, The Role of Idioms in Sentiment Analysis, Expert Systems with Applications, Elsevier, (2015).

[22] http://www.iprospect.com/en/ca/blog/10-sentiment-analysis-tools-track-social-marketing-success/

[23] http://marcobonzanini.com/2015/03/02/mining-twitter-data-with-python-part-1/

 

NMED-T: A TEMPO-FOCUSED DATASET OF CORTICAL AND

BEHAVIORAL RESPONSES TO NATURALISTIC MUSIC

Steven Losorelli1,2 Duc T. Nguyen1,2 Jacek P. Dmochowski3 Blair Kaneshiro1,2

1Center for the Study of Language and Information, Stanford University, USA

2Center for Computer Research in Music and Acoustics, Stanford University, USA

3Department of Biomedical Engineering, City College of New York, USA

{slosorelli, dtn006, blairbo}@stanford.edu jdmochowski@ccny.cuny.edu

 

ABSTRACT

Understanding human perception of music is foundational to many research topics in Music Information Retrieval (MIR). While the field of MIR has shown a rising interest in the study of brain responses, access to data remains an obstacle. Here we introduce the Naturalistic Music EEG Dataset—Tempo (NMED-T), an open dataset of electro-physiological and behavioral responses collected from 20 participants who heard a set of 10 commercially available musical works. Song stimuli span various genres and tem¬pos, and all contain electronically produced beats in du¬ple meter. Preprocessed and aggregated responses include dense-array EEG and sensorimotor synchronization (tap¬ping) responses, behavioral ratings of the songs, and basic demographic information. These data, along with illustra-tive analysis code, are published in Matlab format. Raw EEG and tapping data are also made available. In this pa-per we describe the construction of the dataset, present re¬sults from illustrative analyses, and document the format and attributes of the published data. This dataset facilitates reproducible research in neuroscience and cognitive MIR, and points to several possible avenues for future studies on human processing of naturalistic music.

1. INTRODUCTION

Humans possess a unique ability to process music, and many topics in Music Information Retrieval (MIR) involve computational modeling of human perception. Tasks that humans often perform with ease—such as melody extrac¬tion, beat detection, and artist identification—remain open topics in MIR. At the same time, a full understanding of the cognitive and perceptual processes underlying human processing of music has yet to be reached.

Greater cross-disciplinary collaboration between MIR and neuroscience has been proposed [14], and a number of studies have incorporated approaches from both fields.

c Steven Losorelli, Duc T. Nguyen, Jacek P. Dmochowski, and Blair Kaneshiro. Licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0). Attribution: Steven Losorelli, Duc T. Nguyen, Jacek P. Dmochowski, and Blair Kaneshiro. “NMED-T: A Tempo-Focused Dataset of Cortical and Behavioral Responses to Nat¬uralistic Music”, 18th International Society for Music Information Re¬trieval Conference, Suzhou, China, 2017.

 

For example, neural correlates of short- and long-term fea¬tures introduced in MIR for genre classification [34] have been sought [1, 6, 10, 20], and brain responses have been used in MIR-related applications including tempo estima¬tion [29,30] and emotion recognition [5, 21]. Yet even as brain data become more prevalent in MIR research, ex¬perimental design, data collection, and data cleaning can present challenges [14]. Therefore, the research com¬munity can arguably benefit from curated, ready-to-use datasets of brain responses to real-world musical works.

Aiming to provide an open dataset with which sev¬eral MIR and neuroscience topics can be explored, we introduce the Naturalistic Music EEG Dataset—Tempo (NMED-T), a dataset of EEG and behavioral responses to commercially available musical works. The dataset con¬tains dense-array EEG responses from 20 adult participants who listened to 10 full-length songs, as well as tapped re¬sponses to the beat of shorter excerpts (collected in a sep¬arate listen). These responses have been cleaned and ag¬gregated, and are ready to use in Matlab format along with ratings of familiarity and enjoyment, as well as basic de¬mographic information about the participants.

NMED-T contributes to a growing body of publicly available music-related EEG repositories, including the DEAP [17], Music BCI [32], NMED-H [15], and Open-MIIR [31] datasets. It is well suited for MIR research in that the data are cleaned and ready to use but are also made available in raw form; stimuli are complete, naturalistic 1 musical works spanning a wide range of tempos; metadata links to stimulus audio are provided; and behavioral data are included. Moreover, as EEG was recorded while par¬ticipants listened attentively but did not focus on any par¬ticular dimension of the songs, these data are suitable for studying many aspects of music processing.

The remainder of the paper is structured as follows. In  2 we describe stimulus selection, study design, data col¬lection, and data preprocessing. Illustrative analyses of the preprocessed data, which build upon past music perception and MIR approaches and reveal cortical and behavioral ev¬idence of entrainment to musical beat, are presented in  3. In  4 we document the dataset itself. We conclude and discuss potential future uses of the data in  5.

1 Denoting real-world music—i.e., music that was created to be con¬sumed in everyday life, as opposed to controlled stimuli created for ex¬perimental research.

 

339

 

340 Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017

 

2. METHODS

2.1 Stimuli

2.1.1 Stimulus Selection

As the present dataset is focused on naturalistic music and tempo, stimuli were ecologically valid, real-world musical works containing steady, electronically produced beats in duple meter at a variety of tempos. The 10 selected songs are all 4:30–5:00 in length, contain vocals (all but one in English), and are in the Western musical tradition. Song information is summarized in Table 1.

To aid in song selection, we computed objective mea-sures of tempo using publicly available Matlab code [8]. The computed tempos were then validated perceptually by four trained musicians. The final set of selected songs range in tempo from 56–150 BPM—a wide enough range to potentially explore octave errors [11, 35]. To facilitate further research on the audio as well as the responses, we purchased digital versions of all songs from Amazon, and include in Table 1 each song’s Amazon Standard Identifi¬cation Number (ASIN).

These real-world stimuli are complex and contain en-ergy at various frequencies—not just those directly re-lated to the beat. We followed the approach of Nozaradan et al. [27] and visualized low-frequency spectra of the stim¬uli. We extracted the amplitude envelope of each song us¬ing the MIR Toolbox, version 1.5 [18] at a sampling rate of 125 Hz (the sampling rate of the preprocessed EEG), and plotted magnitude spectra up to 15 Hz. As can be seen in Fig. 1, spectral peaks often occur at harmonics and subhar-monics of the beat—implicating the hierarchical timescale of music—as well as at other frequencies.

2.1.2 Stimulus Preparation

To prepare the stimuli for the EEG experiment, full-length songs were first converted to mono using Audacity, ver¬sion 2.1.2.2 We then embedded the second audio channel with an intermittent click that was transmitted directly to the EEG amplifier (not played to participants) to ensure precise time stamping of the stimuli. For the behavioral experiment, we created 35-second excerpts of each song. Using Audacity, we selected the audio from 1:00–1:34 and applied a linear fade-in and fade-out to the first and last 2 seconds, respectively. We then appended 1 second of si¬lence to make the conclusion of each excerpt more obvious to the participant.

2.2 Participants

Twenty right-handed participants, aged 18–29 years (mean age 23 years, 6 female) participated in the experiment. All reported normal hearing, fluency in English, and no cogni¬tive or decisional impairments. We imposed no eligibility criteria related to formal musical training; 17 participants reported having received training (mean 8.4 years among those with training). Participants reported listening to mu¬sic for 14.5 hours per week on average.

2http://www.audacityteam.org

 

2.3 Experimental Specifications & Data Collection

This study was approved by the Stanford University In-stitutional Review Board. All participants provided writ-ten informed consent before participating. Each partici-pant filled out a general demographic and musical back-ground questionnaire, after which the EEG and tapping blocks were completed, with the EEG block always oc-curring first.

2.3.1 EEG Experiment

First, each participant was informed that the general pur-pose of the experiment was to study human processing of music, and that he or she would be completing an EEG ses¬sion and a behavioral test. As the EEG data were collected for the general study of music processing (not limited to beat perception), no explicit mention of beat or tempo was given at this stage of the experiment. Rather, participants were instructed simply to listen attentively to the songs as they played, and to avoid movement of any kind (includ¬ing stretching, yawning, and tapping or moving to the beat) during the trials. Songs were presented in random order. Following each trial, participants delivered ratings of fa¬miliarity and enjoyment for the song just presented, on a scale of 1–9. The EEG experiment was split into two con¬secutive recording blocks in order to mitigate participant fatigue, limit data size of the EEG recordings, and allow for verification of electrode impedances between record¬ings. Therefore, a total of 40 EEG recordings were col¬lected across the 20 participants.

The EEG experiment was programmed in Matlab ver-sion 2013b 3 with a custom template built on the Psy-chophysics Toolbox, version 3 [4]. Each participant sat comfortably in a chair at a desk for the duration of the experiment. Stimuli were presented through magnetically shielded Genelec 1030A speakers at a measured loudness level between 73–78 dB. During the trials, the participant viewed a fixation image presented on a computer monitor located 57 cm in front of him or her.

Dense-array EEG was recorded using the Electrical Geodesics, Inc. (EGI) GES300 system [33]. Data were recorded from 128 electrodes with vertex reference using an EGI Net Amps 300 amplifier and Net Station 4.5.7 acquisition software, sampled at 1 kHz with a range of 24 bits. Electrode impedances were verified to be no greater than 50 kΩ—an appropriate level for this system— at the start of each recording.

2.3.2 Behavioral Experiment

Following the EEG recordings, the electrode net was re-moved from the participant, and the behavioral test began. Here, each participant listened to the 35-second song ex¬cerpts, after receiving instructions to “tap to the steady beat of the song as you perceive it.” If the participant had ques¬tions about tapping to multiple tempos for a given song, he or she was instructed to tap to the steady beat that best re¬flected his or her perception of it in the moment. Excerpts were presented in random order.

3https://www.mathworks.com

 

Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017 341

# Song Title Artist ASIN Tempo (BPM) Tempo (Hz) min:sec

1 “First Fires” Bonobo B00CJE73J6 55.97 0.9328 4:38

2 “Oino” LA Priest B00T4NHS2W 69.44 1.1574 4:31

3 “Tiptoes” Daedelus B011SAZRLC 74.26 1.2376 4:36

4 “Careless Love” Croquet Club B06X9736NJ 82.42 1.3736 4:54

5 “Lebanese Blonde” Thievery Corporation B000SF16MI 91.46 1.5244 4:49

6 “Canop´ee” Polo & Pan B01GOL4IB0 96.15 1.6026 4:36

7 “Doing Yoga” Kazy Lambist B01JDDVIQ4 108.70 1.8116 4:52

8 “Until the Sun Needs to Rise” R¨uf¨us du Sol B01APT6JKA 120.00 2.0000 4:52

9 “Silent Shout” The Knife B00IMN40O4 128.21 2.1368 4:54

10 “The Last Thing You Should Do” David Bowie B018GS2A46 150.00 2.5000 4:58

Table 1. Stimulus set. Songs were selected on the basis of vocals, electronically produced beats, genre, tempo, and length.

 

Figure 1. Low-frequency magnitude spectra of stimulus amplitude envelopes. Frequencies related to the musical beat hierarchy, from 1/4x the tempo (whole notes) to 8x the tempo (32nd notes) are denoted with vertical dashed lines.

 

Tapping responses were collected using Tap-It, an iOS application that plays audio while simultaneously record-ing responses tapped on the touchscreen [16]. We note a tap-to-timestamp latency of approximately 15 msec (st. dev. 5 msec) [16]. An Apple iPad 2 was used for this experiment, with stimuli delivered at a comfortable listen-ing level using over-ear Sony MDR-V6 headphones.

2.4 Data Preprocessing

All data preprocessing and analysis was conducted using Matlab, versions 2013b and 2016b.

2.4.1 EEG Preprocessing

The following preprocessing steps were performed on in-dividual EEG recordings that had been exported from Net Station to Matlab cell arrays. First, data from each electrode in the electrodes-by-time data matrix were zero-phase filtered using 8th-order Butterworth highpass (0.3 Hz) and notch (59–61 Hz) filters, and a 16th-order Chebyshev Type I lowpass (50 Hz) filter. Following this, the filtered data were temporally downsampled by a factor of 8 to a final sampling rate of 125 Hz.

We extracted trial labels, onsets, and behavioral rat-ings, and corrected the stimulus onset times using the click events sent directly from the audio to the EEG amplifier. The data for each trial were epoched, concatenated, and DC corrected (subtracting from each electrode its median value). Bad electrodes were removed from the data ma¬trix, resulting in a reduction in the number of rows. We computed EOG components for tracking vertical and hori¬zontal eye movements, and retained electrodes 1–124 for further analysis, excluding electrodes on the face. We 

 

applied a validated approach using Independent Compo-nents Analysis (ICA) to remove ocular and cardiac artifacts from the data [2,13] using the runica function from the EEGLAB toolbox [7].

As final preprocessing steps, transients exceeding 4 standard deviations of each electrode’s mean power were identified in an iterative fashion and replaced with NaNs. We then reconstituted missing rows corresponding to pre-viously identified bad electrodes with rows of NaNs, en-suring that each data matrix contained the same number of rows. We appended a row of zeros—representing the vertex reference—and converted the data frame to average reference (subtracting from each electrode the mean of all electrodes). All missing values (NaNs) were imputed with the spatial average of data from neighboring electrodes, and a final DC correction was performed. Finally, the epochs were separated once again into single trials. There¬fore, after preprocessing, each recording produced a cell array of EEG data, each element of which contained an electrodes-by-time matrix of size 125  T, where T varied according to the length of the stimulus.

After preprocessing all recordings, we aggregated the data on a per-song basis. The data frame for each song is thus a 3D electrodes-by-time-by-participant matrix of size 125  T  20.

2.4.2 Preprocessing of Tapping Responses

The Tap-It application stores the timestamps of taps, in seconds, measured from the device touchscreen on a per-trial basis, as well as each participant’s randomized stim-ulus ordering array [16]. We aggregated the tapping re-sponses in a cell array and the ordering arrays in a matrix.

 

342 Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017

 

3. ILLUSTRATIVE ANALYSES

The following analyses are presented to illustrate basic properties of the dataset.

3.1 EEG Responses

One approach to studying beat processing using EEG in-volves low-frequency ( 20 Hz) steady-state evoked po-tentials (SS-EPs). In an SS-EP paradigm, stimuli presented (e.g., flashed or sounded) at a particular frequency elicit brain responses at that same frequency. While SS-EPs are more often used to study vision processing [25], the ap¬proach has in recent years been used to study responses to auditory rhythms. Here, SS-EPs have shown evidence of entrainment to musical beat, peaking at beat- and meter-related frequencies even when metrical accents are imag¬ined [26] or when beat frequencies do not dominate low-frequency stimulus spectra [27]. To our knowledge, mu¬sic SS-EP studies have to date used simple, synthesized rhythmic patterns as stimuli. Our first illustrative analysis extends this approach to complex, naturalistic music.

Spatial filtering is a technique for EEG analysis whereby a weighted sum of electrodes is computed sub-ject to some criterion [3]. Advantages of concentrating ac¬tivity of interest from many electrodes to a few spatially filtered components include dimensionality reduction, im¬proved SNR, and a reduction in multiple comparisons. For the present analysis we consider two simple spatial filters. The first is simply the mean across all electrodes (ME), which can be thought of as a constant weight applied to each electrode. For the second, we perform Principal Com¬ponents Analysis (PCA), and analyze the first PC of data.

We first averaged each song’s 3D electrodes-by-time-by-participant matrix across participants, producing an electrodes-by-time matrix for each song. Then, so that we analyzed the same amount of data for each song and to account for the time course of listener entrainment to the beat [9], we retained 4 minutes of data from each song, starting 15 seconds into the song.

To compute the spatial filters, we concatenated the participant-averaged data frames across all songs, produc-ing an electrodes-by-aggregated-time matrix. Then, for the ME spatial filter, we computed the mean across electrodes, while for the PCA filter we computed electrode weightings for PC1 using Singular Value Decomposition (SVD). Fi¬nally, we reshaped each resulting song-concatenated com¬ponent vector into a songs-by-time matrix. As our current interest is on SS-EPs, we present the magnitude spectrum of each component on a per-song basis.

The SS-EPs are shown in Fig. 2; y-axis scaling is con-sistent within each spatial filtering technique. By inspec-tion of the plots, low frequencies (<15 Hz) of ME spectra occasionally contain peaks at frequencies in the musical beat hierarchy (e.g., Song 5). PC1 performs better, elic-iting more robust spectral peaks at beat-related frequen-cies. Moreover, EEG PC1 appears to peak at frequencies directly related to musical beat, while suppressing many of the other spectral peaks that were observed in the magni¬tude spectra of stimulus amplitude envelopes (Fig. 1).

 

Spatial filters can be visualized by projecting the filter weights on a 2D scalp topography. While it is common to convert the spatial filter weights to a so-called “forward model,” which captures the projection of filtered activity on the scalp, for PCA the spatial filter is equivalent to the forward model [28]. The ME filter, applying a constant weight to all electrodes, would reveal no spatial variation. However, the PC1 filter topography (Fig. 2, bottom right) applies a range of positive and negative weights to the elec¬trodes, which may help to explain why this filter produces more prominent spectral peaks at beat frequencies.

3.2 Behavioral Ratings

Participant ratings of familiarity and enjoyment are shown in Fig. 3. Familiarity with the songs was low overall; rat¬ings of enjoyment tended to be higher, and also varied more across participants.

3.3 Tapped Responses

For each trial of tapping data, we first converted each inter-tap interval to an instantaneous measure of tempo in Hz, mapped it to the midpoint of the interval, and then linearly interpolated the result to a consistent timing grid with a temporal resolution of 2 Hz. We analyze and plot data from a 17-second interval starting 15 seconds into the excerpt (i.e., starting at time 1:15 in the complete song).

The aggregate tapping responses are shown in Fig. 4. We present two visualizations of these results. First, the top figure for each song shows instantaneous tempo over the time of the excerpt for individual participants (gray curves), with the median across participants plotted in black. In bottom figures, we compute the median tempo across time for each individual participant, and summa-rize with histograms. Beat-related frequencies are shown in the orange (1/2x tempo frequency), green (tempo fre-quency), and red (2x tempo frequency) lines. To a large extent, participants tended to tap at what we had previously determined to be the tempo frequency. However, there are cases of lower agreement, particularly for the slowest songs (Song 1 and Song 2). Here, the histograms suggest a nearly bimodal distribution of tapped tempos, split be¬tween the computational measure and twice that, with the higher measure lying closer to what is considered the pre¬ferred tempo region for humans [23].

4. PUBLISHED DATASET

We publish the cleaned EEG data, aggregated behavioral ratings, aggregated tapped responses, and basic demo-graphic data about the participants in Matlab .mat for-mat. Example code and helper functions for the illustrative analyses are provided, also in Matlab format. Finally, we publish raw EEG recordings (for researchers who wish to apply their own preprocessing pipelines) as well as individ¬ual files of the tapped responses. The dataset is available for download from the Stanford Digital Repository [22], 4 published under a Creative Commons CC-BY license.

4https://purl.stanford.edu/jn859kj8079

 

Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017 343

 

 

Figure 2. Low-frequency EEG spectra using a mean-electrode spatial filter (top) and PC1 spatial filter (bottom) for each song. Beat-related frequencies are shown with dashed vertical lines. Bottom right: PC1 spatial filter weights.

 

 

Figure 3. Participant ratings of familiarity and enjoyment.

4.1 Cleaned EEG Data

The .mat file songSS Imputed.mat contains the cleaned EEG records, aggregated across participants, for song SS (§ 2.4.1). There are 10 such files, one per song. Each .mat file contains the following variables:

• dataSS: 3D electrodes-by-time-by-participant data frame. The size is 125 × T × 20, with T varying according to the song.

• subsSS: Cell array of participant ids. Contents are the same for all songs, but are included in order to link these data to raw EEG files, raw tapping re-sponses, and participant demographics.

• fs: Sampling rate, in Hz (always 125).

4.2 Raw EEG Data

We provide the raw EEG records in their exported state before preprocessing. No filtering, epoching, or cleaning has been performed. As each participant underwent two recordings, there are a total of 40 raw EEG files. The file PP R raw.mat refers to recording R 1, 2 from partici¬pant PP. Each file contains the following variables:

• X: Raw data frame. Size is electrodes-by-time, 129× T, where T is the total length of the recording, including time periods not related to the experimen¬tal trials. The vertex reference electrode is row 129.

 

• DIN 1: Cell array containing all event labels (trig-gers) and times. We provide the helper function parseDIN.m to extract the labels and onsets into numeric vectors. Full specification on labels is pro¬vided in the README file accompanying the dataset.

• fs: Sampling rate, in Hz (always 1000).

4.3 Behavioral Ratings

Participants delivered ratings of familiarity (Q1) and en-joyment (Q2) of each song during the EEG session. The file behavioralRatings.mat contains a single vari¬able behavioralRatings, which is a 3D participant-by-song-by-question (20 × 10 × 2) matrix.

4.4 Tapping Responses

Aggregated and raw tapping responses are stored in the file TapIt.zip. This archive contains the file TapIt.mat, which comprises the following variables:

• allTappedResponses: Aggregated tapped response times across all participants and songs. This is a participants-by-song (20 × 10) cell array. Each entry is a column vector of tap times in seconds, recorded from the device touchscreen.

• allSongOrders: Song-order vectors, aggregated across all participants. This is a participants-by-trial (20×10) matrix, where each row contains the stimu¬lus presentation order for the respective participant. Numbering starts at 1.

Individual response files are also included in the .zip file:

• PPP SS.txt: Single trial of tapped responses, in sec-onds, for participant PPP and song SS.

• PPP play order.txt: Stimulus presentation ordering for participant PPP. Numbering starts at 0.

4.5 Participant Demographics

The file participantInfo.mat contains a struct ar¬ray participantInfo with participant demographics. Fields

 

344 Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017

 

 

Figure 4. Tapping responses. Top: Instantaneous tempo over time for individual participants (gray), with median across participants in black. Bottom: Histograms of median tempo, over time, for individual participants. Ground-truth tempos are shown with orange (1/2x tempo frequency), green (tempo frequency), and red (2x tempo frequency) lines.

 

include age, nYearsTraining, weeklyListening (hours), and id (participant identifier link to raw filenames).

4.6 Code

The file Code.zip contains the Matlab scripts for the analyses performed in  3. A variety of helper functions and files (e.g., electrode location map, script to parse the DIN 1 variable in raw EEG files) are also provided here.

5. DISCUSSION

This paper introduces NMED-T, an open dataset of elec-trophysiological and behavioral responses collected from 20 participants listening to real-world musical excerpts. The published data include both raw and preprocessed dense-array EEG and tapping responses, behavioral ratings of the songs, and basic demographic information.

Our illustrative analyses validate the frequency-tagging, SS-EP approach [26,27] with responses to complex, nat¬uralistic music (Fig. 2). Even a simple PCA filter com¬puted from trial-averaged responses highlights beat-related frequencies in the EEG spectra. Many PC1 spectra show prominent peaks between 5–10 Hz, regardless of tempo; future research could use this dataset to investigate further the stimulus and response attributes contributing to this phenomenon. The variability in tapping responses (Fig. 4) highlights the challenge of defining a ‘ground truth’ for tempo and beat identification, particularly for complex mu¬sic [24]. Here we see various, sometimes conflicting re¬sults across and within participants’ tapped responses. Past research has suggested that humans inherently prefer cer¬tain frequencies related to natural movement [23, 35]. This may help to explain why some participants tapped at twice the tempo for the slowest songs, tending toward the postu¬lated 2-Hz natural resonant frequency.

We faced several trade-offs when designing the study. Collection of EEG data, while relatively inexpensive [14], still incurs costs of equipment and time. Participant fa-tigue must also be taken into account when planning the overall duration of an experiment. As we wished to col-lect EEG responses to a set of full-length songs from ev-ery participant, we were limited in the number of songs 

 

we could use, and relegated the secondary tapping task to shorter excerpts. Stimulus selection, too, is often a com-promise of breadth and depth. For example, the OpenMIIR dataset [31] uses shorter stimuli from a variety of genres, but at the expense of depth within any one genre; while the NMED-H [15] includes various stimulus manipulations of complete songs, but only four songs from a single genre. Our focus on full-length songs with a steady beat and a va¬riety of tempos limited the range of genres somewhat. We also deliberately avoided massively popular songs in or¬der to minimize possible effects, on the brain responses, of varying familiarity, established personal preferences, and autobiographical associations with the songs [12].

There are shortfalls to the dataset. One potential con-found is that the EEG session always preceded the behav¬ioral task; thus, participants were more familiar with the music during the tapping task. As a result, the tapping data may not be suitable for studying the time course of beat entrainment. However, we chose this arrangement so that participants would not be focused specifically on beat while EEG responses were recorded. Second, the tapping data show variations in tapped tempo across participants and within-participant over time. Whether this reflects our participant pool (not all trained musicians), inadequate in¬struction for the task, or is merely characteristic of this re¬sponse is not addressed in the present illustrative analyses. Finally, listeners are known to exhibit variations in tempo octave during tapping while largely agreeing on whether a song is fast or slow [19], but we unfortunately did not collect data here to explore this distinction.

Generally speaking, this dataset facilitates research on encoding and decoding of naturalistic music. While the study design and initial analyses focused primarily on beat and tempo, the EEG responses can be analyzed in conjunc¬tion with various other stimulus features as well. Investiga¬tion of individual differences is also possible (e.g., predict¬ing a particular participant’s tapping tempo or preference rating from his or her own EEG). Other researchers might consider augmenting the dataset with complementary re¬sponses to the same songs. Ideally, the dataset will find ap¬plications in MIR and neuroscience research beyond those envisioned by the authors of this study.

 

Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017 345

 

6. ACKNOWLEDGMENTS

This research was supported by the Patrick Suppes Gift Fund and the Wallenberg Network Initiative: Culture, Brain, Learning. The authors thank Anthony Norcia, Dan Ellis, and the anonymous ISMIR reviewers for helpful feedback on the manuscript.

7. REFERENCES

[1] V. Alluri, P. Toiviainen, I. P. J¨a¨askel¨ainen, E. Glerean,

M. Sams, and E. Brattico. Large-scale brain networks emerge from dynamic processing of musical timbre, key and rhythm. NeuroImage, 59(4):3677–3689, 2012.

[2] A. J. Bell and T. J. Sejnowski. An information-maximization approach to blind separation and blind deconvolution. Neural Computation, 7(6):1129–1159, 1995.

[3] B. Blankertz, R. Tomioka, S. Lemm, M. Kawanabe, and K. R. Muller. Optimizing spatial filters for ro¬bust EEG single-trial analysis. IEEE Signal Processing Magazine, 25(1):41–56, 2008.

[4] D. H. Brainard. The psychophysics toolbox. Spatial Vi¬sion, 10(4):433–436, 1997.

[5] R. Cabredo, R. S. Legaspi, P. S. Inventado, and M. Nu-mao. An emotion model for music using brain waves. In ISMIR, pages 265–270, 2012.

[6] F. Cong, V. Alluri, A. K. Nandi, P. Toiviainen, R. Fa, B. Abu-Jamous, L. Gong, B. G. W. Craenen, H. Poiko-nen, M. Huotilainen, and T. Ristaniemi. Linking brain responses to naturalistic music through analysis of on¬going EEG and stimulus features. IEEE Trans. Multi¬media, 15(5):1060–1069, 2013.

[7] A. Delorme and S. Makeig. EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics in¬cluding independent component analysis. Journal of Neuroscience Methods, 134(1):9–21, 2004.

[8] D. P. W. Ellis. Beat tracking by dynamic programming. Journal of New Music Research, 36(1):51–60, 2007.

[9] P. Fraisse and B. H. Repp. Anticipation of rhythmic stimuli: Speed of establishment and precision of syn-chronization (1966). Psychomusicology: Music, Mind, and Brain, 22(1):84, 2012.

[10] N. Gang, B. Kaneshiro, J. Berger, and J. P. Dmo-chowski. Decoding neurally relevant musical features using Canonical Correlation Analysis. In ISMIR, 2017.

[11] J. Hockman and I. Fujinaga. Fast vs slow: Learning tempo octaves from user data. In ISMIR, pages 231– 236, 2010.

[12] P. Janata. The neural architecture of music-evoked autobiographical memories. Cerebral Cortex, 19(11):2579–2594, 2009.

[13] T.-P. Jung, C. Humphries, T.-W. Lee, S. Makeig, M. J. McKeown, V. Iragui, and T. J. Sejnowski. Extended ICA removes artifacts from electroencephalographic recordings. In NIPS, pages 894–900, 1998.

 

[14] B. Kaneshiro and J. P. Dmochowski. Neuroimaging methods for music information retrieval: Current find¬ings and future prospects. In ISMIR, pages 538–544, 2015.

[15] B. Kaneshiro, D. T. Nguyen, J. P. Dmochowski, A. M. Norcia, and J. Berger. Naturalistic music EEG dataset—Hindi (NMED-H). In Stanford Digital Repos¬itory, 2016.

[16] H.-S. Kim, B. Kaneshiro, and J. Berger. Tap-It: An iOS app for sensori-motor synchronization experiments. In ICMPC12, 2012.

[17] S. Koelstra, C. Muhl, M. Soleymani, J. S. Lee, A. Yaz-dani, T. Ebrahimi, T. Pun, A. Nijholt, and I. Patras. DEAP: A database for emotion analysis using phys¬iological signals. IEEE Trans. Affective Computing, 3(1):18–31, 2012.

[18] O. Lartillot and P. Toiviainen. A Matlab toolbox for musical feature extraction from audio. In DAFx, pages 237–244, 2007.

[19] M. Levy. Improving perceptual tempo estimation with crowd-sourced annotations. In ISMIR, pages 317–322, 2011.

[20] Y. P. Lin, J. R. Duann, W. Feng, J. H. Chen, and T. P. Jung. Revealing spatio-spectral electroencephalo-graphic dynamics of musical mode and tempo percep¬tion by independent component analysis. Journal of NeuroEngineering and Rehabilitation, 11(1), 2014.

[21] Y. P. Lin, C. H. Wang, T. P. Jung, T. L. Wu, S. K. Jeng, J. R. Duann, and J. H. Chen. EEG-based emo-tion recognition in music listening. IEEE Transactions on Biomedical Engineering, 57(7):1798–1806, 2010.

[22] S. Losorelli, D. T. Nguyen, J. P. Dmochowski, and B. Kaneshiro. Naturalistic music EEG dataset— Tempo (NMED-T). In Stanford Digital Repository, 2017.

[23] D. Moelants. Preferred tempo reconsidered. In ICMPC7, pages 1–4, 2002.

[24] D. Moelants and M. F. McKinney. Tempo perception and musical content: What makes a piece fast, slow, or temporally ambiguous? In ICMPC8, pages 558–562, 2004.

[25] A. M. Norcia, L. G. Appelbaum, J. M. Ales, B. R. Cot-tereau, and B. Rossion. The steady-state visual evoked potential in vision research: A review. Journal of Vi¬sion, 15(6):4, 2015.

[26] S. Nozaradan, I. Peretz, M. Missal, and A. Mouraux. Tagging the neuronal entrainment to beat and meter. The Journal of Neuroscience, 31(28):10234–10240, 2011.

[27] S. Nozaradan, I. Peretz, and A. Mouraux. Selective neuronal entrainment to the beat and meter embed-ded in a musical rhythm. The Journal of Neuroscience, 32(49):17572–17581, 2012.

 

346 Proceedings of the 18th ISMIR Conference, Suzhou, China, October 23-27, 2017

[28] L. C. Parra, C. D. Spence, A. D. Gerson, and P. Sajda. Recipes for the linear analysis of EEG. NeuroImage, 28(2):326–341, 2005.

[29] A. Sternin, S. Stober, J. A. Grahn, and A. M. Owen. Tempo estimation from the EEG signal during percep¬tion and imagination of music. In BCMI/CMMR, 2015.

[30] S. Stober, T. Pr¨atzlich, and M. Meinard. Brain beats: Tempo extraction from EEG data. In ISMIR, pages 276–282, 2016.

[31] S. Stober, A. Sternin, A. M. Owen, and J. A. Grahn. Towards music imagery information retrieval: Intro-ducing the OpenMIIR dataset of EEG recordings from music perception and imagination. In ISMIR, 2015.

[32] M. S. Treder, H. Purwins, D. Miklody, I. Sturm, and B. Blankertz. Decoding auditory attention to instruments in polyphonic music using single-trial EEG classification. Journal of Neural Engineering, 11(2):026009, 2014.

[33] D. M. Tucker. Spatial sampling of head electrical fields: The geodesic sensor net. Electroencephalogra-phy and Clinical Neurophysiol., 87(3):154–163, 1993.

[34] G. Tzanetakis and P. Cook. Musical genre classifica-tion of audio signals. IEEE Trans. Speech and Audio Processing, 10(5):293–302, 2002.

[35] L. van Noorden and D. Moelants. Resonance in the per¬ception of musical pulse. Journal of New Music Re¬search, 28(1):43–66, 1999.

 

Curriculum Vitae

&

Faculty Personal Record

Inas S. Khayal

Assistant Professor

The Dartmouth Institute of

Health Policy & Clinical Practice

Geisel School of Medicine at Dartmouth

Adjunct Assistant Professor

Department of Computer Science

Dartmouth College

 

Modified: August 11, 2017

46 Centerra Pkwy, Rm 256

Lebanon, NH 03766

Email: Inas.Khayal@dartmouth.edu

Of~ice Phone: (603) 653-6866

Website: http://sustainablehealth.dartmouth.edu

 

(This page is intentionally left blank.)

 

Contents

1 Education 2

1.1 Degreed Programs  2

1.2 Professional Development & Training 3

2 Professional Experience 4

2.1 Academic Appointments  4

2.2 Relevant Research & Development Experience  5

2.3 Consulting Experience  6

2.4 Entreprenurial Experience  6

2.5 Honors and Awards  7

3 Research 8

3.1 Publications, Written Works & Presentations  8

4 Teaching 20

4.1 Academic Supervision 20

4.1.1 Summary 20

4.1.2 Other Advising: Research Projects & Internships 20

4.1.3 Postdoctoral Advising 21

4.2 Curriculum Development 21

4.2.1 New Syllabus & Course Development 21

4.3 Teaching Experience 21

4.3.1 Faculty Teaching Experience 21

5 Academic & Professional Service 22

5.1 Academic Service 22

5.1.1 Program & Departmental Service 22

5.1.2 Institutional Service 22

5.2 Professional Service 23

5.2.1 Professional Committees 23

5.2.2 Funding Review 23

5.2.3 Journal Review 23

5.2.4 Conference Review 23

5.2.5 Conference Organization 23

 

1

 

1 Education

1.1 Degreed Programs

Sep 2009 University of California, Berkeley & San Francisco: San Francisco, CA

Ph.D., Joint Bioengineering Doctoral Program

Developed techniques for acquisition, reconstruction, and quantitative analysis of Dif-fusion Tensor Imaging for brain tumor diagnosis throughout the diseased state (i.e. de¬termined tumor type, determined the most pathologic region to be surgically removed, evaluate the effectiveness of drug therapies, etc.) and designed clinical studies to test these developments.

Seven journal papers published, 12 oral and poster conference presentations.

Dec 2008 University of California, Berkeley: Berkeley, CA

2008 Berkeley Mayfield Fellows Program, The Walter A. Haas School of Business – a uni¬versity program that offers 10 selected students per year in-depth training and experi¬ence in high-tech entrepreneurship. It also includes a summer internship at a venture capital firm, a VC mentor, and a study abroad experience.

Dec 2007 University of California, Berkeley: Berkeley, CA

Management of Technology Program, The Walter A. Haas School of Business, the College of Engineering and the School of Information

May 2004 Boston University: Boston, MA

Bachelor of Science, Biomedical Engineering, Summa Cum Laude

Dean’s List (all semesters)

Most Outstanding Senior Project in Biomedical Engineering for ”Angular Dependence

of Light Scattering By Apoptotic Cells”

 

2

 

1. EDUCATION

1.2 Professional Development & Training

2011 MIT Course MAS 961 Networks, Complexity and Its Applications

2010 New England Complex Systems Institute

CX102: Computer Programming and Complex Systems

CX201: Complex Physical, Biological, and Social Systems

CX202: Complex Systems Modeling and Networks

2010 MIT Course ESD 15.871 Introduction to Systems Dynamics

2010 MIT Course ESD 69 Seminar on Healthcare Systems Innovation

2010 MIT Course ESD 34 Systems Architecture

2008 California Institute for Quantitative Biosciences

QB3 Global Bio-Entrepreneurship Course

 

3

 

2 Professional Experience

2.1 Academic Appointments

July 2016 - Dartmouth College, Department of Computer Science Hanover, NH

Present Adjunct Assistant Professor of Computer Science

Developed & teaching CS 89/189 Health Informatics and advise students.

Oct 2015 - Dartmouth College, Geisel School of Medicine Hanover, NH

Present Assistant Professor of Health Policy & Clinical Practice

The Dartmouth Institute for Health Policy & Clinical Practice

Leads the Sustainable Health Lab: Develops systems solutions to curb the growth of chronic disease by improving health outcomes, minimizing healthcare costs and as-sessing technical and social biomedical innovations.

Dartmouth Hitchcock Collaboratory for Population Health Faculty Member: Preven-tion and management of complex conditions in primary care through innovative mod-els of care, uses of technology and new measures and predictive analytics.

Awarded as a Robert Wood Johnson Foundation 2017 New Connections Scholar. Awarded as an AcademyHealth 2017 Systems Science Scholar - to stimulate systems ap¬proaches to complex social, behavioral, and environmental influences.

July 2010 - Masdar Institute of Science & Technology Abu Dhabi, UAE

Aug 2015 Assistant Professor of Engineering Systems & Management

Leads the Sustainable Health Lab: Working with Stakeholders, designing human sub-jects research studies and building quantitative models based on Biological, Social and Environmental Factors affecting Non-communicable Chronic Diseases.

July 2011- Massachusetts Institute of Technology Cambridge, MA

Aug 2015 Research Affiliate at The Media Lab

Collaborating on mobile monitoring technologies to measure social and biological fac 

tors affecting health state with Professors Alex (Sandy) Pentland and Kent Larson.

July 2010 - Massachusetts Institute of Technology Cambridge, MA

June 2011 Visiting Scholar at The Media Lab

Collaborating on mobile monitoring technologies to measure social and biological fac 

tors affecting health state with Professors Alex (Sandy) Pentland and Kent Larson.

 

4

 

2. PROFESSIONAL EXPERIENCE

2.2 Relevant Research & Development Experience

Sep 2004 - University of California, SF, Surbeck Lab of Advanced Imaging San Francisco, CA

Sep 2009 Graduate Researcher

Developed techniques for acquisition, reconstruction, and quantitative analysis of Dif-fusion Tensor Imaging for brain tumor diagnosis throughout the diseased state (i.e. de¬termined tumor type, determined the most pathologic region to be surgically removed, evaluatet the effectiveness of drug therapies, etc.) and designed clinical studies to test these developments.

Sep 2003 - Boston University, Biomedical Optics and Biophotonics Laboratory Boston, MA

May 2004 Independent Student Researcher, Undergraduate Bioengineering Senior Thesis Designed and constructed an instrument to measure light scattering from cells as a function of angle.

Data analysis and modeling to assess changes in the cell nuclei of apoptotic cells. Project includes written proposal, formal technical research report, and a research presentation conference. Program included experience in patent procedures, federal regulations, project planning and quality control.

Summer GE Medical Systems, Early Identification Program Milwaukee, WI

2003 Intern, CT Advanced Technology and Applications (included Six Sigma Training), Designed a tissue phantom and experiment to evaluate the impact of human tissue elasticity on Image Quality for various CT acquisition parameters. Required Advanced Workstation and Animation Shop.

Determined the effect of multiple post-processing filters on 16-slice cardiac image data using Minitab.

Developed a web-based database of human biophysical data and anatomical specifica-tions to aid in design of emerging CT Technologies.

Designed a study to investigate the impact of patient breath-hold and motion affects on CT image quality in the lung. The results were presented at the Conference on Screen¬ing for Lung Cancer mentioned above.

Summer Boston Scientific Corporation Natick, MA

2002 Intern, Molecular Interventions

Developed injectable polymeric drug delivery systems for delivery with minimally in-vasive medical devices. Used spectrofluorometric methods to quantitate DNA release kinetics from polymer systems. Submitted four invention disclosures on novel injec-tion catheter designs; Disclosure received patent approval.

Intern, Urology

Optimized injectable formulations for urological applications. Performed viscometry measurements on these formulations, evaluated the systems in bench-top tissue mod-els and participated in acute animal studies (prepared materials, recorded data and summarized results).

Summer Boston University, Hearing Research Center Boston, MA

2001 Independent Student Researcher

Designed an adaptive test program to measure just noticeable differences (JNDs) us-ing MATLAB interfaced with TDT System II that created a signal according to the speci¬fied frequency. The project determined the correlation between interference and hand dominance with JND values.

 

5

 

2. PROFESSIONAL EXPERIENCE

Jun 1999 - ImmunoGen, Inc. Cambridge, MA

Dec 2000 Independent Researcher and Intern

Performed a toxicity test for various carcinogens on the growth of breast and skin can¬cer cells.

Awarded Bridgewater Regional Science Fair- First Place 2000, Massachusetts State Sci¬ence Fair- Third Place 2000, Naval Academy Award, and $1,500 Sea Education Associ¬ation Award for an ImmunoGen sponsored science fair project. Skills gained: microp-pipetting, ELISA, gel electrophoresis and Western Blot.

Oct 1998 - Boston University, Biology Department Boston, MA May 1999 Independent Researcher

Determined the effects of ultraviolet radiation on Escherichia coli’s RecA and UvrA pathways. Biology Lab sponsored science fair project; Awarded Bridgewater Regional Science Fair Second Place 1999 and Massachusetts State Science Fair Third Place 2000.

2.3 Consulting Experience

Oct 2008 - QB3 New Ventures Consulting (QNVC) San Francisco, CA

Aug 2009 Consultant,

Served as a consultant in the QNVC team to explore market opportunities for scientific

platforms developed at UCSF.

Jun 2008 - 360Fresh, Inc. San Francisco, CA May 2009 Project Manager Consultant,

Developed data mining and analysis tools. Researched and compiled Market Research reports

Jun 2007 - Ortho Kinematics, Inc. San Francisco, CA

May 2008 Clinical Analyst Consultant,

Performed clinical assessment and clinical competitive analysis

Sep 2002 - Boston University Information Technology, ResNet Boston, MA May 2004 Student Consultant,

Diagnosed networking problems and determined the proper resolutions to customers’ computer issues.

2.4 Entreprenurial Experience

Jan 2013 - Engineering Systems Analytics LLC Mansfield, CT

Present Founder, CFO, & COO

A family-owned engineering consultancy startup that applies control, automation, sig¬nal processing and information technologies to the domains of smart power grids, wa 

ter distribution, transportation, manufacturing, and sustainable health.

 

6

 

2. PROFESSIONAL EXPERIENCE

2.5 Honors and Awards

2017 New Connections Scholar Award Robert Wood Johnson Foundation

2017 Systems Science Scholars Award Academy Health

2017 Best Paper Award 19th International Conference on Bioinformatics, Com¬putational Biology and Biomedical Engineering

2013 Honorable Mention Paper Award 7th International Conference on Axiomatic Design

2008 Mayfield Fellow, Management of Technology Program University of California, Berkeley

2004-

2007 Graduate Opportunity Fellowship University of California, Berkeley

2004 Most Outstanding Senior Project Award Boston University

 

7

 

3 Research

3.1 Publications, Written Works & Presentations

Summary

Patents: [5, 4, 3, 2,1]

Peer-Reviewed (Journal) Articles: [18,17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, Peer-Reviewed (Conference) Articles: [17,16, 15, 14, 13, 12, 11, 10, 9,8,7, Reviewed Abstracts: [24,23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, Op-Eds and Media: [4, 3, 2,1]

Presentations (International): [24,23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2,1]

Presentations (National): [24,23, 22, 21, 20, 19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1]

Presentations (Regional/Local): [9, 8, 7, 6, 5, 4, 3, 2,1]

Theses: [2, 1]

White Papers: [5,4, 3, 2, 1]

Students for which Prof. Inas Khayal was the advisor or co-advisor are highlighted in forest green. Prof. Khayal’s role in the publication is highlighted in blue.

Presentations where Prof. Inas Khayal was extended an invitation to present are indicated with an (*). Presentations where Prof. Inas Khayal presented a poster/talk are indicated with an (#).

Patents

[5] I. Khayal, W. Naimark, and M. Palasis, “Multiple needle injection catheter and method of use of same,” United States Patent US7 442187 B2, October 28, 2008.

[4] I. Khayal, W. Naimark, and M. Palasis, “Multiple needle injection catheter and method of use of same,” European Patent EP 1841486, October 10, 2007.

[3] T. Freyman, I. Khayal, and W. Naimark, “Apparatus and methods for delivering a bolus of thera-peutic material,” United States Patent US 20 060 217 656, September 28, 2006.

[2] T. Freyman, I. Khayal, and W. Naimark, “Apparatus and methods for delivering a bolus of thera-peutic material,” International Patent WO/2006/098 816, September 21, 2006.

 

8

 

PEER-REVIEWED (JOURNAL) ARTICLES

[1] I. Khayal, W. Naimark, and M. Palasis, “Multiple needle injection catheter and method of use of same,” International Patent WO/2006/081 026, March 08, 2006.

Peer-Reviewed (Journal) Articles

[18] I. Khayal and A. Farid, “A Dynamic System Model for Personalized Healthcare Delivery and Man-aged Individual Health Outcomes,” Submitted to Enterprise Information Systems, 2017.

[17] I. Khayal and A. Farid, “Architecting a System Model for Personalized Healthcare Delivery and Managed Individual Health Outcomes,” Complexity (in press), 2017.

[16] I. Khayal, W. Zhou, and J. Skinner, “Structuring and Visualizing Healthcare Claims Data Using Systems Architecture,” International Journal of Biological, Biomolecular, Agricultural, Food and Biotechnological Engineering, vol. 11, no. 4, pp. 342–346, 2017.

[15] I. S. Khayal and A. M. Farid, “The Need for Systems Tools in the Practice of Clinical Medicine,” Systems Engineering, vol. 20, no. 1, pp. 3–20, Jan 2017. [Online]. Available: http://dx.doi.org/10.1002/sys.21374

[14] I. Khayal and A. M. Farid, “Axiomatic Design Based Human Resources Management for the En-terprise Transformation of the Abu Dhabi Healthcare Labor Pool,” Journal of Enterprise Transfor-mation, vol. 5, no. 3, pp. 162–191, 2015.

[13] M. Butt, S. Quan, A. S. Pentland, and I. Khayal, “Gender Differences in Real-Home Sleep of Young and Older Couples,” Southwest Journal of Pulmonary and Critical Care, vol. 10, no. 1, pp. 289–299, 2015.

[12] M. Butt, T. B. M. J. Ouarda, S. F. Quan, A. S. Pentland, and I. Khayal, “Technologically sensed social exposure related to slow-wave sleep in healthy adults,” Sleep and Breathing, no. 19, pp. 255–261, 2015.

[11] T. Ouarda, C. Charron, K. N. Kumar, P. R. Marpu, H. Ghedira, A. Molini, and I. Khayal, “Evolution of the rainfall regime in the United Arab Emirates,” Journal of Hydrology, vol. 514, pp. 258–270, 2014.

[10] R. O. Oloritun, T. B. Ouarda, S. Moturu, A. Madan, A. S. Pentland, and I. Khayal, “Change in BMI Accurately Predicted by Social Exposure to Acquaintances,” PLOS One, vol. 8, no. 11, p. e79238, 2013.

[9] R. Oloritun, A. Madan, A. Pentland, and I. Khayal, “Identifying Close friendships in a Sensed So-cial Network,” Procedia-Social and Behavioral Sciences, vol. 79, pp. 18–26, 2013.

[8] N. Aharony, W. Pan, C. Ip, I. Khayal, and A. Pentland, “Social fMRI: Investigating and shaping social mechanisms in the real world,” Pervasive and Mobile Computing, vol. 7, no. 6, pp. 643–659, 2011. [Online]. Available: http://linkinghub.elsevier.com/retrieve/pii/S1574119211001246

[7] I. S. Khayal, S. R. Vandenberg, K. J. Smith, C. P. Cloyd, S. M. Chang, S. Cha, S. J. Nelson, and T. R. McKnight, “MRI apparent diffusion coefficient reflects histopathologic subtype, axonal disrup-tion, and tumor fraction in diffuse-type grade II gliomas.” Neuro-oncology, vol. 13, no. 11, pp. 1192–1201, 2011.

 

9

 

PEER-REVIEWED (CONFERENCE) ARTICLES

[6] I. S. Khayal, M.-Y. C. Polley, L. Jalbert, A. Elkhaled, S. M. Chang, S. Cha, N. A. Butowski, and S. J. Nelson, “Evaluation of diffusion parameters as early biomarkers of disease progression in glioblastoma multiforme,” Neurooncology, vol. 12, no. 9, pp. 908–916, 2010. [Online]. Available: http://www.ncbi.nlm.nih.gov/pubmed/20501631

[5] I. S. Khayal and S. J. Nelson, “Characterization of low-grade gliomas using RGB color maps derived from ADC histograms,” Journal of Magnetic Resonance Imaging, vol. 30, no. 1, pp. 209–213, Jul 2009. [Online]. Available: http://dx.doi.org/10.1002/jmri.21810

[4] I. S. Khayal, T. R. McKnight, C. McGue, S. Vandenberg, K. R. Lamborn, S. M. Chang, S. Cha, and S. J. Nelson, “Apparent diffusion coef~icient and fractional anisotropy of newly diagnosed grade II gliomas,” NMR in Biomedicine, vol. 22, no. 4, pp. 449–455, may 2009. [Online]. Available: http://dx.doi.org/10.1002/nbm.1357

[3] W. Bian, I. S. Khayal, J. M. Lupo, C. Mcgue, S. Vandenberg, K. R. Lamborn, and S. M. Chang, “Mul-tiparametric Characterization of Grade 2 Glioma Subtypes Using Magnetic Resonance Spectro-scopic, Perfusion and Diffusion Imaging,” Translational Oncology, vol. 2, no. 4, pp. 271–280, 2009.

[2] F. Crawford, I. Khayal, C. McGue, S. Saraswathy, A. Pirzkall, S. Cha, K. Lamborn, S. Chang, M. Berger, and S. Nelson, “Relationship of pre-surgery metabolic and physiological MR imaging parameters to survival for patients with untreated GBM,” Journal of Neuro-Oncology, vol. 91, no. 3, pp. 337–351, 2009. [Online]. Available: http://dx.doi.org/10.1007/s11060-008-9719-x

[1] I. S. Khayal, F. W. Crawford, S. Saraswathy, K. R. Lamborn, S. M. Chang, S. Cha, T. R. McKnight, and S. J. Nelson, “Relationship between choline and apparent diffusion coefficient in patients with gliomas,” Journal of Magnetic Resonance Imaging, vol. 27, no. 4, pp. 718–725, apr 2008. [Online]. Available: http://dx.doi.org/10.1002/jmri.21288

Peer-Reviewed (Conference) Articles

[17] I. Khayal and A. Farid, “A Dynamic Model for a Cyber-Physcal Healthcare Delivery System with Human Agents,” in Proceedings of the 2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC2017), Intelligent Industrial System Special Issue, October 2017.

[16] I. Khayal and A. Farid, “Designing Smart Cities for Citizen Health & Well-being,” in Proceedings of the 2017 IEEE International Summer School on Smart Cities (IEEE S3C), August 2017.

[15] I. Khayal and A. Farid, “An Architecture for a Cyber-Physical Healthcare Delivery System with Human Agents,” in Proceedings of the 2017 IEEE International Summer School on Smart Cities (IEEE S3C), August 2017.

[14] I. Khayal, M. McGovern, M. Bruce, and S. Bartels, “Developing an Integrated Behavioral Health System Model using Engineering Design,” in Proceedings of the 2017 Institute of Industrial and Systems Engineering Annual Conference, Pittsburgh, PA, May 2017.

[13] I. Khayal, W. Zhou, and J. Skinner, “Structuring and Visualizing Healthcare Claims Data Using Sys¬tems Architecture,” in Proceedings of the 19th International Conference on Bioinformatics, Compu¬tational Biology and Biomedical Engineering, Boston, MA, April 2017.

[12] I. Khayal and A. M. Farid, “The Application of Model-Based Systems Engineering to the Practice of Clinical Medicine,” in Proceedings of the 2017 Annual IEEE International Systems Conference (SysCon), Montreal, Quebec, Canada, April 2017, pp. 1–6.

10

 

PEER-REVIEWED (CONFERENCE) ARTICLES

[11] A. Popoola, M. Butt, T. B. M. J. Ouarda, and I. Khayal, “Automatically Capturing Sleep and Social Factors to Understand Ramadan in the Real World,” in IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI), Valencia, Spain, jun 2014.

[10] R. Oloritun, A. Pentland, and I. Khayal, “Dynamics of Human Social Networks: People, Time, Relationships, and Places,” in Collaborative Innovation Networks Conference (COINs13). Chile: eprint arXiv:1308.1287, aug 2013. [Online]. Available: http://adsabs.harvard.edu/abs/ 2013arXiv1308.1287O

[9] I. S. Khayal and A. M. Farid, “Axiomatic Design Based Volatility Assessment of the Abu Dhabi Healthcare Labor Market: Part II - Case Study,” in The Seventh International Conference on Ax-iomatic Design (ICAD2013), Worcester, MA, USA, jun 2013, pp. 77–83.

[8] A. M. Farid and I. S. Khayal, “Axiomatic Design Based Volatility Assessment of the Abu Dhabi Healthcare Labor Market: Part I - Theory,” in The Seventh International Conference on Axiomatic Design (ICAD2013), Worcester, MA, USA, jun 2013, pp. 71–76.

[7] M. Butt, S. T. Moturu, A. S. Pentland, and I. Khayal, “Automatically captured sociability and sleep quality in healthy adults,” in 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC’13). Osaka, Japan: Ieee, jul 2013, pp. 4662–4665. [Online]. Available: http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6610587

[6] R. Oloritun, A. Madan, A. Pentland, and I. Khayal, “Evolution of Social Encounters in Ad-Hoc Mobile Face-to-Face Interaction Networks,” in 2012 ASE International Conference on Social Informatics (SocialInformatics), vol. 0, Alexandria, VA, USA, dec 2012, pp. 192–198. [Online]. Available: http://doi.ieeecomputersociety.org/10.1109/SocialInformatics.2012.101

[5] R. O. Oloritun, A. Pentland, A. Madan, and I. Khayal, “Identifying Close friends interactions in Face-to-Face Networks,” in Applications of Social Network Analysis (ASNA2012), Zurich, Switzer-land, Sep 2012, pp. 1–19.

[4] N. Aharony, W. Pan, C. Ip, I. Khayal, and A. Pentland, “The Social fMRI: Measuring, Understanding, and Designing Social Mechanisms in the Real World,” in 13th International Conference on Ubiq¬uitous Computing (UbiComp 2011). Beijing, China: Massachusetts Institute of Technology, Sep 2011, pp. 445–454.

[3] S. T. Moturu, I. Khayal, N. Aharony, W. Pan, and A. S. Pentland, “Using Social Sensing to Understand the Links between Sleep, Mood, and Sociability,” in 2011 IEEE Third Intl Conference on Privacy Security Risk and Trust and 2011 IEEE Third Intl Conference on Social Computing, no. SocialCom. Boston, MA, USA: IEEE, Sep 2011, pp. 208–214. [Online]. Available: http://web.media.mit.edu/{~}panwei/pub/mood{_}socialcom2011.pdf

[2] S. T. Moturu, I. Khayal, N. Aharony, W. Pan, and A. S. Pentland, “Sleep, mood and sociability in a healthy population,” in 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBS’11), vol. 2011, Media Lab, Massachusetts Institute of Technology, Cambridge, MA 02139 USA. Boston, MA, USA: IEEE, Sep 2011, pp. 5267–5270. [Online]. Available: http://www.ncbi.nlm.nih.gov/pubmed/22255526

[1] J. Sairamesh, K. Argenbright, H. C. Simmons, R. Nemana, R. Rajagopal, I. Khayal, and P. Anderson, “Early Warning Methods for Improving Patient Care: Harnessing Insight from Raw Patient Notes,” in 3rd INFORMS Workshop on Data Mining and Health Informatics

 

11

 

REVIEWED ABSTRACTS

(DM-HI 2008), vol. 12, no. 10, Washington, D.C., USA, oct 2008, p. s146. [Online]. Available: http://www.ncbi.nlm.nih.gov/pubmed/24091256

Reviewed Abstracts

[24] I. Khayal and A. Farid, “The Application of Systems Tools in the Practice of Clinical Medicine,” in Proceedings of the 5th International Engineering Systems Symposium (CESUN2016), Washington, D.C., June 2016.

[23] I. Khayal and A. Farid, “Developing Sustainable Health Frameworks in Smart Cities,” in Proceed-ings of the First IEEE International Smart Cities Conference, Guadalajara, Mexico, September 27, 2015.

[22] M. Butt, S. Moturu, A. Pentland, and I. Khayal, “Deep Sleep and Automatically Captured Face-to-Face Interactions in Healthy Adults,” in Proceedings of the 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Baltimore, MD, Jun 2013.

[21] M. Butt, S. Moturu, A. Pentland, and I. Khayal, “Gender Differences in Real-Home Sleep of Young Couples,” in Proceedings of the 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Baltimore, MD, USA, Jun 2013.

[20] I. Khayal, M. Butt, S. Moturu, and A. Pentland, “Sociability and Sleep Quality in Healthy Adults,” in Proceedings of the 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Baltimore, MD, USA, Jun 2013.

[19] F. A. Baloushi, M. Butt, T. B. Ouarda, and I. Khayal, “Capturing Sleep ‘in the wild’ during the Fast¬ing month of Ramadan,” in Proceedings of the 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Baltimore, MD, USA, jun 2013.

[18] I. Khayal, T. B. Ouarda, M. Butt, and F. A. Baloushi, “Capturing Sleep ‘in the wild’ during the Fasting month of Ramadan,” in in Women’s Panel at Abu Dhabi Sustainability Week, Abu Dhabi, Abu Dhabi, UAE, Jan 2013.

[17] R. O. Oloritun, A. Pentland, A. Madan, and I. Khayal, “Exploring the role of duration of interac-tion in ad-hoc mobile face-to-face networks,” in Interdisciplinary Workshop on Information and Decision in Social Networks (WIDS), MIT, Cambridge, MA, USA, nov 2012.

[16] R. O. Oloritun, A. Pentland, A. Madan, and I. Khayal, “Exploring the effect of the duration and period of Face-to-Face interactions on Close Friendships,” in Interdisciplinary Workshop on Infor¬mation and Decision in Social Networks (WIDS), MIT, Cambridge, MA, USA, nov 2012.

[15] R. Oloritun, T. B. Ouarda, S. Moturu, A. Madan, A. Pentland, and I. Khayal, “Change in BMI Accu-rately Predicted via Social Exposure,” in Interdisciplinary Workshop on Information and Decision in Social Networks (WIDS), MIT, Cambridge, MA, USA, nov 2012.

[14] M. Butt, S. T. Moturu, N. Aharony, A. S. Pentland, and I. Khayal, “Effects of Alcohol on Sleep in a Real-home Environment,” in Proceedings of the 26th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Boston, MA, USA, jun 2012.

[13] S. Moturu, I. Khayal, N. Aharony, and A. S. Pentland, “Mood, Sleep and Face-to-Face Interactions in a Co-located Family Community,” in Interdisciplinary Workshop on Information and Decision in Social Networks (WIDS), Boston, MA, USA, may 2011.

12

 

REVIEWED ABSTRACTS

[12] R. Parvataneni, A. Kadambi, I. Park, A. Elkhaled, E. Essock-Burns, I. Khayal, N. Butowski, K. Lam-born, S. M. Chang, and S. J. Nelson, “Finding Early Prognostic Markers From Metabolic and Physi-ological Imaging Parameters For Newly Diagnosed GBM Patients Receiving Radiation, Temozolo-mide and A PKC Inhibitor,” in Proceedings of the 15th Annual Meeting of the Society of Neuro-Oncology (SNO), Montreal, Quebec, Canada, Nov 2010.

[11] I. Khayal, L. Jalbert, A. Elkhaled, S. M. Chang, S. Cha, and S. J. Nelson, “Evaluation of diffusion parameters as Early biomarkers of progression in high-grade gliomas,” in Proceedings of the In-ternational Society for Magnetic Resonance in Medicine (ISMRM) Seventeenth Scienti~ic Meeting, Honolulu, Hawaii, USA, Apr 2009.

[10] I. Khayal, T. McKnight, W. Bien, S. Cha, S. Chang, and S. Nelson, “ADC histogram derived RGB color-maps for characterizing low grade glioma subtypes,” in Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Sixteenth Scienti~ic Meeting, Toronto, Canada, May 2008.

[9] I. Khayal and S. Nelson, “Multi-exponential Diffusion Imaging: Clinically feasible Multiple B-value Diffusion Imaging for high grade gliomas,” in Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Sixteenth Scienti~ic Meeting, Toronto, Canada, May 2008.

[8] W. Bian, I. Khayal, C. McGue, S. Saraswathy, S. Cha, S. Chang, and S. Nelson, “Differentiation of Low-Grade Glioma Subtypes Using Mutiparametric MR Data,” in Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Sixteenth Scienti~ic Meeting, Toronto, Canada, May 2008.

[7] I. Khayal, T. McKnight, C. McGue, K. Lamborn, S. Chang, S. Cha, and S. Nelson, “Diffusion Imaging Parameter values aid in subtyping newly diagnosed low grade gliomas,” in Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Fifteenth Scienti~ic Meeting, Berlin, Germany, May 2007.

[6] I. Khayal, T. McKnight, S. Chang, S. Cha, and S. Nelson, “Apparent diffusion coefficient and frac-tional anisotropy may predict newly diagnosed low-grade glioma subtypes,” in Proceedings of the American Society of Neuro-Radiology 45th Annual Meeting. Chicago, IL: American Society of Neuro-Radiology, Jun 2007.

[5] B. Lee, I. Khayal, S. J. Chang, N. Butowski, S. Chang, S. Nelson, and S. Cha, “The utility of apparent diffusion coefficients in predicting therapeutic outcome in patients with recurrent glioblastoma multiforme treated with convection enhanced delivery of il 13-pe39qqr,” in Proceedings of the American Society of Neuro-Radiology 45th Annual Meeting. Chicago, IL: American Society of Neuro-Radiology, Jun 2007.

[4] I. Khayal, S. Cha, S. Chang, and S. Nelson, “Double b-value difference map better distinguishes tumor from edema than adc and fa maps for patients with high grade gliomas,” in Proceedings of the 2007 Joint Molecular Imaging Conference, Providence, RI, Sep 2007.

[3] I. Khayal, K. Lamborn, F. Crawford, S. Saraswathy, S. Chang, S. Cha, T. McKnight, and S. Nel¬son, “Correlation of Apparent Diffusion Coefficient and Choline in newly diagnosed patients with Glioblastoma Multiforme,” in Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Fourteenth Scienti~ic Meeting, Seattle, WA, USA, May 2006.

 

13

 

PRESENTATIONS (INTERNATIONAL)

[2] F. Crawford, I. Khayal, K. Lamborn, R. Choy, S. Saraswathy, I. Park, J. Oh, S. Cha, S. Chang, and S. Nelson, “Pre-surgery lactate, lipid, and adc predict survival for patients with glioblastoma mul-tiforme,” in Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Fourteenth Scienti~ic Meeting, Seattle, WA, May 2006.

[1] K. Karau, I. Khayal, S. Sirohey, J. Hsieh, and M. Ferrant, “Minimizing Clinical CT Variability in Sequential Volumetric Sizing of Pulmonary Nodules,” in Proceedings of the 9th International Con-ference on Screening for Lung Cancer, Jackson Memorial Hospital, Miami, FL, Oct 2003.

Op-Eds and Media

[4] M. A. Chaudhry, L. A.-R. Kadhim, I. Khayal, and A. Zaheer, “An introduction to diffusion-weighted Magnetic Resonance Imaging in oncology,” Arab Health Imaging and Diagnostics, no. 1, pp. 17–19, 2013.

[3] I. Khayal and T. B. Ouarda, “Groundbreaking Ramadan sleep study to offer fasting insight,” The National, Nov 2012. [Online]. Available: http://www.thenational.ae/news/uae-news/ groundbreaking-ramadan-sleep-study-to-offer-fasting-insight

[2] R. Ghazal, “How does Ramadan affect our sleep? UAE researchers find out,” The Na¬tional, Oct 2012. [Online]. Available: http://www.thenational.ae/news/uae-news/health/ how-does-ramadan-affect-our-sleep-uae-researchers-find-out

[1] I. Khayal, “Health is vital for the future development of the UAE,” The Na¬tional, Sep 2011. [Online]. Available: http://www.thenational.ae/news/uae-news/health/ health-is-vital-for-the-future-development-of-the-uae

Presentations (International)

[24] I. S. Khayal*, “Designing and Assessing Healthcare Systems,” Keynote Presentation at the Inter-national Conference for Axiomatic Design (ICAD2017), Iasi, Romania, September 2017.

[23] I. S. Khayal# and M. Farid, Amro, “An Architecture for a Cyber-Physical Healthcare Delivery Sys-tem with Human Agents,” The 2017 IEEE International Summer School on Smart Cities (IEEE S3C), Natal, Brazil, August, 2017.

[22] I. S. Khayal# and M. Farid, Amro, “Designing Smart Cities for Citizen Health & Well-being,” The 2017 IEEE International Summer School on Smart Cities (IEEE S3C), Natal, Brazil, August, 2017.

[21] I. Khayal# and A. M. Farid, “The Application of Model-Based Systems Engineering to the Practice of Clinical Medicine,” The 11th Annual IEEE International Systems Conference (IEEE SysCon), Montreal, Quebec, Canada, April 2017.

[20] I. Khayal#, W. Zhou, and J. Skinner, “Structuring and Visualizing Healthcare Claims Data Using Systems Architecture,” The 19th International Conference on Bioinformatics, Computational Bi-ology and Biomedical Engineering (ICBCCBE), Boston, MA, April 2017.

[19] I. Khayal# and A. Farid, “The Application of Systems Tools in the Practice of Clinical Medicine,” The 5th International Engineering Systems Symposium (CESUN2016), Washington, D.C., Jun 2016.

 

14

 

PRESENTATIONS (INTERNATIONAL)

[18] I. Khayal# and A. Farid, “Developing Sustainable Health Frameworks in Smart Cities,” 2015 IEEE Smart Cities Conference, September 2015.

[17] I. S. Khayal# and A. M. Farid, “Axiomatic Design Based Volatility Assessment of the Abu Dhabi Healthcare Labor Market: Part II - Case Study,” The Seventh International Conference on Ax-iomatic Design (ICAD2013), Worcester, MA, USA, Jun 2013.

[16] A. M. Farid and I. S. Khayal#, “Axiomatic Design Based Volatility Assessment of the Abu Dhabi Healthcare Labor Market: Part I - Theory,” The Seventh International Conference on Axiomatic Design (ICAD2013), Worcester, MA, USA, Jun 2013.

[15] M. Butt, S. T. Moturu, A. S. Pentland, and I. Khayal#, “Automatically captured sociability and sleep quality in healthy adults,” 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC’13), Osaka, Japan, Jul 2013.

[14] R. Oloritun, A. Pentland, and I. Khayal#, “Dynamics of Human Social Networks: People, Time, Relationships, and Places,” Collaborative Innovation Networks Conference (COINs13), Chile, Aug 2013.

[13] R. Oloritun, A. Madan, A. Pentland, and I. Khayal#, “Evolution of Social Encounters in Ad-Hoc Mobile Face-to-Face Interaction Networks,” 2012 ASE International Conference on Social Infor-matics (SocialInformatics), Alexandria, VA, USA, Dec 2012.

[12] R. O. Oloritun, A. Pentland, A. Madan, and I. Khayal#, “Identifying Close friends interactions in Face-to-Face Networks,” Applications of Social Network Analysis (ASNA2012), Zurich, Switzer-land, Sep 2012.

[11] N. Aharony, W. Pan, C. Ip, I. Khayal#, and A. Pentland, “The Social fMRI: Measuring, Understand-ing, and Designing Social Mechanisms in the Real World,” 13th International Conference on Ubiq¬uitous Computing (UbiComp 2011), Beijing, China, Sep 2011.

[10] S. T. Moturu, I. Khayal#, N. Aharony, W. Pan, and A. S. Pentland, “Using Social Sensing to Under-stand the Links between Sleep, Mood, and Sociability,” 2011 IEEE Third Intl Conference on Pri-vacy Security Risk and Trust and 2011 IEEE Third Intl Conference on Social Computing, Boston, MA, USA, Sep 2011.

[9] S. T. Moturu, I. Khayal#, N. Aharony, W. Pan, and A. S. Pentland, “Sleep, mood and sociability in a healthy population,” 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBS’11), Boston, MA, USA, Sep 2011.

[8] I. Khayal#, L. Jalbert, A. Elkhaled, S. M. Chang, S. Cha, and S. J. Nelson, “Evaluation of diffusion parameters as Early biomarkers of progression in high-grade gliomas,” Proceedings of the Inter¬national Society for Magnetic Resonance in Medicine (ISMRM) Seventeenth Scientific Meeting, Honolulu, Hawaii, USA, Apr 2009.

[7] I. Khayal#, T. McKnight, W. Bien, S. Cha, S. Chang, and S. Nelson, “ADC histogram derived RGB color-maps for characterizing low grade glioma subtypes,” Proceedings of the International Soci¬ety for Magnetic Resonance in Medicine (ISMRM) Sixteenth Scientific Meeting, Toronto, Canada, May 2008.

[6] I. Khayal# and S. Nelson, “Multi-exponential Diffusion Imaging: Clinically feasible Multiple B-value Diffusion Imaging for high grade gliomas,” Proceedings of the International Society for Mag¬netic Resonance in Medicine (ISMRM) Sixteenth Scientific Meeting, Toronto, Canada, May 2008.

15

 

PRESENTATIONS (NATIONAL)

[5] W. Bian, I. Khayal#, C. McGue, S. Saraswathy, S. Cha, S. Chang, and S. Nelson, “Differentiation of Low-Grade Glioma Subtypes Using Multiparametric MR Data,” Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Sixteenth Scientific Meeting, Toronto, Canada, May 2008.

[4] I. Khayal#, T. McKnight, C. McGue, K. Lamborn, S. Chang, S. Cha, and S. Nelson, “Diffusion Imag-ing Parameter values aid in subtyping newly diagnosed low grade gliomas,” Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Fifteenth Scientific Meeting, Berlin, Germany, May 2007.

[3] I. Khayal#, K. Lamborn, F. Crawford, S. Saraswathy, S. Chang, S. Cha, T. McKnight, and S. Nel¬son, “Correlation of Apparent Diffusion Coefficient and Choline in newly diagnosed patients with Glioblastoma Multiforme,” Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Fourteenth Scientific Meeting, Seattle, WA, USA, May 2006.

[2] F. Crawford, I. Khayal#, K. Lamborn, R. Choy, S. Saraswathy, I. Park, J. Oh, S. Cha, S. Chang, and S. Nelson, “Pre-surgery lactate, lipid, and adc predict survival for patients with glioblastoma mul-tiforme,” Proceedings of the International Society for Magnetic Resonance in Medicine (ISMRM) Fourteenth Scientific Meeting, Seattle, WA, May 2006.

[1] K. Karau, I. Khayal#, S. Sirohey, J. Hsieh, and M. Ferrant, “Minimizing Clinical CT Variability in Sequential Volumetric Sizing of Pulmonary Nodules,” Proceedings of the 9th International Con-ference on Screening for Lung Cancer, Jackson Memorial Hospital, Miami, FL, Oct 2003.

Presentations (National)

[24] I. Khayal*, “Applying Systems Science Methods in Health Research,” AcademyHealth’s Annual Research Meeting, New Orleans, LA, June 2017.

[23] I. Khayal#, M. McGovern, M. Bruce, and S. Bartels, “Developing an Integrated Behavioral Health System Model using Engineering Design,” The 2017 Institute of Industrial and Systems Engineer¬ing Annual Conference (IISE), Pittsburgh, PA, May 2017.

[22] I. Khayal#, M. Butt, S. Moturu, and A. Pentland, “Sociability and Sleep Quality in Healthy Adults,” The 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Baltimore, MD, USA, jun 2013.

[21] M. Butt, S. Moturu, A. Pentland, and I. Khayal#, “Deep Sleep and Automatically Captured Face-to-Face Interactions in Healthy Adults,” 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), jun 2013.

[20] M. Butt, S. Moturu, A. Pentland, and I. Khayal#, “Gender Differences in Real-Home Sleep of Young Couples,” The 27th Annual Meeting of the Associated Professional Sleep Societies (SLEEP), Bal¬timore, MD, USA, Jun 2013.

[19] F. A. Baloushi, M. Butt, T. B. Ouarda, and I. Khayal#, “Capturing Sleep ‘in the wild’ during the Fasting month of Ramadan,” The 27th Annual Meeting of the Associated Professional Sleep Soci-eties (SLEEP), Baltimore, MD, USA, jun 2013.

[18] I. Khayal#, T. B. Ouarda, M. Butt, and F. A. Baloushi, “Capturing Sleep ‘in the wild’ during the Fasting month of Ramadan,” Women’s Panel at Abu Dhabi Sustainability Week, Abu Dhabi, Abu Dhabi, UAE, Jan 2013.

16

 

PRESENTATIONS (NATIONAL)

[17] I. Khayal*, “Towards Sustainable Health in Abu Dhabi,” Crown Prince Court Brown Bag Lunch, Abu Dhabi, UAE, October 3, 2012.

[16] R. O. Oloritun, A. Pentland, A. Madan, and I. Khayal#, “Exploring the role of duration of inter-action in ad-hoc mobile face-to-face networks,” Interdisciplinary Workshop on Information and Decision in Social Networks (WIDS), MIT, Cambridge, MA, USA, nov 2012.

[15] R. O. Oloritun, A. Pentland, A. Madan, and I. Khayal#, “Exploring the effect of the duration and period of Face-to-Face interactions on Close Friendships,” Interdisciplinary Workshop on Infor-mation and Decision in Social Networks (WIDS), MIT, Cambridge, MA, USA, nov 2012.

[14] R. Oloritun, T. B. Ouarda, S. Moturu, A. Madan, A. Pentland, and I. Khayal#, “Change in BMI Accu¬rately Predicted via Social Exposure,” Interdisciplinary Workshop on Information and Decision in Social Networks (WIDS), MIT, Cambridge, MA, USA, nov 2012.

[13] M. Butt, S. T. Moturu, N. Aharony, A. S. Pentland, and I. Khayal#, “Effects of Alcohol on Sleep in a Real-home Environment,” The 26th Annual Meeting of the Associated Professional Sleep Soci-eties (SLEEP), Boston, MA, USA, Jun 2012.

[12] S. Moturu, I. Khayal#, N. Aharony, and A. S. Pentland, “Mood, Sleep and Face-to-Face Interactions in a Co-located Family Community,” Interdisciplinary Workshop on Information and Decision in Social Networks (WIDS), Boston, MA, USA, May 2011.

[11] R. Oloritun, T. B. Ouarda, S. Moturu, A. Madan, A. Pentland, and I. Khayal#, “Sustainable Healthy Living: Social Interactions via embedded sensing techniques and lifestyles choices as early warn¬ing markers for Obesity,” in EPFL UAE-Swiss Research Day, Abu Dhabi-Dubai, UAE, Nov 2012.

[10] M. Butt, S. T. Moturu, N. Aharony, A. S. Pentland, and I. Khayal#, “Sustainable Healthy Living: Sociability and Sleep Quality in Healthy Adults,” in EPFL UAE-Swiss Research Day, Abu Dhabi-Dubai, UAE, Nov 2012.

[9] I. Khayal*, “Applications of Mobile Technologies in Sustainable Health,” 4th Annual SEHA Re-search Conference, Abu Dhabi, UAE, December 11, 2012.

[8] I. Khayal*, “Opportunities for Collaboration in Sustainable Health,” Visiting Innovation Profes-sor, Children’s National Medical Center, Sheikh Zayed Institute for Pediatric Surgical Innovation, Washington, D.C., May 26 2011.

[7] I. Khayal*, “Investment Opportunities in Sustainable Health Research,” Mubadala Healthcare, Mubadala Development Company, Abu Dhabi, UAE, November 27, 2011.

[6] I. Khayal*, “Innovation and Technology for Sustainable Health,” Science and Society Innovation, Technology and Biotechnology in the 21st Century Economic Opportunities and Ethical Chal-lenges, New York University Abu Dhabi, Abu Dhabi, UAE, November 16, 2011.

[5] R. Parvataneni, A. Kadambi, I. Park, A. Elkhaled, E. Essock-Burns, I. Khayal#, N. Butowski, K. Lam¬born, S. M. Chang, and S. J. Nelson, “Finding Early Prognostic Markers From Metabolic and Phys¬iological Imaging Parameters For Newly Diagnosed GBM Patients Receiving Radiation, Temo-zolomide and A PKC Inhibitor,” Proceedings of the 15th Annual Meeting of the Society of Neuro-Oncology (SNO), Montreal, Quebec, Canada, Nov 2010.

 

17

 

PRESENTATIONS (REGIONAL/LOCAL)

[4] J. Sairamesh, K. Argenbright, H. C. Simmons, R. Nemana, R. Rajagopal, I. Khayal#, and P. Anderson, “Early Warning Methods for Improving Patient Care: Harnessing Insight from Raw Patient Notes,” 3rd INFORMS Workshop on Data Mining and Health Informatics (DM-HI 2008), Washington, D.C., USA, Oct 2008.

[3] I. Khayal#, T. McKnight, S. Chang, S. Cha, and S. Nelson, “Apparent diffusion coefficient and fractional anisotropy may predict newly diagnosed low-grade glioma subtypes,” Proceedings of the American Society of Neuro-Radiology 45th Annual Meeting, American Society of Neuro-Radiology, Chicago, IL, Jun 2007.

[2] B. Lee, I. Khayal#, S. J. Chang, N. Butowski, S. Chang, S. Nelson, and S. Cha, “The utility of appar-ent diffusion coefficients in predicting therapeutic outcome in patients with recurrent glioblas-toma multiforme treated with convection enhanced delivery of il 13-pe39qqr,” Proceedings of the American Society of Neuro-Radiology 45th Annual Meeting, American Society of Neuro-Radiology, Chicago, IL, Jun 2007.

[1] I. Khayal#, S. Cha, S. Chang, and S. Nelson, “Double b-value difference map better distinguishes tumor from edema than adc and fa maps for patients with high grade gliomas,” Proceedings of the 2007 Joint Molecular Imaging Conference, Providence, RI, Sep 2007.

Presentations (Regional/Local)

[9] I. Khayal*, “Using engineering systems methods to understand variation of healthcare delivery systems at the end-of-life,” PO1 Research Meeting, Greenfield, NH, September 22, 2016.

[8] I. Khayal*, “Developing a Model & Assessment of Clinical Practice: An Engineering Perspective,” Dartmouth-Hitchock Nashua Medical Center, Nashua, NH, August 11, 2016.

[7] I. Khayal*, “Healthcare Systems Modeling and Analysis: The role of Human Resources in Abu Dhabi from 1967-2012,” The Dartmouth Institute Research in Progress Talk, Geisel School of Medicine@Dartmouth, Lebanon, NH, April 22, 2016.

[6] I. Khayal*, “Aging with Technology Support,” Tri-State Summit on Aging, Concord, NH, April 1, 2016.

[5] I. Khayal*, “Sensing Health ’in the Wild’,” Center for Technology and Behavioral Health Affiliates Meeting, Lebanon, NH, February 22, 2016.

[4] I. Khayal*, “Big Data Enabling Sustainable Health,” IBM, Abu Dhabi, UAE, October 2, 2012.

[3] K. Al Dalel and I. Khayal*, “Health and Social Networks in the Workplace and Who we are,” Pre-sentation to the Abu Dhabi Police HQ, Abu Dhabi, UAE, July 2012.

[2] I. Khayal*, “Sustainable Healthy Living,” Masdar Insitute Day at MIT, Cambrige, MA, March 1, 2011.

[1] I. Khayal*, “Nothing To Fear But Fear Itself,” Syrian-Lebanese Women’s Club of Greater Boston Scholarship Awards, Dedham, MA, USA, May 19, 2011.

 

18

 

WHITE PAPERS

Theses

[2] I. Khayal, “Characterization of Diffusion Weighted Magnetic Resonance Imaging for Patients with Brain Tumors,” Ph.D. Dissertation, University of California, Berkeley and San Francisco, 2009.

[1] I. Khayal, “Angular Dependence of Light Scattering by Apoptotic Cells,” B.Sc. Thesis, Boston Uni-versity, 2004.

White Papers

[5] I. Khayal and T. B. Ouarda, “Study of the health benefits of urban green spaces in Abu Dhabi and the United Arab Emirates,” to Mubadala Healthcare, Abu Dhabi, UAE, Tech. Rep., Apr 2012.

[4] T. B. Ouarda and I. Khayal, “Effects of climate on cardiovascular diseases in the UAE for current and anticipated conditions,” to Mubadala Healthcare, Abu Dhabi, UAE, Tech. Rep., Apr 2012.

[3] I. Khayal and T. B. Ouarda, “Characterization of MRI parameters as early warning markers of can-cer in the UAE,” to Mubadala Healthcare, Abu Dhabi, UAE, Tech. Rep., Apr 2012.

[2] I. Khayal and T. B. Ouarda, “Disease Prevention and Health Monitoring for Improved Diabetes Care in the UAE,” to Mubadala Healthcare, Abu Dhabi, UAE, Tech. Rep., Apr 2012.

[1] J. Yoo and I. Khayal, “Proactive Healthcare with Continuous Health Monitoring By Wearable Sen-sors,” to Mubadala Healthcare, Abu Dhabi, UAE, Tech. Rep., Apr 2012.

 

19

 

4 Teaching

4.1 Academic Supervision

4.1.1 Summary

Master’s Theses: [7, 6, 5, 4, 3, 2,1]

Other Advising: Research Projects & Internships: [1, 2, 3, 4, 5, 6, 7]

Post-Docs: [1]

Master’s Theses

[7] H.-W. Yang, “Environment and Health (co-advisor),” Master’s thesis, Masdar Institute of Science and Technology.

[6] R. Al Hashemi, “Environment and Health (co-advisor),” Master’s thesis, Masdar Institute of Science and Technology, 2015.

[5] K. Al Dalel, “Social Interactions and Health in the Workplace: Exploring Sickness Absence Behav-ior in the United Arab Emirates,” Master’s thesis, Masdar Institute of Science & Technology, 2014.

[4] M. Butt, “Understanding Social and Behavioral Factors Affecting Sleep in Healthy Adults,” Master’s thesis, Masdar Institute of Science & Technology, 2013.

[3] Y. Al-Abd, “UAE Aircraft Manufacturing Sector’s Potentials and Challenges (thesis defense mem-ber),” Master’s thesis, Masdar Institute of Science and Technology, 2012.

[2] R. Oloritun, “Exploring Health-Related Behaviors in Face-To-Face Networks,” Master’s Thesis, Masdar Institute of Science & Technology, 2012.

[1] D. Cihan, “Designing Robust Tuberculosis Treatment Networks in Rural India (co-advisor),” Mas-ter’s thesis, Masdar Institute of Science and Technology, 2011.

4.1.2 Other Advising: Research Projects & Internships

1. Dakota Thompson. Discrete-Event Simulation Visualization. Research Assistant Fall 2016.

2. Abdulfatai Popoola. Ramadan Sleep Study. Research Assistant Summer 2013.

3. Kefah Alqawasmeh. Practicum in Population Health Risk Assessment and Management. Visiting Student Summer 2013.

 

20

 

MASTER’S THESES

4. Hazza BaniMalek. Developing a graphical user interface design for Health Data. Summer 2012 Internship.

5. Hanin Almuhallabi. Web visualization tools for Sleep Study. Summer 2012 Internship.

6. Abdulla Al-Kayoumi. Mapping Location and Use of UAE Activity Centers. Summer 2011 Intern-ship.

7. Mouza Al Zaabi. Mapping and understanding Abu Dhabi Food Sources. Summer 2011 Internship. 4.1.3 Postdoctoral Advising

1. Ali Mutlu. Research Area: Statistical signal processing algorithms. Electrical and Computer En-gineering at Michigan State University. February 2013 - October 2013.

4.2 Curriculum Development

4.2.1 New Syllabus & Course Development

Spring 2017: CS 89/189 Health Informatics

Fall 2011: ESM 507 Management and Entrepreneurship for Engineers

Spr 2012: ESM 615 Sustainable Health

Fall 2012: Young Future Energy Leaders (YFEL) Entrepreneurship Course

Fall 2012: ESM 519 Innovation, Creativity and Entrepreneurship

Apr 2013: ESM 620 Analysis of Complex System Networks

4.3 Teaching Experience

4.3.1 Faculty Teaching Experience

Term Course Code Title Role Course

Type

Spring 2014 ESM 620 Analysis of Complex System Networks Co 

Lecturer Lecture

Fall 2013 ESM 507 Management and Entrepreneurship for Engineers Co 

Lecturer Lecture

Spring 2013 ESM 620 Analysis of Complex System Networks Lead Lecturer Lecture

Fall 2012 YFEL Entrepreneurship Course Lead Lecturer Seminar

Fall 2012 ESM 507 Management and Entrepreneurship for Engineers Lead Lecturer Lecture

Fall 2011 ESM 507 Management and Entrepreneurship for Engineers Lead Lecturer Lecture

21

 

5 Academic & Professional Service

5.1 Academic Service

5.1.1 Program & Departmental Service

The Dartmouth Institute. Geisel School of Medicine at Dartmouth.

Dates Role Scope

Summer 2016 - Member TDI Partnership Workgroup

Present

Engineering Systems & Management Department. Masdar Institute of Science & Technology.

Dates Role Scope

Fall 2013 - Fall 2014 Member MSc Admissions Committee Member

Fall 2012 - Fall 2014 Member Faculty Search Committee

5.1.2 Institutional Service

Dartmouth Hitchock Medical Center. Lebanon, NH

Dates Role Committee

Fall 2016 - Present Member D-H Population Health Initiative on Addiction & Mental

Health

Fall 2016 - Summer Member D-H Behavioral Health Planning Team - Screening &

2016 Treatment Work Group

Masdar Institute of Science & Technology. Abu Dhabi, UAE.

Dates Role Committee

Fall 2010 - Fall 2015 Chair & Founder Research Ethics Committee

Fall 2012 - Fall 2013 Member MI/MIT Joint Committee for Collaborative Research

Conference Committee

Fall 2012- Spring 2013 Member Center for Innovation Systems & Entrepreneurship

Working Group

Fall 2012- Spring 2013 Faculty Rep. Japan International Cooperation Center (JICE) Faculty

Representative

Fall 2011- Fall 2012 ESM Rep. PhD Admissions Committee

Fall 2011- Fall 2014 Member Women’s Advisory Committee

22

 

5. ACADEMIC & PROFESSIONAL SERVICE

5.2 Professional Service

5.2.1 Professional Committees

Dates Role Committee Name

Fall 2012 - Fall 2013 Member Health Authority of Abu Dhabi (HAAD) Committee on

Setting Health Research Priorities in Abu Dhabi

Fall 2012 - Fall 2013 Member Weqaya Advisory Task Force Member, Health Authority

of Abu Dhabi

Fall 2011 - Fall 2012 Judge Fifth MIT Arab Business Plan Competition, Round I&II

Fall 2011 Moderator Fifth MIT Arab Business Plan Competition Launch

Event

Fall 2010 - Fall 2011 Judge Fourth MIT Arab Business Plan Competition, Round I&II

5.2.2 Funding Review

2015 MIT Deshpande-Masdar Institute Grants

2011 MIT Energy Initiative Seed Fund Program

5.2.3 Journal Review

2016 - Present Journal of Substance Abuse Treatment

2014 - Present American Journal of Public Health

2013 - Present Journal of Enterprise Transformation

2012 - Present International Journal of Computer and Information Technology (IJCIT)

2011 - Present Psychiatric Services

2010 - Present Neuroradiology Journal

5.2.4 Conference Review

2016 - Present Council of Engineering Systems Universities (CESUN)

2012 - Present International Conference on Mobile Ubiquitous Computing Systems

5.2.5 Conference Organization

Role Scope

Panel Moderator 2017 Institute of Industrial and Systems Engineering Annual Meeting (IISE)

Panel Moderator 2017 Council of Engineering Systems Universities (CESUN) Annual Meeting

Prog. Committee 2015 IEEE Smart Cities Conference, Smart Health Track Track Chair

23

 

CIMCA 2008, IAWTIC 2008, and ISE 2008

 

Comparing Agile Software Processes Based on the Software

Development Project Requirements

Malik Qasaimeh, Hossein Mehrfard, Abdelwahab Hamou-Lhadj

Department of Electrical and Computer Engineering

Concordia University

1455 de Maisonneuve West

Montreal, Quebec H3G 1M8 CANADA

{m_qasaim, h_mehrfa, abdelw}@encs.concordia.ca

 

Abstract

Agile software development principles and methodologies have gained a lot of popularity in recent years. Unlike traditional software processes, agile processes advocate flexible practices that allow coping with changing requirements. Recently, several agile processes have been proposed, which differ in a variety of ways. In this paper, we analyze the characteristics of many agile processes and classify them based on key requirements for a software development project. The objective is to help project managers and software engineers select an agile process that is most suitable to the requirements of their software projects.

1. Introduction

Software development practices have evolved significantly since the term software engineering was popularized by F. L. Bauer during the NATO Software Engineering Conference in 1968. There exist today a large number of software processes that have been introduced and studied at a great extent, but up to date none has proven to be the golden standard.

A software process is defined as a set of activities, methods, practices, and transformations that are used to develop and maintain software and its associated products [3].

Agile processes have been proposed to overcome the flexibility issues of traditional processes. They have been developed by practitioners based on their experience working on several software development projects [8,10]. Although existing agile processes can differ significantly in the way they approach software development, they all share one key characteristic which consists of favoring close collaboration between software development and business teams via face-to-face communication, as opposed to putting an emphasis on written documentation [3, 8]. However, selecting the appropriate agile process is not an easy task due to many differences that are associated with each agile software process. These differences must be 

 

studied before choosing the appropriate methodology that fit the stakeholders’ needs [3, 4, 5]. In this paper, we surveyed and analyzed several agile processes, with the ultimate objective being:

n To help software engineers and project managers understand the strengths and weaknesses of existing agile processes.

n To help managerial staff understand the capabilities of agile processes to support software project requirements.

n To provide researchers in the area of software process engineering with a reference work in which key characteristics of several agile processes are discussed.

The remaining part of this paper is organized as follows. In Section 2, we describe the agile software processes considered in this study. In Section 3, we analyze these processes based on software development project requirements. We conclude the paper in Section 4.

2. Agile Processes

In this section, we describe the agile processes studied in this paper. We selected these particular processes in order to achieve good coverage of the types of activities available.

The processes selected in this study are the following:

n Extreme Programming (XP)

n Scrum

n Feature Driven Development (FDD)

n Adaptive Software Development (ASD)

n Crystal Methodologies (CM)

2.1. Extreme Programming (XP)

To many people, extreme programming (XP) has come to embody the agile methodology itself. This is, perhaps,

 

 

Authorized licensed use limited to: KING SAUD UNIVERSITY. Downloaded on October 4, 2009 at 07:29 from IEEE Xplore. Restrictions apply.

 

because it is one of the first agile processes that has been proposed. In general, XP consists of a set of individual practices that when put together yield a successful software practice [1].

 

Figure 1. Extreme programming (taken from [14])

Figure 1 depicts the activities involved in XP. Using XP, a development team starts by a planning phase, referred to as planning the game. During this phase, the technical team evaluates the feasibility of the project and the effort required as well as the schedule for the first release. The customer then defines the most valuable features, represented as stories, which need to be developed. Every story represents the smallest possible increment to which new features of the system can be added, which usually takes only a few weeks to be developed.

XP supports the idea that the design should be as simple as possible. To achieve this objective, XP puts an emphasis on using refactoring techniques such as removing duplicated code, improving as well as to improve the existing design. It should be mentioned that programmers must verify that the system is still operational after a refactoring activity takes place. The XP process requires that design, implementation, and testing of the system should be carried out by a pair of programmers sharing one computer. This allows programmers to spend more time finding solutions to challenging problem and less time doing routine debugging. Pair programming has been shown to be a useful technique for building robust software [7].

XP is a test driven development method such that, before writing code for a story, programmers must implement the automated tests that will verify the story functionality. As shown in Figure 1, programmers rely on unit tests to verify the correctness of the story. The work on a story is not considered complete until it has been shown to be defect free. Integration tests are run to verify that the overall functionality of the system is bug free. The story is not considered successfully implemented until it passes the acceptance test, which is normally written by the customer so as to verify that the system functionality satisfies the customer’s needs [1].

Using XP, the resulting implementation is owned by all team members. This collective ownership of the artifacts of the system allows programmers to make modifications to parts of the code that have been created by others. The main 

 

advantage of this practice is to speed up the development process such that when programmers detect a fault in the code he has the right to fix it. A coding standard is used to make sure that the development team use the same design and coding conventions. To keep the development team motivated, XP discourages team members from working more than 40 hours a week. In addition, overtime weeks are usually limited to no more than two weeks in a row [1].

2.2. Scrum

Scrum is an agile software development method that inherits many of the characteristics of the traditional iterative and incremental approaches (e.g., [1]). The term "scrum" is originated from the game of rugby, referring to a strategy of getting a ball back into play.

 

Figure 2. Steps of the Scrum process (taken from [14])

Figure 2 describes the activities involved in the Scrum process. The key activity consists of the concept of “sprint”, which consists of a 30-day working period with a set of specific goals. The Scrum process starts with a planning phase, during which a backlog list is developed to define the functionality of one or more releases of the system along with the risk associated with each release. The appropriate risk controls are also determined. The product backlog lists contain the total work of the project to be done. After that, a sprint planning meeting takes place. It usually starts every 15 to 30 days after the planning phase. During this meeting, customers, users, managers, and developers discuss the objectives of the next sprint release and the sprint backlog lists to be completed.

Scrum encourages team members to meet every day for 15 minutes in order to improve the overall knowledge of each member. Non-members may also attend but do not have the right to speak during the meeting, and any further discussions are avoided. These meetings are led by the Scrum master, a role given usually to the project manager. During these meetings, the Scrum master ensures that standards, conventions, and best practices are followed throughout the project. It is also the responsibility of the Scrum master to provide the necessary resources to team members so as to complete the sprint goals.

Scrum requires at least one daily integration and regression test of the code. In addition, a sprint review session of four hours maximum is organized regularly to discuss and report to the manager and the customer what has been

 

50

Authorized licensed use limited to: KING SAUD UNIVERSITY. Downloaded on October 4, 2009 at 07:29 from IEEE Xplore. Restrictions apply.

 

accomplished so far during the sprint. The sprint review session is also a way to receive feedback on regular basis from the various stakeholders involved in the project.

2.3. Crystal Methodologies

The Crystal methodologies are a set of processes that can be applied to different projects depending on the size and the complexity of a project. The framework in Figure 3 includes the factors that influence the selection of a particular methodology. The X-axis indicates staff size while the Y-axis represents the system criticality. The more critical the project, the more rigorous and formal processes are required. Crystal methods define four levels of critically:

n Life (L): A system failure is critical and may cause loss of life.

n Essential money (E): A system failure may cause loss of money.

n Discretionary money (D): A system failure may cause loss of money but can be fixed by referring to the system’s user manual.

n Comfort (C): A system failure may cause a loss of customer comfort.

Crystal methods put an emphasis on a set of policy standards that govern the way the project is managed. These standards are common among all crystal methods [1] and include incremental delivery of releases, progress tracking, direct user involvement, etc.

 

Figure 3. Crystal methods (taken from [14])

Different processes are assigned a different color that represents the heaviness of the process. Currently two crystal methodologies have been defined: Crystal clear and crystal orange.

Crystal clear is designed for small project with a maximum of six developers as shown in Figure 3. It can be used for different levels of criticality. For example, the D6 category 

 

indicates the use of the crystal clear method with a critical level of discretionary money.

The developers should be located in a shared space to improve the commutation between them. Developers can use any tool to improve the overall work, in other word, crystal clear keeps the choices open for developers to choose the appropriate tools. The documentation in crystal clear is very light. The requirements are expressed using UML use cases. The first incremental cycle must not exceed three months and a workshop meeting is usually held after each delivery [11].

Crystal orange is targeted for a project with a maximum of 40 developers. The project duration is usually between one to two years. The crystal orange method is suitable for a project of category D40 and may extend to E40 if necessary. Due to lack of rigorous verification techniques, the crystal orange method is not appropriate for life critical projects. Similar to crystal clear, developers are encouraged to work in a shared space. Crystal orange requires more documentation than crystal clear. For example, the requirements should be expressed in a natural language, as well as, the design document should be clear enough for developers and stockholders. The design documents are expressed using formal specifications such as state chart diagrams. The first incremental delivery must not exceed four months and more formal testing methods are encouraged in crystal orange [1, 11].

2.4. Feature-Driven Development (FDD)

The Feature-Driven Development (FDD) approach focuses on the software features of the system as the main driver of the development process. It differs significantly from the other agile processes by putting a strong emphasis on on planning and upfront design [9].

 

Figure 4. The simple steps of FDD (taken from [14])

As shown in Figure 4, the first step of the FDD process is to build a detailed model of the system to be developed, which captures the stakeholders' assumptions and requirements. Once the domain model is built, the team members print a list of the features of the system. Each feature should be developed in a few hours or days, but no longer than 2 weeks. Using FDD, development teams are formed specifically to design and implement a particular feature. The work is usually performed in parallel on multiple

 

51

Authorized licensed use limited to: KING SAUD UNIVERSITY. Downloaded on October 4, 2009 at 07:29 from IEEE Xplore. Restrictions apply.

 

features. The teams are disbanded once the feature is completed and verified. Each team is led by a feature owner, who is responsible of the code segment that implements the feature. This is contrasted with the XP approach where the ownership of the code belongs to the whole development team and not to a specific member [15].

The FDD process utilizes rigorous inspection guidelines in order to find defects in the system. It also enforces coding standards. It also encourages regular builds on a daily or weekly basis in order to add newly designed features to the baseline system. Due to the fact that features are developed in parallel, it is important to have a configuration management system that allows proper integration of the changes made to the system.

Unique to the FDD approach is a tracking and papering mechanism that assesses the project status based on the number of features that have been implemented as well as the overall progress of the design, coding, and testing activities. Each feature is scored using a value ranging between 0 (for a feature that has not yet been worked on) and 1 (a completed feature) and anything in between refers to a feature in progress.

2.5. Adaptive Software Development (ASD)

Highsmith defines the adaptive software development as “a complex adaptive process that involves interaction between agents (stockholders), environment (organization) and the product (software)” [2]. The ASD process involves essentially three phases that are geared towards planning, re¬assessment and review. These phases are: Speculate, Collaborate, and Learn. During the speculation phase a definition of what to achieve in each iteration is determined. The collaboration phase highlights the importance of team work, by sharing the knowledge among the software developers. The learning phase is carried out after each iteration in order to improve the developer’s expertise as well as to enhance the quality of the work [1].

The ASD life cycle focuses on results rather than the tasks performed to reach these results. The results are mapped directly to form the application features. In other words, the results are combined together to form the overall functionality of the system. Similar to other agile processes, ASD encourages interaction between the developers, managers, and customers. In addition, the ASD approach favors sharing of knowledge among the software developers since building a complex system would typically require the collaboration of multiple domain experts bringing the needed knowledge and experience.

Figure 5 depicts the main phases of the ASD process. ASD starts with a project initiation phase. During this phase the project mission statement is established, which is defined to guides the overall process. It must be clear and well organized. The project normally starts with unclear 

 

requirements but after each adaptive loop the overall mission becomes clearer. ASD is a feature-oriented approach rather than task-oriented. The main focus is always on the features of the systems rather than the tasks needed to implement these features. During the concurrent component engineering phase, the developers may work in parallel to implement one or more features at the same time. One of the most important aspects of ASD is the quality review phase where the customers, developers and managers meet to discuss and assess the overall quality of the work performed. The review phase session, known as the joint application development session (JAD), is important for demonstrating the functionality of the system developed as well as to keep the project within the boundaries of the mission statement. Finally, a quality assurance and release phase is held at the end of the project to fix any problems regarding the quality of the work performed [1].

 

Figure 5. ASD process activities (taken from [14])

ASD is a risk-driven process, that is, the risk management in ASD should start as early as possible in the cycle and should be implemented in all stages, i.e., Speculate, Collaborate, and Learn.

3. Requirements for Software Development Projects

In this section, we discuss key criteria that characterize software development projects against which the above software processes have been compared (see Table 1). The list of requirements is by no means exhaustive, but we believe that it captures the most common requirements found in the literature (e.g., [4, 6, 8, 12, 13]).

3.1. Customer Involvement

Customer involvement is a key practice in all agile processes, as shown in Table 1. From this table, we can clearly state that agile processes consider customers as an integral part of the development process. For example, XP, crystal methodologies, and Scrum require on-site visits to customer’s venues to allow end users to verify and prioritize the requirements during the requirement phase. The

 

52

Authorized licensed use limited to: KING SAUD UNIVERSITY. Downloaded on October 4, 2009 at 07:29 from IEEE Xplore. Restrictions apply.

 

involvement of customers is also reflected during acceptance testing, where most agile processes require these tests to be written and executed by customers.

3.2. Time to Market

Most agile processes favor early delivery of software. The motivation behind this is to allow customers to start using the software system and provide early feedback that would further enhance the upcoming releases. Early deliverables help also improve the customers’ understanding of the expected software features. Agile processes provide a delivery schedule time which varies between two weeks (e.g., in FDD) to almost six months such as in Crystal orange.

3.3. Responding to Change

As changes are costly to accommodate later in the project, the ability to respond rapidly to changes would therefore reduce project risks and costs.

The surveyed agile processes rely on some sort of review sessions involving customers to identify changes in the requirements. XP and crystal methodologies, for example, rely on frequent on-site customer visits and direct user intervention techniques to identify any possible changes triggered by customers. Scrum and ASD rely on daily meetings where customers are involved. FDD, on the other hand, identify changes during the specification of the features that need to be implemented.

3.4. Documentation

The documentation in agile processes is reduced to informal documentation, face to face communication, and on-site customer visits. Using informal documentation reduces the cost associated with maintaining it [4]. The agile processes studied in this paper vary in the level and the type of documentation they provide. For example, XP uses user stories to capture the software features that need to be implemented. Scrum’s main documentation consists of product and spring backlog lists. FDD and Crystal methodologies use UML diagrams such as use cases, class diagrams, and object models to document the design. Test cases have also been used by XP and Crystal methodologies as documentation artifacts.

3.5. Verification and Validation

Agile software processes vary in the amount of verification and validation required. This requirement will enable a project manager to select an agile process that would fit the verification and validation requirements of the software project at hand. For example, a critical safety system will 

 

most likely require extensive testing than other types of software.

Agile processes put a strong emphasis on unit and integration testing. For example, XP is known as a test-driven design approach where test cases for unit and integration testing are written before writing code. In addition to unit testing, regression testing is used in Crystal methodologies and Scrum.

Other quality review techniques are also used such as design and code inspection (in FDD), sprint review (in Scrum), and code quality review (in ASD).

3.6. Team Management

This requirement is important for organizing the team from many prospective such as team size, team communication and the use of standardized procedures (e.g., design conventions), etc. Team size is one of the important factors that may affect the selection of the development process. Although agile processes emphasize a face-to-face communication instead of formal documentation, the number of developers considered is a serious obstacle to the effectiveness of the communication. Except crystal orange, all other agile processes suggest at most 20 persons per team. Team communication considered as the second factor in team management. Agile processes tend to be people-oriented processes by allowing team members to take appropriate decisions when required without being restricted to any procedure or technique.

The use of code standard guidelines has been proposed in XP and Crystal methodologies to facilitate exchange of information among team members. This is also due to the fact that these processes favor collective ownership of the system artifacts. In other words, any member can modify the code or design of someone else. In such cases, standard coding guidelines facilitate the collaborative work.

3.7. Collaborative Work

Some software projects require coordination of project teams located in different regions, countries, and time zones. This is typically the case for large companies that have shifted parts of their software development activities to geographically distributed locations. A software process that is selected for this type of projects needs to allow for smooth collaboration among geographically distributed team members.

Agile processes do not discuss how collaborative work can be performed. This lack of support for collaborative work has also been reported by Turk et al. in [13].

 

53

Authorized licensed use limited to: KING SAUD UNIVERSITY. Downloaded on October 4, 2009 at 07:29 from IEEE Xplore. Restrictions apply.

 

Table1: Mapping agile process to software development project requirements

 

 

4. Conclusion and Future Directions

In this paper, we presented our analysis of five agile software processes and compared them based on criteria relate to software development projects. The objective is to help software engineers and project managers understand the key characteristics of these processes and therefore select the most suitable process with respect to the type of software projects they develop.

As future work, there is a need to include other agile processes not covered in this paper such as the Dynamic Systems Development Method (DSDM), the Lean Software Development (LSD), etc.

5. References

[1] P. Abrahamsson, O. Salo, J. Ronkainen, and J. Warsta, “Agile software development methods: Review and Analysis”. Espoo, Finland: Technical Research Centre of Finland, VTT Publications 478.

[2] J. A. Highsmith. Adaptive Software Development: A Collaborative Approach to Managing Complex Systems. Dorset House, New York 2000.

[3] G. Cugola and C. Ghezzi, "Software Processes: a Retrospective and a Path to the Future", In Proc. of the Software Process Improvement and Practice Conference, 1998, pp. 101-123.

[4] M.Coram and S.Bohner, "The Impact of Agile Methods on Software Project Management" In Proc. of the 12th International Conference and Workshops on the Engineering of Computer-Based Systems, 2005, pp. 363-370. 

 

[5] P. Abrahamsson, J. Warsta, M. T. Siponen, J. Ronkainen, "New Directions on Agile Methods: A Comparative Analysis", In Proc. of the International Conference on Software Engineering, 2003, pp.244- 254.

[6] M.Huo, J.Verner, L.Zhu, M.A.Babar, "Software Quality and Agile Methods”, In Proc, of the 28th Annual International Computer Software and Applications Conference, 2004, pp. 520-525.

[7] M. Fowler, K. Beck, J. Brant. Refactoring: Improving the Design of Existing Code. Addison-Wesley, 1999.

[8] S. Nerur, R. Mahapatra and G.Mangalaraj, “Challenges of migrating to agile methodologies”. Communications of the ACM, New York, 2005, pp. 72-78/

[9] S. R. Palmer and J. M. Felsing, A Practical Guide to Feature-Driven Development, Prentice Hall PTR, 2002.

[10] B. Boehm, R. Turner, “Observations on balancing discipline and agility”, In Proc. of the Agile Development Conference, 2003, pp. 32- 39.

[11] A. Cockburn. Crystal Clear: A Human-Powered Methodology for Small Teams. Addison-Wesley Professional, 2004.

[12] K. Molokken-Ostvold, K. M. Furulund, “The Relationship between Customer Collaboration and Software Project Overruns”, In Proc. of the AGILE 2007, 2007, pp 72¬83.

[13] D. Turk, R. France, and B. Rumpe, “Limitations of agile software processes”. In Proc. of the 3th International Conference on eXtreme Programming and Agile Processes in Software Engineering, 2002, pp. 43-46.

[14] D-B.Cao, “An Empirical Investigation of Critical Success Factors in Agile Software Development Projects", PhD thesis, Capella University, USA, 2006.

 

54

Authorized licensed use limited to: KING SAUD UNIVERSITY. Downloaded on October 4, 2009 at 07:29 from IEEE Xplore. Restrictions apply.

 

CV Michael Kuperberg

Personal data

Private address: Am Wald 40, 76149 Karlsruhe, Germany

Private phone: +49 721384-3825, mobile +49 163 806-4228

E-Mail: kuperberg@web.de

Citizenship: German

Family status: unmarried

Date of birth: June 23rd, 1981

Work experience

11.2010 – now Manager of the Informatics Innovation Center (IIC) at the Karlsruhe In-stitute of Technology (KIT), in charge for teaching and research on IBM System z in cooperation between KIT, IBM and FZI Karlsruhe. Involved in the acquisition of third-party funds over public-private partnerships.

03.2006 – 11.2010 Researcher and PhD candidate at the chair of Prof. Reussner, KIT. Multiple teaching and organizational responsibilities in addition to research activities.

08.2004 – 02.2006 Student research assistant (analyzing JTA transaction managers) in the Con-troMAS project at the chair of Prof. Lockemann, University of Karlsruhe.

09.2003 – 05.2004 Teaching assistant in mathematics and computer science at the UMass Boston.

08.2002 – 08.2003 Working student at SAP AG (Walldorf, Germany): working in a team de¬veloping GUI-supported Python tests for a search engine built by SAP AG.

07.2000 – 08.2002 Working student at the asknet AG in Karlsruhe, Germany. Tasks: security, smartcard and cryptography (PKCS) programming in Java.

Technologies and tools portfolio

07.2000 – now 10+ years Java programming experience incl. JTA/JTS, JNI, Reflection etc.

Software performance Palladio Component Model and its performance prediction tooling; JProfiler; VisualVM; Jamon; JavaSimon; perf4j; Dtrace and Apple Instruments

JVM analysis JMX; JConsole; JVMTI (JVM tooling interface); JIT compiler internals

Bytecode engineering ASM bytecode engineering library; Javassist; java.lang.instrument

Other languages/tools Author of TIMERMETER and BYCOUNTER; contributed to Faban; experi-ence with C#, C/C++, Scala, X10, MPI and Weka data mining framework

Dissertation and university education

03.2006 – 11.2010 Dr.-Ing. magna cum laude in Computer Science at the Department of Infor¬matics, Karlsruhe Institute of Technology (KIT). Dissertation title: “Quanti¬fying and Predicting the Influence of Execution Platform on Software Compo¬nent Performance”, Advisors: Prof. Ralf Reussner and Prof. Walter F. Tichy. 17 peer-reviewed publications (see full list starting on page 3).

10.2000 – 11.2005 Undergraduate and graduate studies in informatics (computer science) at the University of Karlsruhe, diploma grade “A” (“sehr gut”). Diploma thesis on using rule-based multi-agent systems for controling and monitoring produc¬tion scheduling. Minors in operations research and abstract algebra.

University education continues on the next page 

 

Karlsruhe, 4th May 2011 Page 1 of 4

 

CV Michael Kuperberg

08.2003 – 06.2004 Baden-W¨urttemb erg state scholarship for abroad studies (1 year) at the Uni-versity of Massachusetts in Boston as graduate student. First publication.

Professional functions, community service and memberships

2011 Session Chair at ICPE’11 (Intern. Conference on Performance Engineering).

2010 – now Elected officer and release manager of the SPEC RG (Standard Performance Evaluation Corporation Research Group), member of the steering committee.

March and April 2010 Lightning Talks “Exact Timing Measurements in Java” and “Performance measurements with perf4j” at the Java User Group Karlsruhe, Germany.

2009 – now Member of ACM, IEEE, the German Informatics Society (GI e.V.) and VKSI.

2008 Copy editor of the “Dependability Metrics” Springer LNCS volume 4909.

March 29th, 2008 Workshop host for FESCA’08 (Formal Foundations of Embedded Software and Component-Based Software Architectures), Budapest, Hungary.

2006 – now Reviewer for the IEEE TSE journal (Transactions on Software Engineering) as well as for QoSA, CBSE and other conferences and workshops.

10.2001 Tutor during the freshmen orientation phase of the computer science depart-ment of the University of Karlsruhe.

Awards and prizes

02.2011 Co-recipient of the “Best Lecture Award” for the lecture “Component-based software development” (as voted by students during the teaching evaluation at the KIT), taught during summer term 2010.

06.2009 Co-recipient of “Best Paper Award” for the QoSA 2009 conference paper “Improved Feedback for Architectural Performance Prediction using Software Cartography Visualizations” (published in Springer LNCS 5581, pages 52–69)

1999, 1997 and 1996 Prizes for Germany-wide and statewide mathematics competitions. Teaching at university level

Summer term ’11 Planning, development and teaching of two new courses at the Informatics Innovation Center (IIC), KIT: project group on software development for IBM System z and a lab course on performance analysis of applications on IBM z10.

Winter term ’10/’11 Lecturer on Scala and X10 programming languages as well as MPI (Message Passing Interface) in the lecture “Programming paradigms”.

Summer term ’10 Teaching assistant for the award-winning lecture “Component-based soft¬ware development” (voted by students as the best elective lecture of the department of computer science during the teaching evaluation at the KIT).

Winter term ’09/’10 Teaching assistant for the lecture “Software Architecture” (also winter terms ’06/’07, ’07/’08 and ’08/’09), authoring its “Middleware/SOA” part.

2006 – now Advisor of 5 diploma/master theses, 7 study/bachelor theses, 10 lab course participants and 12 seminar participants.

Karlsruhe, 4th May 2011 Page 2 of 4

 

CV Michael Kuperberg

School and college education

06.2000 Abitur at the “Goethe-Gymnasium” in Karlsruhe as top of the class, average grade “excellent” (1.0) with distinction, majors: mathematics and physics.

09.1999 – 06.2000 Coorganization of the transportation study group at the Goethe-Gymnasium.

1999 3rd prize in Germany-wide mathematics competition.

09.1997 School exchange with Lyc´ee Michelin, Marseille, France.

1997 and 1996 2nd and 3rd prize in statewide mathematics competition for high schools.

Miscellaneous

Languages: German (native language), English (fluent), French (business fluent), Russian (native language), Ukrainian (understanding and reading)

Driving license: available

Interests and hobbies: Architectural photography and history of architecture;

public transportation and high-speed trains;

playing piano; volleyball; geotagging

Peer-reviewed publications in journals and conference proceedings

Top 3 publications highlighted in red; publications listed in reverse chronological order:

1. Ginpex: Deriving Performance-relevant Infrastructure Properties Through Goal-oriented Experiments (Michael Hauck, Michael Kuperberg, Nikolaus Huber, and Ralf Reuss-ner), accepted for publication in QoSA 2011: Proceedings of the Seventh International ACM Sigsoft Conference on the Quality of Software Architectures, to appear 2011.

2. Metric-based Selection of Timer Methods for Accurate Measurements (Michael Ku-perb erg, Martin Krogmann, and Ralf Reussner), in ICPE’11: Proceedings of the 2nd ACM / SPEC International Conference on Performance Engineering, 2011.

3. Analysing the Fidelity of Measurements Performed With Hardware Performance Counters (Michael Kuperberg and Ralf Reussner), in ICPE’11: Proceedings of the 2nd ACM / SPEC International Conference on Performance Engineering, 2011.

4. Using Invariant Analysis For Improving Instrumentation-based Performance Evala-tion Of SPECjvm2008 Benchmarks (Michael Kuperberg, Martin Krogmann, and Ralf Reussner) in Proceedings of the International SPEC Benchmark Workshop 2010, 2010.

5. Using Genetic Search for Reverse Engineering of Parametric Behaviour Models for Performance Prediction (Klaus Krogmann, Michael Kuperberg and Ralf Reussner), in IEEE Transactions on Software Engineering Vol. 36, No. 6, 2010.

6. Automated Benchmarking of Java APIs (Michael Kuperberg and Fouad Omri), in Pro¬ceedings of Software Engineering 2010 (SE2010), Feb. 2010.

7. Modelling Layered Component Execution Environments for Performance Prediction (Michael Hauck, Michael Kuperberg, Klaus Krogmann and Ralf Reussner), in Proceedings of the 12th International Symposium on Component Based Software Engineering (CBSE 2009), LNCS 5582, Pages 191–208. Springer-Verlag Heidelberg, 2009.

8. Improved Feedback for Architectural Performance Prediction using Software Car¬tography Visualizations (Klaus Krogmann, Christian M. Schweda, Sabine Buckl, Michael Ku-perberg, Anne Martens and Florian Matthes), in Architectures for Adaptive Systems (Proceeding

Karlsruhe, 4th May 2011 Page 3 of 4

 

CV Michael Kuperberg

of QoSA 2009), LNCS 5581, pages 52–69. Best paper award. Springer-Verlag Heidelberg, 2009.

9. TimerMeter: Quantifying Accuracy of Software Timers for System Analysis (Michael Kuperberg, Martin Krogmann and Ralf Reussner), in Proceedings of the 6th International Con¬ference on Quantitative Evaluation of SysTems (QEST) 2009, IEEE Computer Society, Wash¬ington, DC, USA.

10. Using Heuristics to Automate Parameter Generation for Benchmarking of Java Methods (Michael Kuperberg, Fouad Omri and Ralf Reussner), in Proceedings of the 6th In-ternational Workshop on Formal Engineering approaches to Software Components and Archi¬tectures, York, UK, (ETAPS 2009, 12th European Joint Conferences on Theory and Practice of Software), March 2009.

11. Performance Prediction for Black-Box Components using Reengineered Parametric Behaviour Models (Michael Kuperberg, Klaus Krogmann and Ralf Reussner), in Proceedings of the 11th International Symposium on Component Based Software Engineering (CBSE 2008), LNCS 5282, pages 48–63. Springer-Verlag Heidelberg, 2008.

12. Reverse Engineering of Parametric Behavioural Service Performance Models from Black-Box Components (Klaus Krogmann, Michael Kuperberg and Ralf Reussner), in MDD, SOA und IT-Management (MSI 2008), Pages 57–71, GITO-Verlag, Berlin, 2008.

13. ByCounter: Portable Runtime Counting of Bytecode Instructions and Method In-vocations (Michael Kuperberg, Martin Krogmann and Ralf Reussner), in Proceedings of the 3rd International Workshop on Bytecode Semantics, Verification, Analysis and Transformation, ETAPS 2008, 11th European Joint Conferences on Theory and Practice of Software, Budapest, Hungary, April 2008.

14. Markov Models (Michael Kuperberg), in Dependability Metrics, LNCS 4909, Springer-Verlag Heidelberg, 2008.

15. Predicting Software Component Performance: On the Relevance of Parameters for Benchmarking Bytecode and APIs (Michael Kuperberg, Steffen Becker), in Proceedings of the 12th International Workshop on Component-Oriented Programming, Berlin, July 2007.

16. Influence of Execution Environments on the Performance of Software Components (Michael Kuperberg), in Proceedings of the International Research Training Groups Workshop 2006, GITO-Verlag, Berlin, 2006.

17. Metric Incremental Clustering of Nominal Data (Dan Simovici, Namita Singla and Michael Kuperberg), in Proceedings of International Conference on Data Mining 2004, Brighton, UK, 2004.

 

Karlsruhe, 4th May 2011 Page 4 of 4

 

Outline for publication

1

SCHEME OF AID AND ADVICE ON HOME MORTGAGE ARREARS

OUTLINE OF THE SCHEME

1. OVERVIEW

The Government has put in place an innovative new Scheme to help people who are struggling to pay their debts, and are at risk of losing their homes due to mortgage arrears.

The overall objective of the Scheme is to ensure that a person in this situation can access free, independent expert financial and legal advice and support, which will help them to identify and put in place their best option to get back on track. Priority is given to finding solutions which will allow the person to remain in their home, wherever that is a sustainable option.

Government Departments and bodies are working together to deliver the Scheme, which is coordinated by the Department of Justice and Equality and the Department of Social Protection, and implemented by the Money Advice and Budgeting Service (MABS), working with the Insolvency Service of Ireland (ISI), the Legal Aid Board, and the Citizens’ Information Board.

Help is already available under the Scheme, which issued its first vouchers for free advice on 22 July 2016. Contact the MABS Helpline at 076-1072000, or the local MABS office, for help and information. 

The Scheme is an important component of the new national Mortgage Arrears Resolution Service announced under the Programme for Government. It is expected to be formally launched by the Tánaiste and Minister for Justice and Equality, Frances Fitzgerald TD, and by the Minister for Social Protection, Leo Varadkar TD, in October 2016.

The Scheme is being put in place for a maximum duration of three years from July 2016, at an expected total cost of up to € 15 million. This includes funding for a significant information and promotion campaign, targeted to ensure that those most in need of help are aware of the supports available, which is expected to roll out in autumn 2016.

The Scheme also builds on recent legal changes and on previous Government initiatives to help people to recover from indebtedness and return to solvency, particularly in relation to home mortgage arrears.

2. HOW DOES THE SCHEME WORK?

The Scheme is designed to help borrowers who are insolvent and at risk of losing their homes, to get the help they need as quickly and effectively as possible. Aid and advice provided under the Scheme are provided free of charge to eligible borrowers.

 

Outline for publication

2

Anyone interested in seeking help under the Scheme should start by contacting MABS. They can call the MABS Helpline for help and information, at 076-1072000 (Monday-Friday, 9 a.m. to 8 p.m.), or call to their local MABS office (there are over 50 MABS offices around the country). Details are also available on www.keepingyourhome.ie .

MABS is the Government’s ‘one-stop shop’ point of information and guidance for homeowners in mortgage arrears, and will be the single universal gateway for aid and advice under the Scheme. All applications for aid and advice under the Scheme must go through MABS.

MABS will quickly assess the borrower’s financial situation. If the borrower is eligible for the Scheme (see section 5 below), MABS will direct them to the expert likely to be best placed to help them. Depending on the borrower’s situation, this expert may be a dedicated mortgage arrears adviser in MABS, or it may be a personal insolvency practitioner (‘PIP’), a solicitor, or an accountant. If MABS directs the borrower to a professional adviser, it will give the borrower a voucher for free provision of the relevant services.

Priority is given to obtaining financial analysis and advice, as this is the most likely to identify a long-term financial solution which can help the borrower to get out of debt and get back on track.

Once the financial analysis and advice is available, the borrower can also be referred to a ‘consultation solicitor’, for advice on any legal issues which arise.

The Scheme will also provide a new ‘duty solicitor’ service at all Circuit Court repossession hearings, to help unrepresented borrowers who are receiving advice under the Scheme.

In addition, the Scheme can provide legal aid for a borrower who considers that their creditors have unfairly refused a personal insolvency proposal which includes the home mortgage arrears, to seek review of that refusal by the Courts.

The Scheme provides five different fixed packages of professional financial or legal advice and help, all free of charge to the borrower. These are: the PIP Service, the Accountant Service, the Consultation Solicitor Service, the Duty Solicitor Service, and the Personal Insolvency Court Review Service. Each of these is explained in more detail below, at section 6.

MABS can refer a borrower to one or more of these services, depending on their needs. The intention is that a borrower in need can be routed as quickly as possible to the expert best placed to help them, and that expert advisers can work together to help the borrower to get the best available solution into place.

Advice under the Scheme can cover all options for resolving the borrower’s home mortgage arrears. Depending on the borrower’s individual circumstances, these options may include restructure, personal insolvency, bankruptcy, sale or surrender of the home. The decision on what option to take remains the borrower’s, but the Scheme aims to ensure that the borrower can make that decision with the benefit of independent expert advice.

For example, a borrower in mortgage arrears may turn out to be able to pay their debts, once they can access advice and help from MABS. Another borrower may be insolvent, yet able

 

Outline for publication

3

to keep their home under a personal insolvency arrangement or alternative restructuring arrangement, once they have access to expert financial advice. Or the borrower may not have the financial capacity to keep their home, or may not wish to keep their home, and may need independent advice and assistance on their best alternative options.

3. WHO ARE THE ADVISERS?

MABS has put in place a new network of in-house Dedicated Mortgage Arrears (DMA) advisers, who are located in MABS offices around the country. These DMA Advisers are able to provide a high level of independent expert advice to people in difficulty with mortgage arrears.

Under the Scheme, professional panels have also been established, consisting of qualified and regulated personal insolvency practitioners (‘PIPs’) 1, accountants 2 and solicitors, who have appropriate knowledge of mortgages, insolvency, bankruptcy and/or repossession. These panels are established and supervised respectively by the Insolvency Service of Ireland, the professional accountancy bodies and the Citizens’ Information Board, and the Legal Aid Board.

The professionals participating in the panels are private practitioners who have agreed to provide borrowers with independent advice and help under the Scheme, free of charge to the borrower, and at fixed cost to the Scheme. The borrower can choose their advising PIP, solicitor or accountant from the Scheme panels, subject to availability.

In order to provide help to the borrower as quickly and efficiently as possible, any adviser providing analysis and advice to them under the Scheme may share that information with any other expert who is advising that borrower under the Scheme. This will be done only with the borrower’s prior agreement, and only refers to confidential sharing of information within the Scheme, between any experts the borrower has chosen to advise them.

4. OBJECTIVES OF THE SCHEME

The detailed objectives of the Scheme are as follows:

• To provide access to independent expert advice (both financial and legal) to borrowers in serious mortgage arrears on their homes.

• To address fear and stress among many unrepresented borrowers, provide assistance to them, and ensure that sustainable solutions can be identified and put in place as early as possible.

1 PIPs are specialists in debt advice and personal insolvency solutions, and can also give financial advice on bankruptcy. They are supervised and regulated by the Insolvency Service of Ireland.

2 The solicitors’ panel and the personal insolvency practitioners’ panel are already operational. The accountancy panel is currently being established (August 2016) and is expected to become operational very shortly.

 

Outline for publication

4

• To help those in mortgage arrears who have already been issued with repossession proceedings against their homes, out of those proceedings and into alternative solutions, where that is possible.

• To address the high number of borrowers in serious home mortgage arrears, particularly those in arrears of more than 720 days, and to ensure that the home mortgage cases currently in arrears of less than 720 days are helped not to progress beyond that point.

• To support earlier, more effective and more widespread constructive engagement between borrowers and lenders where there are serious arrears on a home mortgage – right from the point where the borrower receives a proposal for surrender or other loss of the home, or is deemed ‘non-cooperating’ by the lender - so that the borrower may be assisted ‘upstream’ of the issue of repossession proceedings.

• To provide solicitor advice and assistance to those homeowners in respect of whom either surrender of ownership, or court-ordered repossession, are likely.

• To support the Courts in ensuring fair process for large numbers of otherwise unrepresented defendants in repossession actions, support MABS in dealing with cases requiring more complex financial advice, and support MABS (or PIPs on referral from MABS) in dealing with legal aspects of debt and repossession.

• To ensure access to legal aid for the new Court review under s. 115A Personal Insolvency Acts, where creditors reject a Personal Insolvency deal proposed by a Personal Insolvency Practitioner on behalf of an insolvent person, which includes mortgage arrears on their home.

5. WHO IS ELIGIBLE?

A person will be eligible for advice and assistance under the Scheme if they:

a) are insolvent (as defined under the Personal Insolvency Act 2012: this simply means that the person is ‘unable to pay their debts in full as they fall due’), and

b) are in mortgage arrears on the home in which they normally reside (their principal private residence), and

c) are at risk of losing their home (for example, they may have received from their mortgage lender repossession proceedings, a letter indicating that such proceedings will issue, a letter indicating that they are deemed non-cooperating, or an invitation to consider sale, surrender or other loss of all or part of the home).

A person will be eligible for the Personal Insolvency Court Review Service under the Scheme if they satisfy conditions (a) and (b) above, and they have already made a proposal for a Personal Insolvency Arrangement including their home, but that proposal has been rejected by their creditors. In this case, the Personal Insolvency Practitioner concerned must also certify to the Legal Aid Board that the borrower satisfies the conditions required under s. 115A of the Personal Insolvency Acts for seeking Court review.

A person will not be eligible for the Scheme, if the above conditions are satisfied, but the home is disproportionate to the reasonable living accommodation needs of the borrower

 

Outline for publication

5

and his or her dependents, taking account of the criteria set out at section 104 of the Personal Insolvency Act 2012.

The Scheme focuses on risks to the borrower’s home, and therefore does not cover a person whose mortgage arrears relate only to a buy-to-let property. However, a person in mortgage arrears on their home, who also owns a buy-to-let, will still be eligible, if he or she satisfies the conditions above.

The eligibility criteria set out above are designed to ensure that a person in serious mortgage arrears can be ‘triaged’ quickly. This will help MABS to easily identify the most suitable likely pathway to resolving their indebtedness, and to refer the borrower quickly to the appropriate expertise. The borrower’s financial situation will be examined in full detail when their Financial Statement (a detailed written analysis of the borrower’s financial situation) is being prepared. The initial triage may be revised by MABS, if information later comes to hand which makes this appropriate.

6. WHAT SORT OF ADVICE AND HELP IS COVERED?

The five services available under the Scheme are explained in detail below.

a. The PIP Service

If the borrower wishes to explore their options under personal insolvency, MABS can give them a voucher for free advice and help from a personal insolvency practitioner (‘PIP’), who is a member of the Scheme panel. The borrower can choose their PIP from the Scheme panel, subject to availability.

The voucher will entitle the borrower to a face-to-face consultation with a PIP, who will carry out a full assessment of the borrower’s financial situation, explain to them all the available options and the best option for them to deal with their mortgage arrears, and confirm that advice to the borrower in writing. As part of this service, the PIP will also prepare, with the borrower’s co-operation, a ‘Prescribed Financial Statement’ (PFS), which is the detailed written analysis of the borrower’s financial situation, as required under the Personal Insolvency Acts.

If the panel PIP advises that the borrower’s best option is personal insolvency, they will also help them in taking the next steps. If the PIP advises that the borrower’s best option is bankruptcy, they will also provide the borrower, under the voucher, with the certificate required by the bankruptcy court, confirming that the borrower has first been advised regarding their options under personal insolvency.

b. The Accountant Service

If the borrower’s case involves more complicated financial issues, but is not suited to personal insolvency, MABS may direct them for financial advice to an accountant who is a member of the Scheme panel. The borrower can choose their accountant from the Scheme panel, subject to availability.

To get free advice from an accountant under the Scheme, the borrower will need to first work with a MABS Adviser to complete the Standard Financial Statement (SFS) required by lenders under the MARP (Mortgage Arrears Resolution Process). The accountant will need a copy of this, in order to give them financial advice.

 

Outline for publication

6

The voucher will entitle the borrower to a face-to-face consultation with the panel accountant, who will advise them on any financial issues relating to resolution of their mortgage arrears and on possible solutions, and will confirm that advice to them in writing.

c. The Consultation Solicitor Service

After the borrower has completed their financial statement (SFS or PFS- see above) and received financial advice under the Scheme, their financial adviser (PIP, MABS adviser or accountant) may recommend that they also consult a solicitor from the Scheme panel, if the case raises legal issues. For example, the borrower may have received correspondence from their lender, seeking repossession or proposing surrender of their home.

The financial adviser can then apply to MABS for a legal advice voucher for the borrower. The borrower can choose their solicitor from the Scheme panel, subject to availability. The solicitor will need to see any relevant documents in the borrower’s case, and particularly their up to date SFS or PFS (see above), and any written financial advice they have received under the Scheme.

The legal advice voucher will entitle the borrower to a face-to-face consultation with a solicitor, who will assess and explain their legal position, and advise them on any repossession proceedings and any legal issues arising for resolving their mortgage arrears. The solicitor will also confirm that legal advice to the borrower in writing. If repossession proceedings have already been issued by the lender, the consultation solicitor may in certain circumstances be able to negotiate on the borrower’s behalf to settle the proceedings out of court.

d. The Duty Solicitor Service

If the borrower is facing Circuit Court repossession proceedings against their home due to mortgage arrears, they may be able to get some help at Court from the Duty Solicitor.

The Duty Solicitor is a solicitor from the Scheme’s solicitor panel, who will normally be on duty at a Circuit Court on the date where the Court is due to deal with repossession cases. The borrower cannot choose their Duty Solicitor, who will be the panel solicitor assigned by the Legal Aid Board to be present at the Court on that date.

Normally, the borrower should already have obtained written financial and legal advice under the Scheme, including advice from the consultation solicitor on any repossession proceedings. This will allow the Duty Solicitor to help the borrower more effectively.

If the borrower has not applied in time to do this, the Duty Solicitor may still be able to give them some more limited help, but only if they have already applied to MABS. Any borrower who has already received repossession proceedings against their home from their mortgage lender, and hasn’t yet applied for the Scheme, should contact MABS as soon  as possible (076 107 2000).

The Duty Solicitor Service is a limited service to provide advice and help to unrepresented borrowers at court. The borrower’s consultation solicitor will explain this in more detail.

 

Outline for publication

7

The Duty Solicitor may be able to speak for the borrower in Court, to explain what steps they are taking to try and deal with their mortgage arrears, and may be able to apply for the court proceedings to be adjourned, if the borrower is trying to put a solution in place.

The Duty Solicitor may also be able to explain to the borrower what is happening in the proceedings. However, he or she may be Duty Solicitor for a number of Scheme borrowers whose repossession cases are listed on that day, so the borrower should make sure to attend Court well in advance of the time shown on the Court papers, and should ask for MABS when they arrive. (MABS Advisers also attend at the Circuit Court on repossession hearing dates. They can provide borrowers with information and support, but they cannot provide legal advice or speak for the borrower in Court.)

The Duty Solicitor cannot act as the borrower’s legal aid solicitor, or defend the repossession proceedings on their behalf. This Scheme does not cover legal aid for defending repossession proceedings. If a borrower is in mortgage arrears on their home, the lender will normally have a legal right to seek to take back the home, and the borrower may not have a valid legal defence to prevent repossession. The consultation solicitor will advise them about this. In a small number of situations, the borrower may have a valid legal defence to repossession, but this applies to relatively few cases. If they do have a valid legal defence and want to apply for legal aid, they will need to apply for that to the Legal Aid Board, separately from this Scheme, and different conditions may apply. The consultation solicitor can explain this to the borrower.

e. The Personal Insolvency Court Review Service

The borrower will only need this service if they have already worked with a PIP, have proposed a Personal Insolvency Arrangement (‘PIA’) to their creditors, and the creditors have refused that proposal although the borrower’s PIP considers it fair and reasonable to all concerned.

Under the change made to the Personal Insolvency Acts in 2015, a borrower in this situation can ask the Courts to review the PIA proposal. If the Court agrees that the proposal is overall fair and reasonable, using the criteria set out in section 115A of the Personal Insolvency Acts, it has power to impose the proposal on the creditors.

Under the Personal Insolvency Court Review Service, the Legal Aid Board can provide the borrower with legal representation by a solicitor and barrister to make the Court review application, as part of their Civil Legal Aid service.

For this service, the borrower’s PIP applies on their behalf to the Legal Aid Board, using a Scheme reference number for the borrower provided by MABS. The PIP must also certify to the Legal Aid Board ‘that the borrower has reasonable grounds for seeking the court review and satisfies the other conditions for review laid down by the Personal Insolvency Acts.’

If the Legal Aid Board is satisfied with the application, it can then issue a Legal Aid Certificate for the borrower. The Legal Aid Board is aware of the time limits under the Personal Insolvency Acts for seeking the Court review, and will decide quickly on any fully completed application for legal aid.

 

Outline for publication

8

If the Legal Aid Board grants a Certificate, the borrower and their PIP can select a solicitor from the Scheme solicitors’ panel to act on their behalf. (Under the Personal Insolvency Acts, it is the borrower’s PIP who must apply to Court on their behalf for the review.) The Legal Aid Board will normally also cover the cost of a barrister from the Civil Legal Aid barristers’ panel. Fixed payment rates apply.

 

August 2016

 

Sri Vasavi College, Erode Self-Finance Wing 3rd February 2017

National Conference on Computer and Communication NCCC’17

http://www.srivasavi.ac.in/ nccc2017@gmail.com

Smokestack Fabric Defects Detection

Ms.P.Banumathi

Asst Professor,Department Of Computer Applications,

Tiruppur Kumaran college for Women,Tirupur

banuctech@gmail.com , 8248792149.

Dr.P.R.Tamil Selvi

Asst Professor,Department Of Computer science,

Govt Arts and Science college,Komarapalayam.

selvipr2003@gmail.com , 9942055733.

 

ABSTRACT- Quality inspection is an important aspect of modern industrial manufacturing. In textile industry Production, automate fabric inspection is important for maintain the fabric quality. For a long time the fabric defects inspection process is still carried out with human visual inspection, and thus, insufficient and costly. Therefore, Smokestack fabric defect inspection is required to reduce the cost and time waste caused by defects. The development of fully automated web inspection system requires robust and efficient fabric defect detection algorithms. The detection of local fabric defects is one of the most intriguing problems in computer vision. Texture analysis plays an important role in the automated visual inspection of texture images to detect their defects. Various approaches for fabric 

 

defect detection have been proposed in past and the purpose of this paper is to categorize and describe these algorithms.

Index Terms: Fabric Defect, Defect Classification,MATLAB,smokestack

I. Introduction

The textile industry, as with any industry today, is very concerned with quality. It is desirable to produce the highest quality goods in the shortest amount of time possible. Fabric faults or defects

are responsible for nearly 85% of the defects found by the garment industry[6]. Manufacturers recover only 45 to 65 % of their profits from seconds or off-quality goods. In this paper a fabric faulty part is taken for analysis from textiles. It is imperative, therefore, to detect, to identify, and to prevent these defects from reoccurring. There is

 

All Rights Reserved ©2017 International Journal of Computer Science (IJCS Journal) &

Department of Computer Science, Sri Vasavi College, Erode, Self-Finance Wing, Erode, Tamil Nadu, INDIA

Published by SK Research Group of Companies (SKRGC) - Scholarly Peer Reviewed Research Journals

http://www.skrgcpublication.org/ Page 1122

 

Sri Vasavi College, Erode Self-Finance Wing 3rd February 2017

National Conference on Computer and Communication NCCC’17

http://www.srivasavi.ac.in/ nccc2017@gmail.com

 

a growing realization and need for an automated woven fabric inspection system in the textile industry. All faults present on fabrics such as hole, scratch, dirt spot, fly, crack point, color bleeding etc. In this paper we analyze the faults using image processing technique. Hence the efficiency is also reduced in this process. Image processing techniques will help to production increase in fabric industry; it will also increase the quality of product. They have to detect small detail that can be located in wide area that is moving through their visual field. For this process we have use MATLAB 7.10 in imageprocessing toolbox. The high cost, along with other disadvantages of human visual inspectionhas led to the development of on-line machine vision systems that are capable of performing inspection tasks automaticallyFabric defectsIt is imperative to detect, to identify and to prevent these defects from reoccurring. There are many kinds of fabric defects[11]. Much of them are caused by machine malfunctions and havethe orientation along pick direction (broken pick yarns or missing pick yarns), they tend to 

 

be long and narrow. Other defects are caused by faulty yarns or machine spoils. Recently, the fault detection is done manually after a sufficient amount of fabric has been produced, removed from the production machine and then batched into larger rolls and then sent to the inspection frame.An optimal solution for this would be to automatically inspect from the fabric as it is being produced and to alert the maintenance personnel when the machine needs attention to preventproduction of defects or to change process parameters [12].This is done by identifying the faults in fabric using the image processing techniques and then based on the dimension of the faults; the fabric is classified and accordingly then graded. Some of the commonly occurring

fabric defects

are:

 

Fig.1 Yarn Defects & Weaving Defects

 

All Rights Reserved ©2017 International Journal of Computer Science (IJCS Journal) &

Department of Computer Science, Sri Vasavi College, Erode, Self-Finance Wing, Erode, Tamil Nadu, INDIA

Published by SK Research Group of Companies (SKRGC) - Scholarly Peer Reviewed Research Journals

http://www.skrgcpublication.org/ Page 1123

 

 

Sri Vasavi College, Erode Self-Finance Wing 3rd February 2017

National Conference on Computer and Communication NCCC’17

http://www.srivasavi.ac.in/ nccc2017@gmail.com

 

Whenever an image is converted from one form to another many types of noise can be present in the image. Here we use the Adaptive filtering to reduce stationary noise. It filters an intensity image that has been degraded by constant power additive noise. It uses pixel wise adaptive wiener method based on statistics estimated from a local neighborhood of each pixel. Thresholding

Thresholding is a process of converting a gray scale input image to a bi-level image by using an optimal threshold. The purpose of thresholding is to extract those pixels from some image which represent an object (such as graphs, maps). Though the information is binary the pixels represent a range of intensities. Here adaptive thresholding is used. In adaptive thresholding, different threshold values for different local areas are used to represents the objects.

Histogram Equalization

Histogram is a representation of the distribution of coloring an image and it represents the number of pixels that have colors in each of a fixed list of color ranges. Histogram equalization 

 

is a method for stretching the contrast by uniformly distribution the gray values enhances the quality of an image useful when the mage is intended for viewing.

V. Results

Following test image of a detected image has been used for defect identification. The image has

been exposed to histogram equalization algorithm for thresholding. The thresholding image is brought under noise removal program, where the uneven weaving is detected as spots shown in fig.3.

VI. Conclusion

The Fabric Defect detection and location identification in the normal fabrics defines the faults by this method[11]. This method classifies 85% of defect in fabric and locates the defect in the normal fabric at an acceptable rate and provides 80% classification accuracy. In the binary output image local defects appear segmented from the background. One of the most important advantages of the method is that it is multipurpose without requiring any adjustment. The versatility of the method has

 

All Rights Reserved ©2017 International Journal of Computer Science (IJCS Journal) &

Department of Computer Science, Sri Vasavi College, Erode, Self-Finance Wing, Erode, Tamil Nadu, INDIA

Published by SK Research Group of Companies (SKRGC) - Scholarly Peer Reviewed Research Journals

http://www.skrgcpublication.org/ Page 1125

 

Sri Vasavi College, Erode Self-Finance Wing 3rd February 2017

National Conference on Computer and Communication NCCC’17

http://www.srivasavi.ac.in/ nccc2017@gmail.com

 

been demonstrated not only by its applicability to different regular textures but also, for a given texture, the method allows to detect a variety of defects.

References

[1] Kaur Navneet and Dalal Mandeep “Application of Machine Vision Techniques in Textile (Fabric) Quality Analysis”, IOSR Journal of Engineering, Vol. 2, No.4, pp. 582-584, Apr. 2012.

[2] Xie Xianghua, “A Review of Recent Advances in Surface Defect Detection using Texture analysis Techniques”, Electronic Letters on Computer Vision and Image Analysis, Vol.7, No.3, pp.1-22, 2008.

[3] J.Wang, R.A. Campbell and R.J. Harwood, “Automated inspection of carpets”, in Proc. SP IE, Vol. 2345, pp. 180-191, 1995.

[4] P.M. Mahajan, S.R. Kolhe and P.M. Patil “A Review of Automatic Fabric Defect Detection Techniques” Advances in Computational Research, ISSN: 0975–3273, Vol.1, Issue 2, pp.18-29.2009.

[5] S. Priya, T. Ashok Kumar and Paul Varghese, “A Novel Approach to Fabric Defect 

 

Detection Using Digital Image Processing”, Proceedings of International Conference on SignalProcessing,Communication, Computing and Networking Technologies (ICSCCN 2011),2011.

[6] X. F. Zhang and R. R. Bresee, “Fabric defect detection and classification using image analysis”, Textile Res. J., Vol. 65, No.1, pp.1-9, 1995.

[7] E. J. Wood, “Applying Fourier and associated transforms to pattern characterization in textiles,” Textile Res. J., Vol. 60, pp. 212-220, 1990.

[8] C. Chan and G. K. H. Pang, “Fabric defect detection by Fourier analysis”, IEEE Trans. on Ind. Appl, Vol.36, No.5, pp.1267-1276, Oct 2000.

[9] T. Ashok kumar, S.O. Priya and M.G. Mini, “Optic disc localizationin ocular fundus images,” Proc. of iCVCi International Conference, India 2011.

[10] T.J. Kang et al., “Automatic Structure Analysis and Objective Evaluation of Woven Fabric Using Image Analysis”, Textile Res. J.Vol.71, No.3, pp.261-270, 2001.

 

All Rights Reserved ©2017 International Journal of Computer Science (IJCS Journal) &

Department of Computer Science, Sri Vasavi College, Erode, Self-Finance Wing, Erode, Tamil Nadu, INDIA

Published by SK Research Group of Companies (SKRGC) - Scholarly Peer Reviewed Research Journals

http://www.skrgcpublication.org/ Page 1126

 

Sri Vasavi College, Erode Self-Finance Wing 3rd February 2017

National Conference on Computer and Communication NCCC’17

http://www.srivasavi.ac.in/ nccc2017@gmail.com

 

[11] Jagrti Patel, Meghna Jain and Papiya Dutta” Detection of Faults Using Digital Image Processing Technique” 1M.Tech Scholar,

2Assistant Professor, 3Assoc. Professor,

Department of Electronics & Communication, Gyan Ganga College of Technology, Jabalpur - 482 003,Madya Pradesh, India

[12] Prof.P.Y. Kumbhar, Tejaswini Mathpati, Rohini Kamaraddiand Namrata Kshirsagar”Textile Fabric Defects Detection and Sorting Using Image Processing” international journal for 

 

research in emerging science and technology, volum-3, issue-3, mar-2016 Asian Journal of Engineering and Applied Technology ISSN 2249-068X Vol. 2 No. 1, 2013, pp.36-39© The Research Publication, www.trp.org.in

 

All Rights Reserved ©2017 International Journal of Computer Science (IJCS Journal) &

Department of Computer Science, Sri Vasavi College, Erode, Self-Finance Wing, Erode, Tamil Nadu, INDIA

Published by SK Research Group of Companies (SKRGC) - Scholarly Peer Reviewed Research Journals

http://www.skrgcpublication.org/ Page 1127

 

Gradient Pile Up Algorithm for Edge Enhancement and

Detection

Leticia Guimarães1, André Soares1,2, Viviane Cordeiro1, and Altamiro Susin1,2

1 Departamento de Engenharia Elétrica, Universidade Federal do Rio Grande do Sul - UFRGS,

Av. Osvaldo Aranha, 103, CEP. 90 035-190,Porto Alegre, RS, Brazil

leticia@eletro.ufrgs.br

http://www.lapsi.eletro.ufrgs.br

2 Instituto de Informática, Universidade Federal do Rio Grande do Sul – UFRGS,

Av. Bento Gonçalves, 9500, CEP. 91501-970, Porto Alegre, RS, Brazil

Abstract. Edge detection plays a fundamental role on image processing. The detected edges describe an object contour that greatly improves the pattern recognition process. Many edge detectors have been proposed. Most of them apply smooth filters to minimize the noise and the image derivative or gradient to enhance the edges. However, smooth filters produce ramp edges with the same gradient magnitude as those produced by noise. This work presents an algorithm that enhances the gradient correspondent to ramp edges without amplifying the noisy ones. Moreover, an efficient method for edge detection without set a threshold value is proposed. The experimental results show that the proposed algorithm enhances the gradient of ramp edges, improving the gradient magnitude without shifting the edge location. Further, we are testing the implementation of the proposed algorithm in hardware for real time vision applications.

1 Introduction

The estimation of the boundary location of an object is an important subject on image processing. Since edge detection have been studied, several techniques for edge enhancement were proposed. Most of the edge enhancement methods apply the derivative to enhance the edge of the objects on an image. Usually, the derivative of an image is produced by the convolution with a gradient operators [1], as Sobel, Roberts, Prewitt, for example.

However, there are three problems on the use of the derivative as edge detector: a) false edges produced by noise are enhanced; b) ramp edges produce lower derivative magnitude; c) the location of the ramp edges are not accurate. Several edge detection methods as proposed by Marr-Hildreth [2], Canny [3], Perona-Malik [4], have been proposed in order to provide tools to overcome the edge detection problems by applying the derivative to edge detection. Most of these methods apply a smooth gaussian operator in order to suppress the noisy edges, then, estimate the location of the edges by detection of the zero-crossing on the second derivative of an image. Edge detectors proposed by Petrou-Kitter [5] and Zang et. al. [6] are based on ramp edges model. Petrou-Kitter[5] propose a ramp edge model to an optimal edge

A. Campilho, M. Kamel (Eds.): ICIAR 2004, LNCS 3211, pp. 187–194, 2004.

© Springer-Verlag Berlin Heidelberg 2004

 

188 L. Guimarães et al.

detector. Zang et. al. [6] applies correlation matching to enhance the gradient at ramp edge location. In [7] D. Wang proposed a multiscale gradient algorithm that enhances the morphological gradient of ramp edges, however, the edge of thin objects can be lost.

Our approach aims to enhance the gradient value of blurred edges without enhances noisy gradient. In order to achieve our objective the enhancement occurs over the two component vector of the gradient, the horizontal and vertical gradient map separately. First, a smooth filter is applied to the original image. Second, the horizontal and vertical components of the gradient of an image are calculated. Third, each component of the gradient generates a connected component map, the gradient enhancement map (GEM), that guide the gradient enhancement process. Finally, the horizontal and vertical gradient map are enhanced by a process similar to a pile up, guided by their respective gradient enhancement map (GEM). The experiments presented in this paper demonstrate the improvement of the performance on edge detection using the gradient map enhanced by the proposed method.

2 Gradient Map Calculation

There are two components of the gradient, the horizontal and vertical component. The two-dimensional function gh(x,y) represents the horizontal gradient map and gv(x,y) represents the vertical gradient map of an image f(x,y), respectively. The horizontal gradient map gh(x,y) is calculated by the convolution of the image f(x,y) with a horizontal gradient operator. The vertical gradient map gv(x,y) is calculated by the convolution of the image f(x,y) with a vertical gradient operator. The gradient magnitude map g(x,y) is the magnitude of the vector composed by two components. An example of horizontal and vertical operators are the Prewitt operators.

3 Gradient Enhancement Map Generation

The gradient enhancement map GEM guides the pile up process. The pile up process occurs on two gradient maps separately following the same rules. Therefore, two gradient enhancement maps are generated, GEMH(x,y) that guides the enhancement of the horizontal gradient gh(x,y) and GEMV(x,y) that guides the enhancement of the vertical gradient gv(x,y).

The GEM is based on connected components map proposed by Bieniek and Moga

[8]. The GEM is composed by seven elements: a) zero (Z); b)right arrow ( ); c)

left arrow ( ); d) up arrow (  ); e) down arrow ; f) plateau and (P); g)

maximum (M). In [8] the connected components map is used to locate the minima, in our approach we intend to locate the maxima. Therefore, the arrows on GEM point to the maximum neighbor in magnitude. Since the direction of the arrows are related to the gradient direction, GEMH is composed only by up and down arrow and GEMV is composed by only left and right arrow.

 

4 Gradient Enhancement Process (Pile Up Process)

The gradient piled-up process is guided by the gradient enhancement maps GEM of the gradient maps, the horizontal and vertical map. The GEM indicate the direction, the start and the end points of the gradient pile up process. The arrows indicate the pile up direction. The start point of the pile up process is an arrow that the neighbor is zero (Z) and points to another arrow or maximum. The gradient value of a point at an

 

190 L. Guimarães et al.

 

(a) (b)

Fig. 1. Example of gradient enhancement process for one dimensional signal; (a) gradient enhancement map; (b) Gradient enhancement process (pile up algorithm) for the one dimensional ramp edge of the Fig. 1(a).

Search for other

START Condition

Fig. 2. Schema of the gradient enhancement process (pile up algorithm) for the vertical gradient enhancement .

 

Gradient Pile Up Algorithm for Edge Enhancement and Detection 191

 

 

(a) (b) (c)

Fig. 3. (a) Synthetic test image (60X60); (b) GEMH of the Prewitt gradient of synthetic image in (a); (c) GEMV of the Prewitt gradient of the synthetic image in (a).

arrow is piled-up over the neighbor gradient value that presents an arrow of same direction or a maximum. The process ends at a maximum. Fig. 1, shows the gradient enhancement process (pile up process) for one-dimensional ramp edge. Observe, the arrows on final GEM in Fig.1(a) are substituted by zeros (Z) as the pile up occurs, in Fig. 1(b). At the end of the pile up process, the GEM presents only maxima (M) and zeros (Z) and the gradient is enhanced from one, the original gradient value, to five, the difference between a and b level in Fig. 1(a). Moreover, the edge location is evident at the maximum on GEM. Fig. 2 shows the flowchart for the left and right pile up process to the vertical gradient enhancement.

The GEM for a synthetic image is shown in Fig. 3. Fig. 4 shows the original Prewitt gradient and the enhanced gradient for a line of the gradient map.

The edge points of edge map are those which correspond to the maximum at the GEM after the pile up process. However, only the piled up maximum is considered edge point. We intend to implement in hardware the proposed pile up algorithm, in order to accelerate the edge detection on vision chips applications.

Fig. 4. The pile up process for the line y=30 on synthetic image in Fig. 4. (a) line y=30 of the original image, Prewitt gradient and enhanced Prewitt gradient; (b) line y=30 of the noise image, Prewitt gradient and enhanced gradient.

5 Performance Evaluation

The performance of the gradient enhancement algorithm was proceeded by two methods: one applying the Pratt´s Figure Of Merit [9] FOM and other by mean absolute error MAE between the ideal edges and the detected edges.

 

192 L. Guimarães et al.

 

 

 

(a) (b) (c) (d)

Fig. 5. Test images; (a) original image; (b) image corrupted by uniform gaussian noise with variance σ=0.001; (c) edge detected at the best FOM using the Prewitt (FOM=83.66% and MAE=30.88%); (d) edge detected at the best FOM using the enhanced Prewitt gradient (FOM= 83.84% and MAE=7.99%).

The Figure of merit FOM is given by:

 

FOM 1 Na 1 (5)

 max( Na, Ni ) i1 1di2

 

where Na is the number of detected edge points, Ni is the number of ideal edge points (ground truth),  a scaling factor (we adopt  =0.1), and di distance for the ith

detected edge point.

The edge map was produced by thresholding the gradient magnitude map, however, for the original gradient the edges were thinned by the morphological skeleton operation. The FOM and MAE curve were calculated for each threshold edge map, shown in Fig. 6. The mean absolute error MAE is calculated by eq. (6).

1 M N (6)

  mi(y,x)me(y, x)

y1 x1

where mi is the ideal edge map, me is the resulting edge map, M and N are height and weight of the mi and me.

The MAE performance at the best FOM for the original Prewitt gradient and the enhance Prewitt gradient of image in Fig. 5 corrupted by gaussian noise, are shown in Fig. 7. Fig. 8 and 9 shows the results of edge detection applied to the Lena and Boat images. The ideal edge map of Lena in Fig. 8(b) was constructed manually in order to compare with the other edge detection techniques by calculating the FOM and MAE.

(a) )

(b

Fig. 6. Performance of the original and enhanced gradient for the image in Fig. 6(b); (a) Figure

Of Merit (FOM) x Threshold value ; (b) Mean Absolute Error (MAE) x Threshold value.

 

Gradient Pile Up Algorithm for Edge Enhancement and Detection 193

Fig. 7. Comparison of the Performance of Prewitt vs. Enhanced Prewitt for the image in Fig. 5.

 

 

(d) (e) (f)

Fig. 8. Edge detection on Lena; (a) original image; (b) ideal edge map of Lena; (c) best FOM

edge map from Canny with standard deviation 1 (FOM=75.28% and MAE=16.37% ); (d) smoothed Lena by mean filter 33; (e) best FOM edge map from the Prewitt gradient of Lena (FOM=79.18% and MAE=27.12%) and; (f) edge map from enhanced Prewitt (FOM=77.43% and MAE=14.72%).

 

(a) (b) (c)

Fig. 9. Edge map of the Boat image after pre-processed by a mean filter 33; (a) Original Boat image; (b) Boat image after the pre-processing; (c) the edge map generated by the proposed algorithm.

 

194 L. Guimarães et al.

6 Conclusions

The experimental results for the synthetic image show that the best FOM for the enhanced gradient is near from the best FOM of the original gradient, while for the enhanced gradient the MAE decreased 50% of the original gradient, at the best FOM threshold. Although the best FOM for original Prewitt gradient of Lena in Fig. 8, (FOM=79.18%) had been greater than the best FOM of the proposed enhanced gradient (FOM=77.43%), the MAE of the edge map obtained applying the proposed pile up method had been reduced from a MAE=27.12%, for the original gradient, to MAE=14.72% at the best FOM. Therefore, the proposed gradient enhancement method is effective to enhance gradient edges and suppress noisy gradient.

Furthermore, the FOM for the edge map produced by the proposed method is higher than the edge map generated applying the Canny edge detector, in Fig. 8(c), without setting any parameter as threshold value.

The proposed pile up process preserves the edges even for thin objects in an image, as shown in Fig. 9(c).

Acknowledgements. This work was supported by Conselho Nacional de Pesquisa CNPq – BRAZIL.

References

1. Gonzalez, R., Woods, R., Digital Image Processing, Addison-Wesley, 1992.

2. Marr, D. and Hildreth, E. C., Theory of Edge Detection, Proc. of the Royal Society of London B207, pp. 187-217, 1980.

3. Canny, J., A Computational Approach to Edge Detection, PAMI, V. 8, No. 6, pp. 679-698, 1986.

4. Perona, P. and Malik, J., Scale-Space and Edge Detection Using Anisotropic Diffusion, PAMI ,V.12 , No. 7, 1990.

5. Petrou, M. and Kitter, J., Optimal Edge Detectors for Ramp Edges, PAMI, V. 13, No. 5, pp. 483-491, 1991.

6. Wang, Z., Rao, K. R. and Ben-Arie, J., Optimal Ramp Edge Detection Using Expansion Matching, PAMI, V. 18, No. 11, pp. 1092-1097, 1996.

7. Wang, D., A Multiscale Gradient Algorithm for Image Segmentation using Watersheds, Pattern Recognition, V. 30, No. 12, pp. 2043-2052, 1997.

8. Bieniek, A and Moga, A., An efficient watershed algorithm based on connected components. Pattern Recognition, V. 33 , No. 6 , pp. 907-916, 2000.

9. Ballard, D.H., Brown, C.M., Computer Vision, Prentice Hall Inc., 1982.

 

Using Recurrent Neural Networks To Forecasting of Forex

V.V.Kondratenko1 and Yu. A Kuperin2

1 Division of Computational Physics, Department of Physics, St.Petersburg State University

2 Laboratory of Complex Systems Theory, Department of Physics, St.Petersburg State University E-mail: kuperin@JK1454.spb.edu

Abstract

This paper reports empirical evidence that a neural networks model is applicable to the statistically reliable prediction of foreign exchange rates. Time series data and technical indicators such as moving average, are fed to neural nets to capture the underlying “rules” of the movement in currency exchange rates. The trained recurrent neural networks forecast the exchange rates between American Dollar and four other major currencies, Japanese Yen, Swiss Frank, British Pound and EURO. Various statistical estimates of forecast quality have been carried out. Obtained results show, that neural networks are able to give forecast with coefficient of multiple determination not worse then 0.65. Linear and nonlinear statistical data preprocessing, such as Kolmogorov-Smirnov test and Hurst exponents for each currency were calculated and analyzed.

Key words: Neural Networks, Foreign Exchange Rate, Statistical Tests, Hurst Exponent, Complex Systems Theory

1. Introduction

Forex is the largest and most liquid of the financial markets, with an approximately $1 trillion traded every day. It leads to the serious interest to this sector of finance and makes clear that for various reasons any trader on Forex wish to have an accurate forecast of exchange rate. Most of traders use in old fashion manner such traditional method of forecast as technical analysis with the combination of fundamental one. In this paper we develop neural network approach to analysis and forecasting of financial time series based not only on neural networks technology but also on a paradigm of complex systems theory and its applicability to analysis of various financial markets (Mantegna et al., 2000; Peters, 1996) and, in particularly, to Forex. While choosing the architecture of neural network and strategy of forecasting we carried out data preprocessing on the basis of some methods of ordinary statistical analysis and complex systems theory: R/S-analysis, methods of nonlinear and chaotic dynamics (Mantegna et al., 2000; Peters, 1996). In the present paper we do not describe all of them. We present here only the results of the Kolmogorov-Smirnov test and results of R/S-analysis. However we stress that the preliminary analysis has allowed to optimize parameters of neural network, to determine horizon of predictability and to carry out comparison of forecasting quality of different currencies.

Below we give some remarks relating to advantages of neural networks technology over traditional methods and compare our approach with the methods of other authors

 

First, neural networks analysis does not presume any limitations on type of input information as technical analysis does. It could be as indicators of time series, as information about behavior of another financial instruments. It is not without foundation, that neural networks are used exactly by institutional investors (pension funds for example), that deal with big portfolios and for whom correlations between different markets are essential.

Second, in contrast to technical analysis, which is based on common recommendations, neural networks are capable to find optimal, for given financial instrument, indicators and build optimal, for given time series, forecasting strategy.

Let us remind that in present study we forecasted the exchange rates of only selected currencies on Forex market. As currencies to deal with, we chose British Pound, Swiss Frank, EURO and Japanese Yen. The following motivates this choice: practically all the main volumes of operations on Forex are made with this currencies.

Let us note that there were published a lot of papers, where similar problems have been studied. (Jingtao Yao et al., 2000; Castiglione, 2001; Kuperin et al., 2001; Lee et al., 1997; Tino et al., 2001; McCluskey, 1993). Let us briefly look round the results some of them.

In (Castiglione, 2001) there were studied the problem of sign of price increments forecasting. As an analyzed data were taken such indices as S&P500, Nasdaq100 and Dow Jones. Has been taken multilayer perceptrons of different configurations, with different number of hidden neurons. As a result there has been shown a possibility of forecasting of the sign of price increments with probability of slightly higher than 50%, i.e. a little bit better then just coin tossing. We suppose, that such kind of results is irrelevant from the practical point of view and has an academic interest.

In (McCluskey, 1993) has been studied the problem of forecasting the price increments of S&P500. As the analyzed data has been taken historical S&P500 data for the period from 1928 to 1993. There have been built and trained a lot of neural networks of different configuration. There has been estimated the profitability of using the neural networks in question. In the result surprisingly almost all networks were able to give a profitable forecast. Any statistical estimates of forecast quality were absent. It is clear that statistical estimates of forecast quality are extremely important, since profitable forecast might be just accidental.

In (Tino et al., 2001) there has been forecasted the sign of price increments. As the analyzed data there have been taken DAX and FTSE100 for some time period. The Elman recurrent neural network has been chosen. To the input there were fed binary signals corresponding to the sign of price increments. As an estimate of forecast quality, the profitability was chosen as in above paper. In the result the authors made a conclusion, that neural networks are not capable to give better results than more simple models, such as Markov models for example.

In (Jingtao Yao et al., 2000) there were carried out a neural network forecast of currencies exchange rates on Forex. As the analyzed data there were chosen weekly data of AUD, CHF, DEM, GBP and JPY. Two neural models in this study have been used:

1. Multilayer perceptron, to inputs of which there were fed values of exchange rates with some time lag.

2. Multilayer perceptron, to inputs of which there were fed values of moving average of exchange rates with different time windows (from 5 to 120 ticks).

In the result there were performed statistical estimates of forecast quality, which showed for the last model, that neural network is able to give a good result. At the same time we have to emphasize, that forecast of the exchange rate by itself is of no practical value. Besides of that, the forecast of weekly data presumes, that trader, who uses this forecast will trade once a week, which is irrelevant from the practical point of view.

We have also to mention, that forecasting of future values of underlying time series gives, besides the direct profit, the opportunity of calculating of some interesting quantities, such as price of derivative or probability of adverse mode, which is essential when one estimate the risk of investment. (Castiglione, 2001).

 

In the reminder of the paper we describe the structure of financial data we used, the data preprocessing that we carried out, i.e. normalization, calculation of Hurst exponent, Kolmogorov-Smirnov test. Next we describe the neural network architecture we used and statistical estimates of the results. Finally we present the results of time series forecasting and discussion. Last section concludes.

2. Time series forecasting

2.1 Structure of Data

In this section we remind the well-known features of economic and financial time series, which should be taken into account in training of neural networks and appropriate forecasting. What is follows are the major features:

Noise: Macroeconomic time series are intrinsically very noisy and generally have poor signal to noise ratios. The noise is due both to the many unobserved variables in the economy and to the survey techniques used to collect data for those variables that are measured. The noise distributions are typically heavy tailed and include outliers. The combination of short data series and significant noise levels makes controlling model variance, model complexity, and the bias / variance tradeoff important issues (Geman, Bienenstock and Doursat, 1992). One measure of complexity for nonlinear models is Peff , the effective number of parameters (Moody, 1992; Moody, 1994). Peff can be controlled to balance bias and variance by using regularization and model selection techniques.

Nonstationarity: Due to the evolution of the world economies over time, macroeconomic time series are intrinsically nonstationary. To confound matters, the definitions of many macroeconomic time series are changed periodically as are the techniques employed in measuring them. Moreover, estimates of key time series are periodically revised retroactively as better data are collected or definitions are changed. Not only do the underlying dynamics of the economy change with time, but the noise distributions for the measured series vary with time also. In many cases, such nonstationarity shortens the useable length of the data series, since training on older data will induce biases in predictions. The combination of noise and nonstationarity gives rise to a noise / nonstationarity tradeoff (Moody, 1994), where using a short training window results in too much model variance or estimation error due to noise in limited training data, while using a long training window results in too much model bias or approximation error due to nonstationarity.

Nonlinearity: Traditional macroeconomic time series models are linear (Granger and Newbold, 1986; Hamilton, 1994). However, recent works by some authors have suggested that nonlinearities can improve macroeconomic forecasting models in some cases (Granger and Terasvirta, 1993; Moody et al., 1993; Natter, Haefke, Soni and Otruba, 1994; Swanson and White, 1995). (See table 1 and figures 2 and 3.) Based upon our own experience, the degree of nonlinearity captured by neural network models of macroeconomic time series tends to be mild (Moody et al., 1993; Levin, Leen and Moody, 1994; Rehfuss, 1994; Utans, Moody, Rehfuss and Siegelmann, 1995; Moody, Rehfuss and Saffell, 1996; Wu and Moody, 1996). Due to the high noise levels and limited data, simpler models are favored. This makes reliable estimation of nonlinearities more difficult.

2.2 Data preprocessing

As it was mentioned above, in this paper there have been analyzed exchange rates of four main currencies on Forex: British pound, Swiss frank, EURO and Japanese Yen in the period from 22.07.1998 to 02.09.2001. Data was taken from www.gelium.net and www.forexite.com.

 

There are diagrams of exchange rates of currencies in question below for entire period and for shorter period. It could be seen that in spite of apparent smoothness graphs are rather heavily jagged.

Figure 1a. Daily exchange rate of CHF for entire period

22/07/1998 – 02/09/2001. Along the coordinate axes are given logarithmic returns

(ordinate axis) and time (abscissa axis), the definition see below

Figure 1b. Daily exchange rate of CHF for shorter period 01/01/2000– 01/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis), the definition see below

 

Figure 2a. Daily exchange rate of EUR for entire period 22/07/1998 – 02/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis), the definition see below.

Figure 2b. Daily exchange rate of EUR for shorter period 15/06/2000– 15/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis), the definition see below.

Figure 3a. Daily exchange rate of GBP for entire period 22/07/1998 – 02/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis),

the definition see below.

 

GBP

1.55

1.5

1.45

1.4

1.35

1.3

09.06.2000 09.07.2000 09.08.2000 09.09.2000 09.10.2000 09.11.2000 09.12.2000 09.01.2001 09.02.2001 09.03.2001 09.04.2001 09.05.2001 09.06.2001 09.07.2001 09.08.2001

Figure 3b. Daily exchange rate of GBP for shorter period 09/06/2000– 09/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis), the definition see below.

JPY

 

 

Figure 4a. Daily exchange rate of JPY for entire period 22/07/1998 – 02/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis),

the definition see below.

Figure 4b. Daily exchange rate of JPY for shorter period 08/10/2000–08/09/2001. Along the coordinate axes are given logarithmic returns (ordinate axis) and time (abscissa axis),

the definition see below.

Financial data has particular property: the higher is the frequency of data collection the worse is signal/noise ratio. Regard to this, weekly data are usually better forecasted, than daily

 

data. On the other hand the more accurate is the forecast the higher is its practical value. In compliance with these principles we chose daily closing prices as input data.

Usually traders are not really interested in forecast of exchange rate itself. Much more of interest is the increment of rate or the sign of increment forecast. Moreover, there is strong correlation between current and previous values of price – most probable value of price in next moment equals to its previous value. Meanwhile, in order to improve the quality of forecast it is necessary to make inputs statistically independent, i.e. to remove correlations.

Because of this in our study we analyzed not the exchange rates but its logarithmic returns:

S

r (1)

ln( )

n 

n Sn1

It is clear, that results of neural network analysis should not depend on the units of measurement. It means that all the input and output values should be reduced to a unit scale, in other words to be normalize. Moreover, in order to improve the learning rate it is effective to carry out additional data preprocessing that smoothes the data distribution before learning phase.

In the present study normalization of data was performed with the use of logistic function – the activation function of neurons. Transformation was performed corresponding to following formula:

x~

i

1  exp(

6

where: x is the mean, 6 is the standard deviation.

This transformation normalizes main volume of data and guarantees that  0,1

x~ i E .

Moreover it increases the entropy of data, since after transformation data fills the interval [0,1] more even.

The most well known but little believed hypotheses are Random Walk Hypothesis and Efficient Market Hypothesis. The Random Walk Hypothesis states that the market prices wander in a purely random and unpredictable way. The Efficient Market Hypothesis roughly speaking states that markets fully reflect all of the available information and prices are adjusted fully and immediately once new information become available. In the actual market, some people do react immediately after they have received the information while other people wait for the confirmation of information. The waiting people do not react until a trend is clearly established. All together (in week form or in stricter form) it means that in frames of Efficient Market Hypothesis is nothing to do with any type of forecasting and, in particularly, by means of neural networks. However, our study is based on the recent empirical evidence that in actual markets, the residual inefficiencies (caused by different reasons) are always present. In general we are not going to discuss these reasons here, however we shall use some of tools from linear and nonlinear statistics in order to analyze the deviation of data in question from efficiency. That is the standard Kolmogorov-Smirnov test and R/S analysis (or Hurst exponents).

The Hurst Exponent is in particular a measure of the bias in the fractional Brownian motion (Peters 1996, Peters 1994, Kantz 1997). In general (modulo any probability distribution) the method could be used in economic and financial market time series to know whether these series are biased random walk, which indicates in some sense the possibility of forecasting.

The Hurst Exponent H describes the probability that two consecutive events are likely to occur (for more details see, e.g. Peters 1996). There are three distinct classifications for the Hurst exponent: 1) H=0.5, 2) 0 <_ H < 0.5, and 3) 0.5< H <1.0. The type of series described by H=0.5 is a pure random, consisting of uncorrelated events. However, value H=0.5 cannot be used as an evidence of a Gaussian random walk. It only proves that there is no evidence of a long memory effect. A value of H different from 0.5 denotes that the observations are correlated. When 0 <_ H

 

<_ 0.5, the system is characterized by antipersistent or ergodic time series with frequent reversals and high volatility. At the same time despite the prevalence of the mean reversal concept in economic and finance literature, only few antipersistent series have yet been found. For third case (0.5 <_ H <_ 1), H describes a persistent or trend reinforcing series, which is characterized by long memory effects. The strength of the bias depends on how far H is above 0.5. However, even in the case that the Hurst process describes a biased random walk, the bias can change abruptly either in direction or magnitude. Therefore, only the average cycle length of observed data can be estimated.

It is useful before neural network forecasting to check if our time series are persistent or antipersisnet and whether returns have Gaussian distribution. This can give some information on the predictability of time series in question. If time series have H0.5 and returns have Gaussian distribution then we can presume that these time series hardly could be accurately forecasted. On the other hand, the closer H is to 1 and the more the distribution of returns differs form Gaussian distribution, the higher is the probability that these time series could be forecasted with high quality.

So to check the predictability of the analyzed time series there has been performed statistical analysis of data. For each of the currencies there were computed first four moments, i.e. mean, variance, skewness and kurtosis. Also there was performed the Kolmogorov-Smirnov test and have been computed Hurst exponents. Results are given below.

Table 1.1. Statistical data for currencies.

Mean Maximum Minimum Variance St. Deviation Skewness Kurtosis

CHF -8,3E-05 -0,03798 0,035216 4,92E-05 0,007017 0,2380862,291049

EUR 0,000246 -0,02335 0,023201 3,2E-05 0,005661 -0,18691 2,138603

GBP -1,7E-05 -0,02245 0,024826 2,42E-05 0,004917 0,009823 1,655181

JPY 4,13E-05 -0,07811 0,035251 6,01E-05 0,007755 -0,94684 7,928037

Table 1.2 The Kolmogorov-Smirnov test results.

CHF 0.05871

EUR 0.010531

GBP 0.04967

JPY 0.06328

Table 1.3 The values of Hurst Exponents

CHF 0.566  0.002

EUR 0.570  0.002

GBP 0.523  0.004

JPY 0.557  0.002

As it is shown in Tables 1.1-1.3, the values of Hurst exponents for the logarithmic returns of daily exchange rates data is higher than 0.5 for all the observed time series. This shows that all the studied currencies are persistent. The highest value is 0.570 for the exchange rate of EUR/USD. In other words, it is the least efficient currency amongst the four studied ones. The results of the Kolmogorov-Smirnov test show that not any returns of observed time series have Gaussian distribution. This means that not any time series are described by Brownian motion. The closest distribution to Gaussian is the distribution of EUR logarithmic returns. It means, that EUR returns are weaker correlated than the others currencies time series returns and probably EUR will have the worst forecast.

 

The results obtained in this part of the study imply that the all time series are nor antipersistent neither Brownian motion. Thus neural network forecasting of the exchange rates is possible and should has a good quality.

 

Figure 5. Distribution of CHF. Here abscissa axis is the upper boundaries of fragmentation of

the logarithmic returns and ordinate axis is the number of observations of return’s value in each

fragmentation interval.

 

 

Figure 6. Distribution of EUR. Here abscissa axis is the upper boundaries of fragmentation of the

logarithmic returns and ordinate axis is the number of observations of return’s value in each

fragmentation interval.

 

Figure 7.Distribution of GBP. Here abscissa axis is the upper boundaries of fragmentation of the

logarithmic returns and ordinate axis is the number of observations of return’s value in each

fragmentation interval.

 

 

Figure 8.Distribution of JPY. Here abscissa axis is the upper boundaries of fragmentation of the

logarithmic returns and ordinate axis is the number of observations of return’s value in each

fragmentation interval.

1.3 Choosing the of neural network architecture

The presence of recurrent feedback in neural network seems to be a positive factor in forecasting of financial time series. (Castiglione, 2001; McCluskey, 1993) This is apparently because the recurrent neural network has a “deeper memory” than feedforward neural network.

In our study the Elman-Jordan neural network has been used. This class of neural networks with learning by back propagation method was successfully used for prediction of financial markets since it is the recurrent neural net which learns the rules in the time series, which is necessary when works with ones. The disadvantage of this class of neural nets is the long learning time. (Kuperin et al., 2001).

The network we built had two inputs and one output. To one input there were fed daily returns, to another one there were fed daily returns, smoothed by moving average with the window equal to 5 observations. Moving average in input was supposed to smooth the noisy time series of returns. As the output we chose the value of moving average of returns with the window equal to 5 observations, shifted to one day forward, because it corresponds to trading purposes better than returns because of more accurate forecast and such fact that if trader would try to trade on every market movement it would lead to very expensive transaction costs., It is partially explain why it is reasonable to forecast the moving average. Thus, the network was supposed to forecast the one-day ahead value of moving average of returns with the window equal to 5 observations. The number of hidden neurons was chosen equal to 100. .

Neurons of input layer had linear activation functions. Neurons of hidden and output layers had logistic activation function. This configuration was determined experimentally as giving the best results.

 

Momentum and learning rate were chosen equal to 0.003. The net terminated its learning if during 20000 epochs it was not capable to improve its results. Training, test and production sets were chosen as following: first 1000 observations were used as the training set, next 200 observations were used as the test set, and last 100 observations were used as the production set.

 

Figure 9. Schematic illustration of the Elman-Jordan network architecture. Here r(t) is

logarithmic return, MA(r,5) is the moving average with window equal to 5 of logarithmic returns

and Loop Back denotes the interlayer feedback.

2.4 Statistical estimation of forecasting results

In the present section we briefly describe some indices we used for statistical estimations of the neural network forecasting quality. This part of the study is quite important both for the proper choice of neural network architecture and for statistical relevance of obtained results. The specific values of obtained indices are discussed below.

In particular, as performance criteria there were chosen the following indices:

R2 The coefficient of multiple determination is a statistical indicator usually applied to multiple regression analysis. It compares the accuracy of the model to the accuracy of a trivial benchmark model wherein the prediction is just the mean of all of the samples. A perfect fit would result in an R squared value of 1, a very good fit near 1, and a very poor fit less than 0. If the neural model predictions are worse than one could predict by just using the mean of the sample case outputs, the R squared value will be less than 0. The formula used for R2 is given by

 ( y  y ˆ)

i 2

R 21

2

i

where: yi is the actual value, yˆ is the predicted values of yi , y is the mean of the yi values

r squared This is the square of the correlation coefficient, described below.

Mean Squared Error This is the mean over all patterns of the square of the actual value

minus the predicted value, i.e., the mean of (actual minus predicted) squared.

 

Min Absolute Error This is the minimum of actual minus predicted values over all patterns.

Max Absolute Error This is the maximum of actual minus predicted values over all patterns.

Correlation Coefficient r (Pearson’s Linear Correlation Coefficient) This is a statistical measure of the strength of the correlation between the actual and predicted outputs. The r coefficient can range from -1 to +1.

Percent within 5%, 10%, 20% and 30% and over 30% It is the percent of network answers that are within the specified percentage of the actual answers used to train the network. If the actual answer is 0, the percent cannot be computed and that pattern is not included in a percentage group. For that reason and rounding, the total computed percentages may not add up to 100%.

Ability of correct prediction for the sign of increments This is the number of correctly predicted signs of increments, divided by the whole number of patterns.

Most representative statistical estimate of forecast quality among of all mentioned above is R2. Other representative indices are: r squared, mean absolute error and maximum absolute error. Mean absolute error and maximum absolute error show the accuracy of forecast, r squared shows the linear correlation, but tells nothing about accuracy. R2 gives information both on accuracy and correlation. It can be shown that

 (1 2 ) 

R (4)

where:  is the mean absolute error,  is the standard deviation.

For example, if we have R2 = 0.8, so we can estimate mean absolute error as  0.447 .

3. Results and discussion

In the present study there were built and trained four neural networks, one for each currency. In Appendix we present the Tables with the statistical results of forecast quality.

Below we give also the examples of graphs, which visualize the quality of forecast, where ordinate axis is the value of logarithmic returns and abscissa axis is the time.

Figure 10. Actual and predicted values of CHF for the period from 18/04/2001 to

01/10/2001.

 

Figure 11. Actual and predicted values of EUR for the period from 12/08/2001 to

04/05/2002.

Figure 12. Actual and predicted values of GBP for the period from 18/04/2001 to

01/10/2001.

Figure 13. Actual and predicted values of JPY for the period from 23/04/2001 to

01/10/2001.

 

We also made some attempts to improve the achieved results in the following direction. Since the amount of data in use was limited by available data on EURO (exchange rate of EURO appeared in 1998), we decided to take more data in spite of its different amount for different currencies. Thus, the amount of data in use was chosen equal to 2200 daily ticks. Moreover, we decided to change the configuration of inputs and to add some technical indicators to the input of the neural net. We have chosen the Relative Strength Index (RSI) and Williams Percent Range (%R) as such inputs. In other words there were built several neural networks with different configuration of inputs. Below we give more detailed description of these models. (See Appendix for the appropriate results and statistical estimation). Namely, we undertaken the following numerical experiment:

1. We have added to the input of neural network such technical indicators as RSI and %R. We expected to improve the quality of forecast due to presence of leading technical indicators in the network input. Since the number of inputs increased, we also increased the number of hidden neurons up to 140.

2. As the next step we have decided to feed only technical indicators in input of the network. Moving average of returns and returns themselves were removed. We supposed to check the prevalent belief that all the necessary information contains in RSI and % R time series and so the excess information could only worsen the quality of forecast.

3. Since it turned out that the results, shown by previous model were essentially worse than the others, we decided to feed to the network input the technical indicators and moving average of returns, removed at the same time returns themselves from the input. We supposed that presence of raw data in the input could worsen the quality of forecast.

4. Since the best results have been shown by the network, which had two indicators, i.e., moving average and returns in input, we have decided to take this model as the basic one and only slightly modify it. We replaced moving average by exponential moving average, since it approximates time series more accurate and contains information on longer time period.

5. Finally, we have used the embedding method. Namely, we fed four series values of returns and supposed to obtain the next day returns from the output.

4.Conclusions

Let us summarize the obtained results as follows.

The statistically best results have been obtained by the model, where logarithmic returns and moving average were taken as inputs. Comparing the results, obtained from this model, to others ones it should be noted that:

1. Model with two added technical indicators: in general, the forecast quality became worse than the initial model (R2 is increased on CHF only, on the others currencies it decreased). It turned out that the probability of correctly predicted increments sign is also lower. Apparently, the information that contains in indicators is excess and makes the problem of forecast more complicated. We could increase, of course, the number of hidden neurons, but it would lead to unwarranted increasing of the learning time.

2. Model with only RSI and %R in input: statistically the forecast quality fell down even more then in other models we studied. This indicates, that the mentioned

 

above prevalent belief about information content of technical indicators is irrelevant to the neural network forecasting.

3. Model with RSI, %R and moving average in input: comparing to previous models the forecast quality has been improved, but it still remained not the best. It looks like the presence of moving average in input of the network is the positive factor at least due to the smoothing of the data.

4. Model with indicators, returns and exponential moving average: results, obtained from this model are analogous to results obtained by basic model. Apparently, there is we have found no essential difference between ordinary moving average and exponential moving average for the problem in question.

5. Embedding method: the particular property of this model is that the results of forecasting of different currencies are much more analogous, than in other models. Nevertheless the forecast quality is worse than in basic model. Apparently, it is possible to obtain better results from this model by increasing the number of inputs and hidden neurons, but it would lead again to unjustified long learning time.

It should be noted, that CHF and JPY are usually better predicted than GBP and EURO. Is seemingly means that regularities, that contain CHF and JPY time series are more complicated than regularities, containing in GBP and EURO time series. In EURO case comparatively lower quality of forecast could be explained by insufficient amount of data.

Thus, in this study neural net forecast of four main currencies on Forex market was carried out. Finally let us conclude that statistical estimates of Forex forecast indicates, that neural network can predict the increments sign with relatively high probability – approximately 80%, which is quite sufficient for practical use. The coefficient of multiple determination R2 in the best our model varied from 0.65 to 0.69. . We believe that such quality of forecast is enough for building an automatic profitable trade strategy. This point of view has been confirmed in (Kuperin et al., 2001) where a profitable trade strategy for GBP, based on forecast with similar quality has been constructed and tested.

 

References

1. R.N.Mantegna, H.E.Stenley: An Introduction to Econophysics. Correlations and Complexity in Finance. Cambridge University Press (2000)

2. E.E.Peters: Chaos and Order in Capital Markets. John Wiley&Sons (1996)

3. E.E.Peters: Fractal Market Analysis. John Wiley&Sons (1994)

4. H.Kanz, T.Schreiber: Nonlinear Time Series Analysis. Cambridge University Press (1997)

5. A.A.Ezhov, S.A. Shumsky. Neurocomputing and its Applications in Economics and Business. Moscow.: MEPHI, 1998.

6. Jingtao Yao, Chew Lim Tan. A Case Study on Using Neural Networks to Perform Technical Forecasting of Forex. Neurocomputing, Vol. 34, No. 1-4, pp79-98 (2000).

7. Filippo Castiglione. Forecasting Price Increments Using an Artificial Neural Network. in Complex Dynamics in Economics, a Special Issue of Advances in Complex Systems, 3(1), 45-56, Hermes-Oxford (2001)

8. Kuperin, Yuri A., Dmitrieva Ludmila, A., Soroka Irina V. Neural Networks in financial markets dynamical studies: Working paper series of the Center for Management and Institutional Studies. School of Management. Saint-Petersburg State University. No 2001-12

9. C.Lee Giles, Steve Lawrence, Ah Chung Tsoi. Noisy Time Series Prediction using a Recurrent Neural Network and Grammar Inference. Proceedings of IEEEE/IAFE Conference on computational intelligence for financial engineering IEEE, Piscantaway. New York, (1997). P. 253- 259.

10. John Moody. Economic Forecasting: Challenges and Neural Network Solutions. Proceedings of the International Symposium on Artificial Neural Networks. Hsinchu, Taiwan, (1995)

11. Peter C. McCluskey. Feedforward and Recurrent Neural Networks and Genetic Programs for Stock Market and Time Series Forecasting. Master's thesis, Brown University, (1993).

12. Peter Tino, Christian Schittenkopf, Georg Dorffner. Financial Volatility Trading using Recurrent Neural Networks. Pattern Analysis and Application, (2001)

13. Moody, J. The effective number of parameters: an analysis of generalization and regularization in nonlinear learning systems, in J. E. (1992)

14. Moody, J., Challenges of Economic Forecasting: Noise, Nonstationarity, and Nonlinearity, Invited talk presented at Machines that Learn, Snowbird Utah, April (1994a).

15. Moody, J., Prediction risk and neural network architecture selection, in V. Cherkassky, J. Friedman and H. Wechsler, eds, `From Statistics to Neural Networks: Theory and Pattern Recognition Applications', Springer-Verlag. (1994b)

16. Moody, J., Levin, A. and Rehfuss, S., `Predicting the U.S. index of industrial production', Neural Network World 3(6), 791--794. Special Issue: Proceedings of Parallel Applications in Statistics and Economics '93. (1993)

17. Natter, M., Haefke, C., Soni, T. and Otruba, H., Macroeconomic forecasting using neural networks, in `Neural Networks in the Capital Markets 1994'. Pi, H. and Peterson, C. (1994), `Finding the embedding dimension and variable dependencies in time series', Neural Computation pp. 509--520. (1994)

18. Swanson, N. and White, H., A model selection approach to real-time macroeconomic forecasting using linear models and artificial neural networks, Discussion paper, Department of Economics, Pennsylvania State University. (1995)

19. Rehfuss, S., Macroeconomic forecasting with neural networks. Unpublished simulations. (1994)

20. Utans, J., Moody, J., Rehfuss, S. and Siegelmann, H., Selecting input variables via sensitivity analysis: Application to predicting the U.S. business cycle, in `Proceedings of Computational Intelligence in Financial Engineering'. (1995)

21. Wu, L. and Moody, J., `A smoothing regularizer for feedforward and recurrent networks', Neural Computation 8(2). (1996)

 

Appendix

Table 2. Statistics of forecast quality for main currencies of Forex (the basic model)

CHF EUR GBP JPY

Number of epochs 2622 27370 14389 10181

Number of patterns in training set: 1200 1200 1200 1200

Number of patterns in production set: 103 103 103 103

R squared: 0,6529 0,6694 0,6714 0.6901

r squared: 0,6550 0,6697 0,6722 0.6917

Mean squared error: 0,003 0,001 0,003 0.002

Mean absolute error: 0,041 0,029 0,041 0.039

Minimum absolute

error: 0 0 0,001 0

Maximum absolute

error: 0,178 0,122 0,146 0.150

Pearson’s correlation coefficient: 0,8093 0,8183 0,8199 0.8317

Percent within 5%: 37,879 52,427 35,960 39.798

Percent within 5% to 10%: 28,283 24,272 30,505 30.101

Percent within

10% to 20%: 26,010 23,301 26,263 24.848

Percent within 20% to 30%: 5,556 0 5,657 3.636

Percent over 30%: 2,273 0 1,616 1.616

Probability of correct

sign of increments

forecasting: 0,82 0,82 0,77 0.83

 

Table 3. Statistics of forecast quality for main currencies of Forex for the network with added technical indicators

CHF EUR GBP JPY

Number of epochs 3921 149 2404 12581

Number of patterns in training set: 1600 1200 1600 1600

Number of patterns in production set: 100 100 100 100

R squared: 0.6836 0.4530 0.5436 0.6168

r squared: 0.6941 0.4584 0.5503 0.6212

Mean squared error: 0.002 0.002 0.003 0.002

Mean absolute error: 0.037 0.038 0.040 0.034

Minimum absolute

error: 0 0.001 0.001 0

Maximum absolute

error: 0.155 0.110 0.139 0.137

Pearson’s correlation

coefficient: 0.8332 0.6770 0.7418 0.7882

Percent within 5%: 45.299 50.000 35,960 39.655

Percent within 5% to 10%: 22.222 20.192 29.310 34.483

Percent within

10% to 20%: 26.496 21.154 25.000 22.414

Percent within 20% to 30%: 3.419 7.692 6.897 2.586

Percent over 30%: 2.564 0.962 0.862 0.862

Probability of correct

sign of increments

forecasting: 0,78 0,78 0,73 0.81

 

Table 4. Statistics of forecast quality for main currencies of Forex for the network with two added indicators and without moving average and returns.

CHF EUR GBP JPY

Number of epochs 1918 114 6271 136

Number of patterns in training set: 1600 1200 1600 1600

Number of patterns in production set: 100 100 100 100

R squared: 0.5153 0.4032 0.4114 0.4693

r squared: 0.5332 0.4061 0.4139 0.4798

Mean squared error: 0.003 0.003 0.003 0.003

Mean absolute error: 0.045 0.039 0.047 0.041

Minimum absolute

error: 0 0 0 0

Maximum absolute

error: 0.168 0.130 0.142 0.142

Pearson’s correlation

coefficient: 0.7302 0.6373 0.6433 0.6927

Percent within 5%: 37.607 42.308 29.310 36.207

Percent within 5% to 10%: 23.077 24.038 33.621 31.897

Percent within

10% to 20%: 26.496 25.000 28.448 24.138

Percent within 20% to 30%: 7.692 7.692 5.172 7.759

Percent over 30%: 5.128 0.962 3.448 0

Probability of correct

sign of increments

forecasting: 0,74 0,75 0,73 0,80

 

Table 5. Statistics of forecast quality for main currencies of Forex for the network

without raw data in input

.

CHF EUR GBP JPY

Number of epochs 3794 149 1885 1997

Number of patterns in training set: 1600 1200 1600 1600

Number of patterns in production set: 100 100 100 100

R squared: 0.6513 0.4541 0.5124 0.6022

r squared: 0.6614 0.4601 0.5177 0.6051

Mean squared error: 0.002 0.002 0.003 0.002

Mean absolute error: 0.039 0.038 0.042 0.036

Minimum absolute

error: 0 0 0 0

Maximum absolute

error: 0.166 0.108 0.140 0.126

Pearson’s correlation

coefficient: 0.8132 0.6783 0.7195 0.7779

Percent within 5%: 39.316 48.077 37.931 41.379

Percent within 5% to 10%: 27.350 21.154 27.586 31.897

Percent within

10% to 20%: 25.641 23.077 26.724 24.138

Percent within 20% to 30%: 5.128 6.731 6.897 2.586

Percent over 30%: 2.564 0.962 0.862 0

Probability of correct

sign of increments

forecasting: 0,78 0,76 0,71 0,81

 

Table 6. Statistics of forecast quality for main currencies of Forex for the network with exponential moving average in input.

CHF EUR GBP JPY

Number of epochs 4143 132 3518 117

Number of patterns in training set: 1600 1200 1600 1600

Number of patterns in production set: 100 100 100 100

R squared: 0.6825 0.4470 0.5106 0.6954

r squared: 0.6841 0.4384 0.5116 0.7024

Mean squared error: 0.002 0.002 0.003 0.001

Mean absolute error: 0.037 0.038 0.042 0.032

Minimum absolute

error: 0.001 0.001 0 0

Maximum absolute

error: 0.128 0.110 0.144 0.084

Pearson’s correlation

coefficient: 0.8271 0.6580 0.7152 0.8381

Percent within 5%: 40.171 50.000 35.345 40.517

Percent within 5% to 10%: 31.624 20.192 31.897 39.655

Percent within

10% to 20%: 22.222 21.154 25.000 18.103

Percent within 20% to 30%: 3.419 7.692 6.034 1.724

Percent over 30%: 2.564 0.962 1.724 0

Probability of correct

sign of increments

forecasting: 0,79 0,78 0,78 0,89

 

Table 7. Statistics of forecast quality for main currencies of Forex for the network with four values of returns in series

CHF EUR GBP JPY

Number of epochs 2282 1764 3517 3513

Number of patterns in training set: 1600 1200 1600 1600

Number of patterns in production set: 100 100 100 100

R squared: 0.6369 0.5946 0.5103 0.5678

r squared: 0.6476 0.6030 0.5212 0.5706

Mean squared error: 0.003 0.002 0.003 0.002

Mean absolute error: 0.039 0.031 0.042 0.036

Minimum absolute

error: 0.001 0.001 0.001 0

Maximum absolute

error: 0.157 0.102 0.143 0.138

Pearson’s correlation

coefficient: 0.8048 0.7765 0.7220 0.7554

Percent within 5%: 43.363 45.545 39.823 46.018

Percent within 5% to 10%: 19.469 34.653 23.894 23.009

Percent within

10% to 20%: 30.973 14.851 29.204 28.319

Percent within 20% to 30%: 3.540 4.950 6.195 2.655

Percent over 30%: 2.655 0 0.885 0

Probability of correct

sign of increments

forecasting: 0,75 0,78 0,72 0,78

 

Hierarchical Extreme Learning Machine for

Unsupervised Representation Learning

Wentao Zhu and Jun Miao

Key Lab of Intelligent

Information Processing of

Chinese Academy of Sciences (CAS),

Institute of Computing Technology,

CAS, Beijing 100190, China

Email: wentao.zhu@vipl.ict.ac.cn;

jmiao@ict.ac.cn

Laiyun Qing

School of Computer and

Control Engineering,

University of Chinese Academy of Sciences,

Beijing 100049, China

Email: lyqing@ucas.ac.cn

Guang-Bin Huang

School of Electrical &

Electronic Engineering,

Nanyang Technological University,

Nanyang Avenue, Singapore 639798

Email: egbhuang@ntu.edu.sg

 

Abstract—Learning representations from massive unlabelled data is a topic for high-level tasks in many applications. The recent great improvements on benchmark data sets, which are achieved by increasingly complex unsupervised learning meth-ods and deep learning models with many parameters, usually requiring many tedious tricks and much expertise to tune. Additionally, the filters learned by these complex architectures are quite similar to standard hand-crafted visual features, and training to fine-tune the weights of deep architectures requires a long time. In this paper, the Extreme Learning Machine-Auto Encoder (ELM-AE) is employed as the learning unit to learn local receptive fields at each layer, and the lower layer responses are transferred to the last layer (trans-layer) to form a more complete representation to retain more information. In addition, some beneficial methods in deep learning architectures such as local contrast normalization and whitening are added to the implemented hierarchical Extreme Learning Machine networks to further boost the performance. The resulting trans-layer representations are processed into block histograms with binary hashing to produce translation and rotation invariant representations, which are utilized to do high-level tasks such as recognition and detection. The proposed trans-layer representa-tion method with ELM-AE based learning of local receptive filters was tested on the MNIST digit recognition data set, including MNIST variations, and on the Caltech 101 object recognition database. Compared to traditional deep learning methods, the proposed ELM-AE based system has a much faster learning speed and attains 65.97% accuracy on the Caltech 101 task (15 samples per class) and 99.45% on the standard MNIST data set.

I. INTRODUCTION

Almost all tasks such as classification, and recognition require us to design fine representations for their specific aims. For classification of images taken in natural settings, numer¬ous factors in the environment, such as lighting conditions, occlusions, corruptions and deformations, lead to large amount of intra-class variability. Good representations should reduce such non-informative intra-class variability, whilst preserving discriminative information across classes. However, designing good feature representations is a quite tough and difficult procedure for pattern recognition tasks [37].

This research is partially sponsored by Natural Science Foundation of China (Nos. 61272320 and 61472387).

 

Two basic approaches to research on feature representations are hand-crafted design and automated machine learning. For decades, enormous numbers of researchers and engineers have devised robust feature representations in their own domains. Many hand-crafted features are successfully proposed such as SIFT [1] and HoG [2] features in computer vision domain. However, these hand-crafted features have poor transferabil¬ity across domains. Novel features need to be redesigned elaborately when the domain of application is changed. The other way is machine learning of representations, which is a quite prevalent topic following the recent successes of deep learning [3]. Nevertheless, these fully learned representations by multilayer unsupervised learning followed by a fine-tuning procedure have too many parameters to be tuned, and require much expertise knowledge and sophisticated hardware.

In this paper, we implement a novel hierarchical neural network using the Extreme Learning Machine Auto Encoder (ELM-AE) [4] to very quickly learn local receptive filters and trans-layer representations. The hierarchical ELM trains the local receptive filters using ELM-AE without trivial BP fine-tuning, and the responses of the previous layer of our model are concatenated to that of the last layer to form a more complete representation. These trans-layer connections effec¬tively reduce the rapid information loss afflicting multi-layer unsupervised learning as commonly suffered by conventional NN architectures. In addition, local contrast normalization [5] and whitening are added in our hierarchical ELM network to boost its learning ability. The major difference between our hierarchical ELM network and conventional deep learning networks is that the hierarchical ELM network requires no back propagation information to fine-tune the feature banks with a trans-layer representation.

Experimental results indicate that the hierarchical ELM network boosts the ELM based methods effectively. The hierarchical ELM network is validated on digit recognition, object recognition tasks. Experimental results show that, the hierarchical ELM representations achieve accuracy of 6597% on Caltech 101 [7] 15 samples per class task and 9945% on MNIST data set [8].

We will start by reviewing related work on representation learning in Section II. Then the idea of the hierarchical

 

ELM network, including the pre-processing, ELM-AE learning local receptive filters, and trans-layer learning, is illustrated detailedly in Section III. How to use the hierarchical ELM network to extract features and tackle applications is also described in Section III. The experimental results and com-parative analysis on MNIST, MNIST variations, Caltech 101 data sets are presented on Section IV. Finally, discussion and conclusion are summarized in Section V.

II. RELATED WORKS

Much research has been conducted to pursuit a good representation by manually designing elaborative low-level features, such as SIFT feature [1], HOG feature [2] in computer vision field. However, these hand-crafted features cannot be easily adapted to new conditions and tasks, and redesigning them usually requires novel expertise knowledge and tough studies.

Learning representations is probably a promising way to handle the required elaborative expertise problem in devising hand-crafted features. Much recent work in machine learning has focused on how to learn good representations from massive unlabelled data, and great progresses have been made by deep neural networks [11]–[13]. The main idea of deep models is to learn multi-level features at different layers. High-level features generated in the upper-layer are expected to extract more complex and abstract semantics of data, and more invariance to intra-class variability, which is quite useful to high-level tasks. These deep learning methods typically learn multi-level features by greedily “pre-training” each layer using the specific unsupervised learning, and then fine-tuning the pre-trained features by stochastic gradient descent method with supervised information [3], [5]. However, these deep architectures have numerous parameters such as the number of features to learn, and parameters of unsupervised learning in each layer. Besides, the stochastic gradient descent also has various parameters such as momentum, weight decay rate, learning rate, and extra parameters including the Dropout rate or DropConnet rate in recently proposed convolutional deep neural networks (ConvNets) [11], [12].

There are also some works on conventional unsupervised learning methods with only single layer [10], [13]. The main idea of these methods is to learn over-complete representa¬tions with dense features. Although these methods have made progresses on benchmark data sets with almost no hyper parameters, these single layer unsupervised representational learning methods require over-complete features of dimensions as high as possible and the parameters need to be elaborately chosen in order to obtain satisfactory results [13].

A major drawback of deep learning methods with fine-tuning for learning representations is the big consumption of expensive computational resources and high complexity of the models. One intuition is that, since the elaborately learned features are quite similar to some conventional un¬supervised features, such as wavelets and PCA, why not jump over the tough and time-consuming fine-tuning procedure and take those features stacked directly. Furthermore, more robust invariant features can be better devised other than various pooling strategies. Wavelet scattering networks (ScatNet) are such networks with pre-fixed wavelet filter banks in the deep 

 

convolutional architectures [14]. The ScatNets have quite solid mathematical analysis of their rotation and translation invariants at any scale. More surprisingly, superior perfor¬mance over ConvNet and deep neural networks is obtained by the ScatNets’ pre-fixed feature bank networks with no fine-tuning phase. However, the ScatNet is shown to have inferior performance in large intra-class variability including great illumination changing and corruption [15].

The other non-propagation deep network with pre-fixed feature banks is PCANet [15]. The PCANet uses two layer cascaded linear networks with na¨ıve PCA filter banks to extract more complex features. The PCANet also utilizes the histogram operator popular in computer vision domain to retain translation and rotation invariance [1], [2]. The output of the two layer cascaded PCA network is processed by the quantized histogram units. The PCANet presents a superior or highly comparable performance over other methods such as ScatNet [14], ConvNet [11] and HSC [16], especially in face recognition tasks with large occlusion, illumination, expression and pose changes. However, the cascaded PCA structure in PCANet will face great information loss and corruption after multilayer transformation. Besides, the learning ability of na¨ıve PCA unsupervised learning is much weaker than other learning units.

On the other hand, Extreme Learning Machine (ELM) is a popular method famous for its fast learning speed and good generalization ability [17], and has been used in many applications [18], [19]. Many methods have been proposed to further boost its performance [20]–[22]. Also, ELM has been extended to multilayer neural networks [4], [23]. However, the accuracies of these methods are lower than the current deep learning methods.

This paper demonstrates a novel method, the hierarchical ELM network, using the Extreme Learning Machine (ELM) to learn the local receptive fields fast without trivial BP fine-tuning. Moreover, the hierarchical ELM network combines the first layer representation with the second layer representation. Furthermore, the local contrast normalization and whitening operators are applied to ameliorate the ELM filter learning in the hierarchical ELM network. Thus the implemented network can form a more complete and efficient representation, whilst retaining the advantages such as faster learning speed and more reliable performance. Also, the successfully used histogram operator in computer vision domain is adopted to preserve translation and rotation invariance after binary hashing quanti-zation. Experimental results confirm that the hierarchical ELM network learns effective feature representations that contribute to object recognition and digital recognition tasks.

III. HIERARCHICAL EXTREME LEARNING MACHINE

In this section, we present a novel framework, the hierar-chical ELM network, for unsupervised representation learning, illustrated in Fig. 1. The procedures of the hierarchical ELM network is similar to other commonly used frameworks in computer vision [15], [24] and other feature learning work [13] as well. Different from the traditional methods, the hierarchical ELM network utilizes the unsupervised Extreme Learning Machine-Auto Encoder (ELM-AE) to learn the local receptive features, and needs no fine-tuning procedure to adjust

 

those local filters. Besides, the previous layer’s unsupervised features transformed by ELM-AE local receptive features are concatenated to the last layer to form a much more completed representation, which is shown quite effective for the following tasks.

Applications: Classification, recognition, -- 

Fig. 1: The framework of the hierarchical ELM network. The hierarchical ELM includes three layer, the first layer of ELM-AE local receptive mapping, the second layer of ELM-AE local receptive mapping, and the output layer with previous responses inputs from previous layers.

In the beginning of each layer (the first and second layer), the system extracts large numbers of random sub-patches from unlabelled input images. Each patch has dimension k1-by-k2, called receptive field size. (Suppose the images used here are all gray images. Actually, if the images are colored images with d channels, the patch dimension is k1-by-k2-by-d. Just process the other d  1 channels the same as following procedures step by step independently.) Then a data set of m patches is constructed, X = x(1), •••, x(), where x() R12 stands for the ith patch extracted from the input images in the first layer or the input feature maps in the second layer. Given the data set before the processing of the first and second layer, we apply the preprocessing of local contrast normalization (LCN) and whitening, and Extreme Learning Machine-Auto Encoder (ELM-AE) unsupervised learning of each layer to learn local receptive fields in the first and second layer. 

 

A. Pre-processing-Local Contrast Normalization and Whiten¬ing

The first pre-processing method of each ELM-AE layer is local contrast normalization (LCN) [5]. For each local patch x() in the extracted patch data set X, we normalize the patch x() by subtracting its mean and dividing by its standard deviation as,

y() = (x() 1 

12

VI~

1: 1 1:2 1: 1 1:2

( 1 (x() 1 x())2 + C),

12 12

=1 =1 =1 =1

j= 1,••• , k1; k = 1,••• , k2; i = 1, ••• , m,

where C is a constant integer to make the model more robust commonly set in practice. In the work, we set C as 10.

( i)

Output

T

P1T PL

1 ... L

(WL, bL)

(W1, b1)

k1k2

( i)

Fig. 2: The framework of ELM-AE. The ELM-AE is a network using ELM to reconstruct the input.

The LCN has explicit explanations both in physics and physiology. The mean of local patch stands for local bright¬ness, and the standard deviation represents contrast normaliza¬tion. By LCN, the illumination and material optimal property effects of the processed patches are removed. On the other hand, the LCN has an effect similar to lateral inhibition found in real neurons. The LCN operator inhibits the responses within each local patch, whilst activating responses in the same location of these patches.

Following LCN, whitening is the second preprocessing method for each ELM-AE unsupervised layer. Whitening is commonly used in various applications and a decorrelation operation, which reduces redundant representation of images. Whitening is commonly used in various applications and we

 

do not describe it in details. The whitening operator transforms the patches as,

[D U]=  ( (Y))

z() = U( + ())12U y() = 1•• • (2)

where Y is formed by  patches y(), () stands for covariance function and the size of output data is 1 * 2, () is the eigenvalue decomposition function, D and U are eigenvalues and eigenvectors respectively,  is set as 01 here. The Zeros Component Analysis (ZCA) whitening also has biological explanation and has been proved its effectiveness by a lot of work.

B. Extreme Learning Machine-Auto Encoder to Learn Local Receptive Filters in Each Layer

In the hierarchical ELM network, we use a novel approach, Extreme Learning Machine-Auto Encoder (ELM-AE), to learn local receptive filters in each layer, illustrated in Fig. 2.

Assuming that the number of first layer feature banks is 1, flatten each pre-processed patch z() extracted from input images, and put the flattened vectors together. The matrix formed by the extracted  patches will be obtained as

Z1= I z1(1) z1(2) ••• z1() I C R12 (3)

The objective function of ELM-AE is

11Z1  2 (1(WZ1 + bi))1122 + 111122

 WW = I1  b b = 1

where W is the orthogonally random weights with size of 1x 12, 1 is the number of local receptive filters in the first layer, I1 is the identity matrix of size 1, b is the normalized random vector of dimensions 1, i is the column vector of size  and full of the element 1 to make b a matrix, () is the element wise sigmoid function, 1 and 2 are the factors changing the hidden nodes and output nodes distribution of ELM-AE based on intrinsic plasticity property respectively, is the output layer weight, C is the trade-off between errors and model complexity. Then the output weight  is calculated as [4]

The first layer local receptive filters are 1 • • •L1. For each input image, the first layer L1 feature maps can be obtained by

I1() = I * ii = 1 •••L1  (6)

where I stands for an input image with zero padded to make I1() has the same size as the input image, and * stands for two dimensional convolution operator.

Patches should be extracted from feature maps I1() ob¬tained from the first ELM-AE layer. These patches are also transformed by the pre-processing LCN and whitening op-erators. Then applying the same process as the first layer, we obtain the second layer feature map for each first layer feature map, I2() = 1 • • •2  where 2 is the number of filters in the second layer.

 

For an input image, we get 1 x (2 + 1) feature maps

after the two ELM convolution layers by concatenating the

first layer maps to the second layer. That is

  

I1 (1) I1 (2) • • •  I1 (1) I21  I2

(1) 1  ••• 

(2)

I21  •••  I2

(2) 1  I2

(1) 1  •••  I2

(2) (2)  (7)

1

where I1() stands for the th feature map in the first layer, and I2  stands for the th feature map of the th first layer ()

feature map in the second layer.

C. Block-wise Histogram

Fig. 3: The output layer of the hierarchical ELM network. Firstly, quantize the responses from previous layers into binary values. Secondly, compress these responses. Thirdly, use the histogram to retain the invariance.

The output layer of the hierarchical ELM network is illustrated in Fig. 3. For an input training image, the first step is to use binary hashing to make the constructed real valued

1 x (2 + 1) feature maps have only two values, 0 and 1. The second step is to compress these binary feature maps by quantizing each 1 binary feature maps. The number of second layer filters 1 is set as 8, and the number of first layer filters

2 is set the same as 1. That is, we compress each 1 binary feature maps into one feature map, and the compressed feature maps have pixel values from 0 to 255 [15]. Then we get 2+1 compressed feature maps for each training image as

(

2+1)1

I (1) I (2) • • •  I(2) I  (8)

The third step is to construct block-wise histogram il¬lustrated as the third procedure in Fig. 3. First, we should partition each encoded feature map. Assuming that the size of

 

compressed map is x x y, and the size of block is w1 x w2 with strides s1xs2, each compressed feature map is partitioned into L(x  w1)/s1 + 1j x L(y  w2)/s2 + 1j blocks. For all the L2 + 1 compressed feature maps of each input image, we get blocks as

{B1, B2, ••• , B(L2+1)L(xw1)/s1+1JL(yw2)/s2+1f ~ , (9)

where Bi stands for the ith blocks constructed from the compressed feature maps. Next step is to build histograms in each of the blocks. In the hierarchical ELM network, we set the number of bins in the histogram as 2L1. It means that each integer of the pixel values is set as a bin and a sparse vector representing the histogram is constructed. Then concatenate these NB histograms to form the completed representation of the input image as

f(Image) = [hist(B1)T , hist(B2)T , •••,hist(BNB )T ]T E RNB(2L1 ), (10)

where hist()stands for histogram operator, and NB = (L2 +

1) x L(x  w1)/s1 + 1j x L(y  w2)/s2 + 1j. Then we use the hierarchical ELM network representation of each training images to learn a dimension reduction weight, or to train a classifier to tackle the next applications directly.

IV. PERFORMANCE EVALUATION

In this section, we will introduce how to train the hierar-chical ELM network, including how to select the parameters. The hierarchical ELM networks are validated on MNIST [8], MNIST variations [9] and Caltech 101 [7] data sets. MNIST and MNIST variation data sets are grey hand-written digits of ten classes and size 28 x 28. Other description of the data sets are illustrated as the Table I. Caltech 101 data set contains color images belonging to 102 categories including a background class. The number of each class’s images varies from 31 to 800.

TABLE I: Data Set Description about the Size of Training, Validation and Testing.

Data Sets Training Validation Testing

MNIST 50,000 10,000 10,000

MNIST basic 10,000 2,000 50,000

MNIST Rotation 10,000 2,000 50,000

MNIST back-rand 10,000 2,000 50,000

MNIST back-image 10,000 2,000 50,000

MNIST back-image-rotation 10,000 2,000 50,000

Convex vs. non-convex 6,000 2,000 50,000

A. How to Train

In the hierarchical ELM network, there are mainly two kinds of parameters, local receptive fields related parameters and ELM-AEs related parameters.

For the local receptive fields related parameters such as the block size, the stride size and the patch size are determined through cross-validation. On all the tasks, the filter size is set to 7 x 7 pixels, the number of filters is set to L1 = L2 = 8 and the size of strides is the half size of the block. For MNIST, MNIST basic, and mnist-rotation data sets, the block 

 

size is 7 x 7 pixels. For mnist-back-rand, mnist-back-image and mnist-rot-back-image data sets, the block size is 4 x 4 pixels. For convex data set, the block size is 28 x 28. For Caltech-101 data set, the block size is set to a quarter of the image size, and the whitening PCA (WPCA) is used to reduce the dimension of each block representation from 256 to 64. A simple linear SVM with default parameters C = 1 is connected to the hierarchical ELM representation to do recognition with much fast learning speed [26].

For the ELM-AEs related parameters such as C, α1, α2 in each layer, we selected these parameters by layer-wise greedy validation on the validation set. We first selected the parameter

C, then fixed it and selected α1, α2. α1, α2 were selected in a decreasing order 1, 0.9, 0.8, • • •, 0.5. For MNIST data set, we used the parameters the same as the MNIST basic data set. The training set of Caltech 101 was constructed as randomly selected 15 samples for each class, and the validation set of Caltech 101 was constructed the same as training set. The parameters of each data set are listed in the Table II.

TABLE II: Parameters of ELM-AEs on Each Data Set

Data Sets The first layer The second layer

α1 α2 C α1 α2 C

Basic 0.9 0.7 106 1 0.6 106

Rot. 1 0.7 105 0.8 1 105

Back-rand 0.7 0.9 102 0.7 1 107

Back-im. 0.8 0.5 108 0.7 0.7 108

Back-im.-rot. 1 0.7 103 1 1 107

Conv. 1 1 104 1 0.9 106

Caltech-101 0.5 1 102 0.6 0.9 107

B. Effect of Whitening

We validated the whitening effect of ELM-AE filter learn-ing. Whitening is an effective and helpful pre-processing procedure for many methods such as PCA and conventional deep models. However, the effect of combining whitening with ELM-AE has not been validated. The parameters of the hierarchical ELM network without whitening or with whitening are both selected as the same procedure as the above description respectively. The performances of the hierarchical ELM network with whitening pre-processing and without pre¬processing on MNIST and MNIST variation data sets are listed in Table III.

From Table III, we observe that whitening pre-processing boosts the hierarchical ELM networks. That is to say, whiten¬ing is also an effective pre-processing method in ELM-AE filter learning. We also printed the filter maps of the hierarchi¬cal ELM networks on the seven data sets. The ELM-AE filters with whitening and without whitening are illustrated from Fig. 4. to Fig. 9.

From these figures, whitening seems to alter the crispness of the images. However, from the performance view, ELM-AEs with whitening have better performance. The seeming ironical results bring two possible causes. The first is whether the seeming better filters are the true better filters in the hierarchical models, at least in the hierarchical ELM networks. The second is the ELM-AEs cannot learn the representation

 

of small patches robustly since the compressive representation loses lots of information in ELM-AE. We will further investi-gate the problem in the future and it seems a quite promising question.

C. Digital Recognition on MNIST and MNIST Variations Data Sets

We tested our implemented hierarchical ELM network on the MNIST, MNIST variations data sets as the protocols of these data sets. The results are listed in Table IV and Table V.

TABLE III: The Error Rates on MNIST Data Set. The Perfor-mance of Hierarchical ELM Is 9945% with A Fast Speed. Note That The Performances Are Only Those of Methods without Using Augmented Samples.

Method Error Rate (%) Training Time

SDAE [35] 1.28 17 hours

SAE [35] 1.4 17 hours

DBM [36] 0.95 19 hours

DBN [36] 1.13 5.7 hours

ML-ELM [4] 0.97 7.5 mins

HSC [16] 0.77 -

ConvNet [5] 0.53 -

Conv. Maxout + Dropout [27] 0.45 -

ScatNet (rbf ) [14] 0.43 -

PCANet [15] 0.66 -

Hierarchical ELM 0.55 1.5 hours

From Table IV, the results show that the hierarchical ELM network is only inferior to ScatNet-2 and Convolution Network related methods with enhanced tricks. It is worthy to mention that the performance of ScatNet-2 is achieved by being con-nected with a non-linear SVM with RBF kernels with tuned parameters, but our model is connected with linear SVM with all default parameters ( = 1). Our models performance (0.55) is highly close to that of ConvNet (0.53) on MNIST dataset [5]. Also, the learning speed of the hierarchical ELM network is fast, 1.5 hours to get the representations of all training samples on Intel(R) Core(TM) i7-3770 CPU @ 3.4GHz 32GB RAM, and it boosts the performances of ELM based learning methods greatly.

TABLE IV: The Error Rates (%) of Hierarchical ELM Network with Whitening and without Whitening. The Performances using Whitening Are Better than Those without Whitening Consistently.

Data Sets Without whitening With whitening

MNIST 0.78 0.55

MNIST basic 1.05 1

MNIST Rotatation 7.09 7.02

MNIST back-rand 6.17 6.13

MNIST back-image 11.54 9.89

MNIST back-image-rotation 39.03 34.45

Convex vs. non-covex 5.88 3.81

From Table V, we observe that the hierarchical ELM network achieves the best performance on four of six data

The matlab code can be downloaded from https://github.com/wentaozhu/ Hierarchical-ELM-Network.git

 

TABLE V: The Error Rates (%) on MNIST Variations Data Sets. The Performance of Hierarchical ELM Is The Best on Average.

Method basic rot bk-rand bk-im bk-im-rot conv

CAE-2 [28] 2.48 9.66 10.9 15.5 45.23 -

TIRBM [29] - 4.2 - - 35.5 -

PGBM+DN-1 [30] - - 6.08 12.25 36.76 -

ScatNet [14] 1.27 7.48 18.4 12.3 50.48 6.5

PCANet [15] 1.06 7.37 6.19 10.95 35.48 4.36

Hierarchical ELM 1 7.02 6.13 9.89 34.45 3.81

sets with a simple linear SVM classifier. It is sufficient to indicate that the hierarchical ELM network works well in the convolutional structure with ELM-AE local receptive filters and trans-layer structure.

D. Object Recognition on Caltech 101 Data Set

The pre-processing of the data set is to convert the images into grey scale, and adjust the longer side of the image to 300 with preserved aspect ratio. Two tasks are conducted. One is with a training set of 15 samples per class. The other is with a training set of 30 samples per class. The training sets are randomly sampled from Caltech 101, and the rest are testing set. Five rounds of experiments are recorded, and the performance is recorded as the average of the five rounds of results. The results are listed in Table VI.

TABLE VI: The testing Accuracy Rates (%) on Grey Level Images of Caltech 101 Data Set. The Hierarchical ELM Achieves 6597% and 7374% on 15 samples per class and 30 samples per class respectively.

Methods 15 samples per class 30 samples per class

CDBN [25] 57.70  1.50 65.40 0.50

ConvNet [31] 57.60  0.40 66.30 1.50

DeconvNet [32] 58.60  0.70 66.90 1.10

Chen et al. [33] 58.20  1.20 65.80 0.60

Zou et al. [34] -66.5

HSC [16] -74.0

PCANet [15] 61.46  0.76 68.56 1.01

Hierarchical ELM 65.97  0.69 73.74  0.37

The Table VI shows that the hierarchical ELM network gets the accuracy of 6597% trained by 15 samples per class and 7374 on 30 samples per class tasks respectively. The demonstrated network with ELM local receptive filters really makes a high progress for the data set and ELM based learning.

V. CONCLUSION AND DISCUSSION

In this paper, a novel unsupervised representation model, the hierarchical ELM network, is implemented. The method learns the local receptive filters by ELM-AE without trivial BP fine-tuning. Several key elements, such as LCN, whitening and trans-layer representation are combined together to boost the performance. We also provide a training procedure of ELM-AE in the hierarchical network. The hierarchical ELM network achieves accuracy of 6597% on Caltech 101 15 samples per

 

class task and 9945% on MNIST data set with relatively fast learning speed.

However, as presented in the experiment section, the hier¬archical ELM network cannot robustly learn the visually fine local receptive filters. There seems a promising research topic in the future to boost the ELM-AE learning ability on filters. Also, learning filters by ELM-AE on other convolutional architectures is a promising topic. And the hierarchical ELM is just a start. There is a lot of room for great improvements to come from further extensions of the hierarchical ELM.

Fig. 4: Local receptive filters of 2 layers’ ELM-AE on MNIST basic data set with whitening (up 16) vs. without whitening (bottom 16)

 

Fig. 5: Local receptive filters of 2 layers’ ELM-AE on MNIST rotation data set with whitening (up 16) vs. without whitening (bottom 16)

 

 

 

Fig. 7: Local receptive filters of 2 layers’ ELM-AE on MNIST back-im data set with whitening (up 16) vs. without whitening (bottom 16)

Fig. 8: Local receptive filters of 2 layers’ ELM-AE on MNIST back-im-rot. data set with whitening (up 16) vs. without whitening (bottom 16)

[2] N. Dalal, and B. Triggs, “Histograms of oriented gradients for human detection,” In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 2005), June 2005, Vol. 1, pp. 886-893, IEEE.

[3] G. E. Hinton, and R. R. Salakhutdinov, “Reducing the dimensionality of data with neural networks,” Science, vol. 313(5786), pp. 504-507, 2006.

[4] L. L. C. Kasun, H. Zhou, G.-B. Huang, and C. M. Vong, “Represen-tational learning with extreme learning machine for big data,” IEEE Intelligent Systems, vol. 28(6), pp. 31-34, December 2013.

[5] K. Jarrett, K. Kavukcuoglu, M. Ranzato, and Y. LeCun, “What is the best multi-stage architecture for object recognition?,” In Proceedings of IEEE 12th International Conference on Computer Vision (ICCV 2009), pp. 2146-2153, September 2009.

[6] A. Coates, A. Y. Ng, and H. Lee, “An analysis of single-layer networks in unsupervised feature learning,” In Proceedings of International Con¬ference on Artificial Intelligence and Statistics, pp. 215-223, 2011.

[7] L. Fei-Fei, R. Fergus, and P. Perona, “Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories,” Computer Vision and Image Understanding, vol. 106(1), pp. 59-70, 2007.

 

 

Fig. 6: Local receptive filters of 2 layers’ ELM-AE on MNIST back-rand data set with whitening (up 16) vs. without whiten¬ing (bottom 16)

REFERENCES

[1] D. G. Lowe, “Distinctive image features from scale-invariant keypoints,” International journal of computer vision, vol. 60(2), pp. 91-110, 2004.

 

Fig. 9: Local receptive filters of 2 layers’ ELM-AE on Convex data set with whitening (up 16) vs. without whitening (bottom 16)

 

[8] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” In Proceedings of the IEEE, vol. 86(11), pp. 2278-2324, 1998.

[9] H. Larochelle, D. Erhan, A. Courville, J. Bergstra, and Y. Bengio, “An empirical evaluation of deep architectures on problems with many factors of variation,” In Proceedings of the 24th international conference on Machine learning, pp. 473-480, ACM, June 2007.

[10] Q. V. Le, A. Karpenko, J. Ngiam, and A. Y. Ng, “ICA with recon-struction cost for efficient overcomplete feature learning,” In Proceedings of Advances in Neural Information Processing Systems, pp. 1017-1025, 2011.

[11] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” In Proceedings of Advances in neural information processing systems, pp. 1097-1105, 2012.

[12] L. Wan, M. Zeiler, S. Zhang, Y. L. Cun, and R. Fergus, “Regularization of neural networks using dropconnect,” In Proceedings of the 30th International Conference on Machine Learning (ICML-13), pp. 1058¬1066,2013.

[13] A. Coates, and A. Y. Ng, “Learning feature representations with k-means,” In Neural Networks: Tricks of the Trade, Lecture Notes in Computer Science, vol. 7700, pp. 561-580, Springer Berlin Heidelberg, 2012.

[14] J. Bruna, and S. Mallat, “Invariant scattering convolution networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35(8), pp. 1872-1886, 2013.

[15] T. H. Chan, K. Jia, S. Gao, J. Lu, Z. Zeng, and Y. Ma, “PCANet: A Simple Deep Learning Baseline for Image Classification?,” arXiv preprint arXiv:1404.3606, 2014.

[16] K. Yu, Y. Lin, and J. Lafferty, “Learning image representations from the pixel level via hierarchical sparse coding,” In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2011, pp. 1713-1720.

[17] G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew, “Extreme learning machine: a new learning scheme of feedforward neural networks,” In Proceedings of International Joint Conference on Neural Networks, vol.2, pp. 985-990, July 2004.

[18] W. Zhu, J. Miao, J. Hu, and L. Qing, “Vehicle detection in driving simulation using extreme learning machine,” Neurocomputing, vol. 128, pp. 160-165, 2014.

[19] Q. Leng, H. Qi, J. Miao, W. Zhu, and G. Su, “One-Class classification with extreme learning machine,” Mathematical Problems in Engineering, in press.

[20] W. Zhu, J. Miao, and L. Qing, “Constrained extreme learning machine: a novel highly discriminative random feedforward neural network,” International Joint Conference on Neural Networks (IJCNN), pp. 800¬807,2014.

[21] W. Zhu, J. Miao, and L. Qing, “Robust regression with extreme support vectors,” Pattern Recognition Letters, vol. 45, pp. 205-210, 2014.

[22] W. Zhu, J. Miao, and L. Qing, “Extreme support vector regression,” In Extreme Learning Machines 2013: Algorithms and Applications, pp. 25-34, Springer International Publishing, 2014.

[23] Z. Bai, G.-B. Huang, L. L. C. Kasun, C. M. Vong, and C. L . P. Chen, “Local receptive fields based extreme learning machine,” IEEE Computational Intelligence Magazine, vol. 10(2), 2015.

[24] Y.-L. Boureau, F. Bach, Y. LeCun, and J. Ponce, “Learning mid¬level features for recognition,” In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 13-18 June 2010, pp. 2559-2566.

[25] H. Lee, R. Grosse, R. Ranganath, and A. Y. Ng, “Convolutional deep belief networks for scalable unsupervised learning of hierarchical repre¬sentations,” In Proceedings of the 26th Annual International Conference on Machine Learning, pp. 609-616, ACM, June 2009.

[26] R. E. Fan, K. W. Chang, C. J. Hsieh, X. R. Wang, and C. J. Lin, “LIBLINEAR: A library for large linear classification,” The Journal of Machine Learning Research, vol. 9, pp. 1871-1874, 2008.

[27] I. J. Goodfellow, D. Warde-Farley, M. Mirza, A. Courville, and Y. Ben-gio, “Maxout networks,” arXiv preprint arXiv:1302.4389, 2013.

[28] S. Rifai, P. Vincent, X. Muller, X. Glorot, and Y. Bengio, “Contractive auto-encoders: Explicit invariance during feature extraction,” In Proceed¬ 

 

ings of the 28th International Conference on Machine Learning (ICML-11), pp. 833-840, 2011.

[29] K. Sohn, and H. Lee, “Learning invariant representations with local transformations,” arXiv preprint arXiv:1206.6418, 2012.

[30] K. Sohn, G. Zhou, C. Lee, and H. Lee, “Learning and Selecting Features Jointly with Point-wise Gated B oltzmann Machines,” In Proceedings of The 30th International Conference on Machine Learning, pp. 217-225, 2013.

[31] K. Kavukcuoglu, P. Sermanet, Y. L. Boureau, K. Gregor, M. Mathieu, and Y. L. Cun, “Learning convolutional feature hierarchies for visual recognition,” In Proceedings of Advances in neural information process¬ing systems, pp. 1090-1098, 2010.

[32] M. D. Zeiler, D. Krishnan, G. W. Taylor, and R. Fergus, “Deconvo-lutional networks,” In Proceedings of IEEE Conferences on Computer Vision and Pattern Recognition (CVPR), June 2010, pp. 2528-2535.

[33] B. Chen, G. Polatkan, G. Sapiro, L. Carin, and D. B. Dunson, “The hierarchical beta process for convolutional factor analysis and deep learn¬ing,” In Proceedings of the 28th International Conference on Machine Learning (ICML-11), pp. 361-368, 2011.

[34] W. Zou, S. Zhu, K. Yu, and A. Y. Ng, “Deep learning of invariant features via simulated fixations in video,” In Proceedings of Advances in Neural Information Processing Systems, pp. 3212-3220, 2012.

[35] P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, and P. A. Manzagol, “Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion,” The Journal of Machine Learning Research, vol. 11, pp. 3371-3408, 2010.

[36] R. Salakhutdinov, and H. Larochelle, “Efficient learning of deep boltz-mann machines,” The Journal of Machine Learning Research, vol. 9, pp. 693-700, 2010.

[37] B. Widrow, and J. C. Aragon, “Cognitive memory,” Neural Networks, vol. 41, pp. 3-14, 2013.

 

 

Sparse Recovery Using Sparse Matrices

 

The MIT Faculty has made this article openly available. Please share how this access benefits you. Your story matters.

Citation Gilbert, Anna, and Piotr Indyk. “Sparse Recovery Using Sparse Matrices.” Proceedings of the IEEE 98.6 (2010): 937–947. Web. ©2010 IEEE.

As Published http://dx.doi.org/10.1109/jproc.2010.2045092

Publisher Institute of Electrical and Electronics Engineers

Version Final published version

Accessed Sat Dec 02 23:35:32 EST 2017

Citable Link http://hdl.handle.net/1721.1/70932

Terms of Use Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use.

Detailed Terms

 

 

 

 

INVITED

PAPER

Sparse Recovery Using

Sparse Matrices

Significant results in predicting the operation of equipment such as network routers,

or the results of group testing for defective items, can often be obtained

from a few samples.

By ANNA GILBERT AND PIOTR INDYK

 

ABSTRACT | In this paper, we survey algorithms for sparse recovery problems that are based on sparse random matrices. Such matrices has several attractive properties: they support algorithms with low computational complexity, and make it easy to perform incremental updates to signals. We discuss applications to several areas, including compressive sensing, data stream computing, and group testing.

KEYWORDS | Compressive sensing; expanders; sparse matrices; sparse recovery; streaming algorithms

I. INTRODUCTION

The past several years have seen a new approach to the acquisition of compressible signals. Traditional approaches first capture the entire signal and then process it for compression, transmission, or storage. In comparison, the new approach obtains a succinct approximate representation directly by acquiring a small number of nonadaptive linear measurements of the signal. For any signal x, of length n, the representation is equal to Ax, where A is an m x n matrix. The vector Ax is often referred to as the measurement vector or sketch of x. Although m is typically much smaller than n, the sketch Ax contains plenty of useful information about the signal x. In particular, the sketch of x retains enough inherent information that we can directly obtain a sparse approximation or compressed form of the signal.

Manuscript received April 20, 2009; revised November 22, 2009; accepted February 26, 2010. Date of current version May 19, 2010. This work was supported by a David and Lucille Packard Fellowship, the MADALGO (Center for Massive Data Algorithmics, funded by the Danish National Research Association), and the National Science Foundation (NSF) under Grants CCF-0728645, CCF-0910765, and DMS-0547744.

A. Gilbert is with the Department of Mathematics, University of Michigan, Ann Arbor, MI 48109-0937 USA (e-mail: annacg@umich.edu).

P. Indyk is with the Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA 02139-4307 USA (e-mail: indyk@mit.edu).

Digital Object Identifier: 10.1109/JPROC.2010.2045092

0018-9219/$2 6. 00 @2010 IEEE

 

This approach has been discovered and explored exten-sively in several different research communities, including theoretical computer science, applied mathematics, and digital signal processing. The goal of that research is to obtain encoding and recovery schemes with good compression rate (i.e., short sketch lengths) as well as good algorithmic properties (i.e., low encoding, update and recovery times).

Linear sketches have found numerous uses in several areas, including compressive sensing, data stream computing, and combinatorial group testing.

• Compressive sensing. In this area [12], [16], the signal or image x is acquired using (analog or digital) hardware, which (approximately) com¬putes a dot product of each row of the matrix A and the signal at a unit cost. Once we obtain the measurement vector Ax, we process it digitally to extract information about the signal, including significant coefficients in an orthonormal basis (e.g., wavelet or Fourier), as well as the original signal. Frequently, the number of measurements we obtain with compressed sensing hardware is much less than that of traditional hardware devices. Nevertheless, we can sample a bandlimited analog signal at a sub-Nyquist rate and still recover significant frequencies in the signal or the entire signal spectrum. For examples of compressive sensing hardware, see [13], [31], [41], and [42].

• Data stream computing. In this area [7], [24], [33], the vectors x are often very large, and cannot be represented explicitly. For example, in network measurement, xi could denote the total number of packets with destination i passing through a network router. Storing such vector x itself is typically infeasible due to its large size: each destination is represented by an IP address that is 32-bit long, and therefore the vector x has dimension n 1/4 232. Thus, it is preferable to maintain a lower dimensional sketch

Vol. 98, No. 6, June 2010 | PROCEEDINGS OF THE IEEE 937

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

Ax instead and recover an approximation to x from the sketch. However, it must be possible to maintain such sketch under incremental updates to x. For example, if a new packet arrives, the corresponding coordinate of x is incremented by 1, which should be reflected in the sketch Ax. Fortunately, this can be easily done if the sketching procedure is linear. Specifically, let Ai denote the update to the vector x after seeing a packet with destination i (i.e., Ai = 1 and 0j = 0 for j =6 i). Then, we have A(x + Di) = Ax + ADi. Since ADi is simply the ith column of A, updating the sketch can be accomplished by simply adding that column to the current sketch Ax. See, e.g., [18] and [30] for more information about using data stream algorithms for network measurement.

• Combinatorial group testing. In pooling designs or more generally combinatorial group testing [15], the vector x represents a universe of n items in total. Moreover, we know k of the elements are defective. More specifically, the vector x is the characteristic vector for the defective set so that x E {0, 1}n has exactly k entries that are 1 and (n - k) zeros. The goal of combinatorial group testing is to construct a collection of tests (called a design) to minimize the number of tests needed to find the defective set for the worst case input. The tests are represented by a matrix A that is binary, with the jth column of the ith rows equal to 1 if and only if the jth item is used by the ith test. In the simplest (boolean) setting, each test returns 1 if at least one of the elements used in the test is defective. In our setting we assume the linear model, where each test returns the number of defec¬tive elements. Note that each such test corresponds to taking the dot product of x and a test vector, and therefore is captured in the linear sketching model.1 See, e.g., [17], [28], and [36] for further discussion and recent developments in the area.

In each of these applications, it is useful (and often crucial) that the measurement matrix A be a sparse matrix, i.e., contain very few nonzero elements per column. In data stream processing, the time needed to update the sketch Ax under the update Ai is proportional to the number of nonzero elements in the vector ADi, which is equal to the number of nonzeros in the ith column of A. In experiment pooling, the design matrix A is a binary matrix that captures which compounds are pooled together and the measure¬ments Ax reflect the activity levels of the pooled compounds. In many chemical and biological applications, the assumption that compound activity is a linear function of the concentra¬tion holds only when there are not many compounds mixed together in a single pool; thus, the design matrix A should be

1In fact, we can assume an even more general setting, where we allow a general vector x E Rn, and our goal is to identify the top k most significant coefficients from the set of linear measurements. This is applicable in a setting where the entries in x represent the activity level of n compounds, or a genetic response in a biological sample.

938 PROCEEDINGS OF THE IEEE | Vol. 98, No. 6, June 2010

 

not only binary but also sparse. In other applications, sparsity can be useful for computational reasons: one can compute the matrix–vector product Ax very quickly.2

proximation error IIx - x'IIq, where x' ranges over all k-sparse vectors (i.e., that have at most k nonzero entries). Our goal is, given Ax, to find a vector x^ such that the 2p approximation error3 IIx - ^xkp is at most c > 0 times Errkq (x), i.e.,

II^x - xIIp < c • Errkq(x). (1)

Note that for any value of p, the error IIx - ^xIIp is minimized when x^ consists of the k largest (in magnitude) coefficients of x. We refer to such x^ as the "head" of the signal x, while x - x^ will be called the "tail" of x.

As mentioned earlier, we aim to design sparse recovery schemes that achieve short sketches, have low algorithmic complexity, and provide "good" recovery guarantees. In addition, the schemes described in this survey can be classified based on other characteristics, such as 1) whether the schemes are randomized or deterministic, or 2) how general is the class of signals x supported by the schemes. In the following, we elaborate on both issues.

• Randomization: we distinguish between two clas¬ses of schemes: for-each and for-all. The latter describes a scheme in which one matrix A works for all signals x. In the former case, the matrix A is chosen at random from some distribution, and for each signal x, the recovery algorithm works "with high probability" (at least 1 - 1/n).4 Naturally, schemes with the for-all property are preferable to those with the for-each guarantee (if all other parameters are the same).We note that "for-all" does not mean that the matrix is constructed in an "explicit" or efficient manner. In fact, most of the constructions presented here use the probabilistic method. Although it is possible to construct recovery schemes explicitly [2], [14], [34], such schemes tend to require more measurements.

2Specifically, the matrix–vector product can be computed in time O(ns), where s is the column sparsity of A. As we will see in Section III, in many settings, one can achieve s = O(log(n/k)), which leads to the running time of O(n log(n/k)). This compares favorably to the O(nm) time achievable for random Gaussian matrices, or to the O(nlogn) time achievable for random Fourier-like matrices.

3It is natural to consider p = q. However, as we will see later, other guarantees are also possible.

4We adopt here the terminology frequently used in computer science. Note that one could require weaker probability bounds, e.g., 1- o(1). However, all algorithms presented in this survey naturally achieve the stronger probability bound without changing the (asymptotic) bound on the number of measurements.

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

• Generality of supported signals: ideally, the recovery schemes should support arbitrary signals x. In this survey, we focus on describing such schemes. However, there has been plenty of work on algorithms supporting more restrictive classes of signals. In particular, there have been several schemes based on sparse matrices that work for (almost) exactly k-sparse signals [27], [29], [32], [37]–[39], [43], [44]. Although we do not cover them in detail, we point out relevant connections and references whenever possible.

B. Survey Summary

We present an overview of the algorithms for sparse recovery that utilize sparse measurement matrices. The description is divided into two sections: for-each algo¬rithms are covered in Section II, while for-all algorithms are described in Section III. Historically, most of the for-each schemes have been developed in the data stream community during 2001–2004. In contrast, most of the algorithms with for-all guarantees have been discovered after 2004, during the process of unifying the ideas of compressive sensing and data stream algorithms. We present the algorithms in the same chronological order.

Almost all schemes described here offer sketch length bounds of Oðklog nÞ or less, which matches or is close to the lower bound of Qðk logðn=kÞÞ shown in [4]. They are supported by efficient algorithms, with running times ranging from polynomial in n to near-linear in n. They offer a variety of approximation guarantees, starting from a"plain vanilla" guarantee of (1) with p 1/4 q 1/4 1 (the l1=l1-guarantee) to more complex (but often stronger) ones. The exact sketch length bounds, approximation guarantees, and algorithm running times are stated in Theorems 1–10.

Due to lack of space, we focus on describing only those algorithms that achieve the best known bounds and solve the sparse recovery problem formulated earlier in this section. See [19] and [20] for some of the earlier work on closely related problems, such as recovering good piece-wise constant approximations from a sketch of a signal.

II. ALGORITHMS WITH

FOR-EACH GUARANTEES

In this section, we describe algorithms that provide for-each guarantees. The algorithms were discovered and described in the context of data stream computing. The descriptions provided here are sometimes simpler than the original versions, since we ignore various issues specific to data streams (such as how to generate the random matrix A using few random bits, how to update the sketch under incremental changes to x, etc.).

A. Count-Min and Count-Median

The count-min and count-median algorithms [9] utilize

sparse random matrices where each entry is either 0 or 1.

 

Both algorithms use the same distribution of matrices, and differ only in the details of the recovery algorithms.

Each matrix A is generated in the following way. Let w be a parameter, and let h be any function from the set H of all functions h : f1; ... ; ng ! f1; ... ; wg. Each such function defines a w  n 0–1 matrix AðhÞ, such that ðAðhÞÞj;i is equal to 1 if j 1/4 hðiÞ, and is equal to 0 otherwise. Note that each column has exactly one 1.

To create the matrix A, we choose d functions h1; ... ; hd independently and uniformly at random from H. Then, we define A to be a vertical concatenation of matrices Aðh1Þ; ... ; AðhdÞ. Note that the number of rows in the matrix A is equal to m 1/4 wd.

For intuition about the construction, observe that, for any signal x, and j 1/4 1;... ; w, l 1/4 1; ... ; d, we have

ðAxÞðl1Þwþj 1/4 ðAðhlÞxÞj1/4 ~ xi:

i:hlðiÞ1/4j

That is, the coordinate of the sketch corresponding to the function hl and value j is simply the sum of all values xi such that i is mapped to j by hl. For a fixed value of i, the sums Et:hlðtÞ1/4hlðiÞ xt contain approximations of xi, contam¬inated by other coordinates mapped together with i. As a result, "aggregating" those sums over different hl provides an approximation of xi. Different aggregation methods will lead to different algorithms.

Count-Min: The count-min algorithm [9] (see also [18]) works under the assumption that x  0. In this case, computing the approximation x from Ax is particularly simple: we define

xi 1/4 minðAðhlÞxÞhlðiÞ1/4 min

l l

i0:hlði0Þ1/4hlðiÞ

xi0:

The guarantees for the estimator x can be derived as follows. First, observe that xi  xi , since the entries xi0 contaminating the estimation of xi can only increase the value of the estimator xi . Thus, the estimator ðAðhlÞxÞhlðiÞ with the minimum value provides the smallest approxi¬mation error. Moreover, for any coordinate xi and function index l, we have

E ðAðhlÞxÞhlðiÞxi 1/4

[ lJ E

i61/4i0 Pr1/2hlðiÞ 1/4 hlði0Þxi0

 1

w

kxk1:

By Markov inequality

~ 1

Pr ðAðhlÞxÞhlðiÞxi  w 2 kxk1  1=2

Vol. 98, No. 6, June 2010 | PROCEEDINGS OF THE IEEE 939

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

and, therefore

I J

Pr xi xi  2 kxk1  1=2d:

w

For d 1/4 Clog n, we have that the above guarantee holds for all i 1/4 1; ... ; n with probability 1  n=2d 1/4 1  1=nC1. Thus, with the same probability, we have

2

kx  xk1  kxk1:

w

The disadvantage of the above guarantee is that the error is a function of the norm of the whole vector x, not its tail. However, the probability that any of the entries in the head of x contaminate an estimator of a specific xi is at most k=w. Thus, a slightly more refined analysis5 shows that, for w 1/4 4=  k,  2 ð0; 1Þ, we have

[

Pr x  xi  =k  Errk]  1=2d:

1

For d 1/4 Clog n, this implies

kx  xk1  =k  Errk1

with probability 1  n=2d 1/4 1  1=nC1.

Count-Median: The count-min algorithm can be extend-ed to work for general signals [9]; the extension is often referred to as the count-median algorithm. The main issue to take care of is that, for general vectors x, the inequality xi  xi no longer holds, since the entries contaminating the estimator might be negative. As a result, we cannot aggregate using min. Instead, we replace the estimator x by

( )

x med i1/4 medianl ðAðhlÞxÞhlðiÞ:

By using the Chernoff bound, we show that, with high probability, the majority of the estimators ðAðhlÞxÞhlðiÞ (and therefore their median) have small error. Specifically, we can show that for any constant C0 > 0, there exists C such that if we set d 1/4 Clog n, then

~ ~

~ xmed  x~1 =k  Errk1

with probability 1  1=nC0.

5The argument is essentially a simplified version of the argument used in [6]. See [10] or [24, Lecture 4] for the proof.

940 PROCEEDINGS OF THE IEEE | Vol. 98, No. 6, June 2010

 

Theorem 1: There exists a distribution over m  n matrices A, m 1/4 Oðk=  log nÞ, such that for any signal x, given Ax, we can recover x^ 1/4 xmed such that

k^x  xk1  =k  Errk1

with high probability. The column sparsity of A is Oðlog nÞ, and the time needed to recover x^ from Ax is Oðn log nÞ.

We conclude by observing that the approximation guarantee in the above theorem implies a weaker but perhaps more intuitive guarantee about the l1 approxima¬tion error. Consider the vector x^ consisting of the k largest (in magnitude) elements of xmed. Then, we have

kx  ^xk1  ð1 þ 3ÞErrk1:

To show this, let S be the set of the k largest in magnitude coordinates of x, and let S^ be the support of ^x. Note that k^xSk1  k^x^Sk1. We have

kx  ^xk1  kxk1  kx^Sk1 þ kx^S  ^x^Sk1

 kxk1  k^x^Sk1 þ 2Errk1

 kxk1  k^xSk1 þ 2Errk1

 kxk1  kxSk1 þ 3Errk1

 ð1 þ 3ÞErrk1:

For more detailed descriptions of the algorithms, see [6], [9], and [18].

B. Count-Sketch

The next6 algorithm, called count-sketch [6], provides error guarantees that are a function of Errk2 as opposed to Errk1. This is accomplished by using a distribution over matrices A very similar to those used by count-min, with one difference: each nonzero entry is chosen indepen¬dently and uniformly at random from f1; 1g (instead just being equal to 1). Formally, let ri;l be independent random variables with values chosen uniformly at random from f1; 1g, and let the functions h1; ... ; hd be defined as in the previous section. Then, the matrix A is a vertical concatenation of matrices Aðh1Þ; ... ; AðhdÞ, where ðAðhlÞÞj;i is equal to ri;l if j 1/4 hlðiÞ, and is equal to 0 otherwise. To estimate the coordinate xi, one then uses the median estimator

xmed0 1/4 medianl ri;lðAðhlÞxÞhlðiÞ:

6Chronologically, the count-sketch algorithm has been invented before count-min. It is easier, however, to describe the ideas in the reverse order.

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

The analysis of count-sketch relies on the observation that

[( ) )2

Pr ri;lAðhl Þx hlðiÞ xi  C=w Errk  1=4

2

for some absolute constant C > 0. The final guarantee is captured by the following theorem.

Theorem 2: There exists a distribution over m  n matrices A, m 1/4 Oðk= log nÞ, such that for any signal x, given Ax, we can recover x^ such that

( )2

k^x  xk2 1  =k  Errk 2

with high probability. The column sparsity of A is Oðlog nÞ, and the time needed to recover x^ from Ax is Oðnlog nÞ.

As before, the approximation guarantee in the theorem implies a weaker but more intuitive guarantee, this time about the l2 approximation error. Consider the vector x^ consisting of the k largest (in magnitude) elements of x

med0. Then, we have [11]

kx ^xk22  ð1 þ 9p---Þ Errk

( )2:

2

The proof proceeds as follows. Let E 1/4 Errk2. Let S be the set of k largest (in magnitude) coordinates of x, and let S^ be the support of ^x. Moreover, for any set P, let P denote the complement of P. We have

~ ~

kx  ^xk22  ðx  ^xÞ^S

~ ~22þkxS^Sk2 2

2 þ xðS[^SÞ

2: (2)

The first term is bounded by k=k  E2 1/4 E2. To bound the second term, we proceed as follows. Consider any i 2 S  S^and j 2 S^  S. We have

 -------  ------ 

jxij  jxjj  j^xij  j^xjj þ 2 =kE  2 =kE: (3)

Let a 1/4 maxi2S^S jxij and b 1/4 minj2^SS jxjj. From (3), we  ------ 

have a  b þ 2 =kE. Thus

 ------ 

kxS^Sk2 2  a2jS  ^Sj  ðb þ 2 =kEÞ2jS  ^Sj:

 

Since kx^SSk2 2  b2j^S  Sj 1/4 b2jS  ^Sj, we continue

( 1------------- ------- J2

kxS^Sk22  kx^SSk2 jS  ^Sj þ 2 =kE jS  ^Sj  kx^SSk2 þ 2 -- 

( p E)2

 kx^SSk2 2 þ 4kx^SSk2 p E þ 4E2

-- 

 kx^SSk2 2 þ 4 -- 

p E2 þ 4E2

kx^SSk22 þ 8p --- E2:

Plugging into (2), we get

kx  ^xk2 2  E2 þ kx^SSk22 þ 8 -- 

p E2 þ xðS[^SÞ

2

2

 9 p E2 þ kxSk2

-- 

2

1/4 ð1 þ 9 p ÞE2:

-- 

C. Sublinear Algorithms

The above algorithms all run in time at least linear in the signal size as they entail estimating a value for each coordinate in the signal, even those that are insignificant. If our goal is to just report k nonzero terms of k-sparse approximation, then it is sufficient to find (or approxi¬mate) the top k values only to achieve similar error guarantees. Sublinear algorithms aim to do just that and to do so in time that scales polynomially with the number of terms k desired and logarithmically with the length of the input signal.

We start with the simplest example of a sublinear algorithm and its associated binary measurement matrix to find the unique nonzero entry in a signal of length n and sparsity 1. Let B be the binary matrix with ith column given by the binary representation of i, beginning with the first column i 1/4 0. We refer to this matrix as a bit-tester matrix.7 We add a row of 1’s to the bit-tester matrix (to estimate the signal value) and refer to this matrix as B1. It has logðnÞ þ 1 rows and n columns and from the measurements B1x of a vector x with a single large entry, we can determine both the value of the entry and its position in time logðnÞ þ 1. The measurements are simply the position in binary plus an estimate of the signal value and the recovery algorithm is trivial. It also can be seen that a similar approach applies even if the signal x is not exactly 1-sparse, but contains some "small" amount of noise.

For general signals, the approach is to "augment" the algorithms and measurement matrix constructions from previous sections with the matrix B1. Recall that those algorithms used simple hash functions which map signal

7Readers familiar with coding theory might recognize B as the parity-check matrix of the Hamming code.

Vol. 98, No. 6, June 2010 | PROCEEDINGS OF THE IEEE 941

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

coordinates xi to rows j of the measurements. Implicit in the correctness proofs was the ability of those hash functions to isolate a few significant signal values from one another. More precisely, if h is chosen uniformly at random from a prespecified family H of hash functions, then

Pr [ h(i) = h(i0) ] = O(1) 

w

for some w; that is, the probability that positions i and i0 are hashed into the same measurement is low. Using arguments similar to those above, we can show that if there are only k large (or nonzero) entries that are hashed into more than k measurements, then with high probabil¬ity, a large fraction of the significant entries are hashed into separate measurements. We can view this process as a random masking of the original signal, leaving a signal with only one significant entry, to which we can apply the bit-tester matrix. More precisely, each row of our final matrix M is the pointwise (Hadamard) product between a

row in A and a row in B1. We say that M is the row tensor



product of B1 and A, M = B1 r A. Note that M has

approximately klog(n) rows.

Once we have a good estimate of a large fraction of the significant entries, we can subtract their contribution from the original measurements (exploiting the linearity of the measurement process algorithmically, in addition to its role in the application). We then repeat the process, using "fresh" measurements.

By using the above techniques, we obtain the following result [21].

Theorem 3: There exists a distribution over m x n matrices A, m = O(klog n), such that for any signal x, given Ax, we can recover x^ such that JJ^x - xJJ1 < CErrk1 with high probability. The column sparsity of A is O(logc n) for some constant c, and the time needed to recover x^ from Ax is polynomial in k and log n.

III. ALGORITHMS WITH FOR-ALL GUARANTEES

In this section, we describe algorithms that provide for-all guarantees. The algorithms have been discovered during the process of unifying the ideas of compressive sensing with those from data stream algorithms. The key part of that process has been to identify concrete properties that 1) hold for a random sparse matrix with a nonzero probability and 2) are sufficient to support efficient and accurate recovery algorithms.

One such property is based on the notion of graph expansion [2], [44]. Consider a bipartite graph G = G(A) between two node sets U and V, with IUI = n and IVI = m, such that an edge (i; j) belongs to G if and only if Aj;i = 1. Informally, such a graph is an expander, if each small

942 PROCEEDINGS OF THE IEEE | Vol. 98, No. 6, June 2010

 

enough set of the nodes in U has many neighbors in V (the formal definition is provided below).

The notion of expansion has been known to be useful for some related problems, such as constructing low-density parity-check codes. In fact, iterative decoding algorithms for such codes have been used, e.g., in [25], [27], and [44], to design sparse recovery algorithms. However, those algorithms were designed and proven to work only for the case where the signal x is either exactly k-sparse or "almost" k-sparse. In contrast, the algorithms we present here work for arbitrary input signals x.

Formally, we define unbalanced expander graphs as follows. Consider a bipartite graph G = (U; V; E), where E C U x V is the set of edges. We refer to U as the "left" part, and refer to V as the "right" part of the graph. A vertex belonging to the left (right) part is called a left (right) vertex. In our constructions, the left part will correspond to the set {1; 2; ... ; n} of coordinate indexes of vector x, and the right part will correspond to the set of row indexes of the measurement matrix. A bipartite graph is called left-d-regular if every vertex in the left part has exactly d neighbors in the right part.

Definition: A bipartite, left-d-regular graph G = (U; V; E) is an (s; d; )-expander if any set S C U of at most s left vertices has at least (1 - )dISI neighbors.

The algorithms described in this section use adjacency matrices A of the expanders graphs G: we simply set Aj;i = 1 if and only if (i; j) E E. Note that the resulting matrices are sparse, with exactly d ones per column.

What are the achievable expansion parameters? Since expander graphs are meaningful only when IVI<dIUI, some vertices must share neighbors, and hence the parameter  cannot be smaller than 1=d. Using the probabilistic method one can show that there exist (s; d; )-expanders with d = O(log(n=s)=) and m = IVI = O(slog(n=s)=2). Since our constructions require s = O(k) and  strictly bounded away from zero, the resulting matrices will have O(klog(n=k)) rows.

For many applications, one often needs an explicit expander, i.e., an expander for which we can efficiently compute the neighbor set of a given left vertex. No explicit constructions with the aforementioned parameters are known. However, it is known [23] how to explicitly construct expanders with left degree d = O((log IUI)(log s)=)1+1= and right set size (d2s1+), for any fixed  > 0. For simplicity, in the remainder of this paper, we will assume expanders with the optimal parameters.

Unlike in the for-each case,8 the algorithms in this section are known to be resilient to the measurement noise. That is, we could assume that we are given a noisy sketch vector b = Ax + , where  is the "measurement noise" vector. In that case, the error bounds in the approximation

8It should be noted that, although the for-each algorithms have not been typically analyzed for the case of noisy sketches, the algorithm themselves could very well be quite resilient to various forms of noise.

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

guarantees would have an additional term depending on  1/4 kk1=d. However, for the sake of consistency, we will focus the description on the noise-free case where b 1/4 Ax. The reader is referred to the original papers for the bounds for the noise-resilient variants of the algorithms.

A. RIP(1) and ‘1-Minimization

In this section, we give an overview of the "geometric" approach to sparse recovery using sparse matrices, introduced in [2]. The approach uses the ‘1-minimization algorithm that has been earlier shown to work for random dense matrices [12], [16]. In the noiseless case b 1/4 Ax, the algorithm proceeds by finding x^ such that A^x 1/4 b and k^xk1 is minimized.

To understand when the above algorithm performs an accurate recovery, we need the following generalized definition of the restricted isometry property.

Definition 2: An m  n matrix A is said to satisfy RIPðp; k; Þ if, for any k-sparse vector x, we have

kxkpð1  Þ  kAxkp  kxkp:

For the case of p 1/4 2, the notion was introduced9 in [12], which also showed that if a matrix A satisfies this property, then the ‘1-minimization procedure produces an accurate solution. Since then there has been a tremendous amount of study of the properties and construction of RIPð2; k; Þ [or RIP(2), for short] matrices. Unfortunately, sparse matrices cannot satisfy the RIP(2) property, unless their number of rows is "large" [8]. In particular, sparse 0–1 matrices must have at least Qðk2Þ rows.

However, it was shown [2] that such matrices can satisfy RIPðpÞ for p equal (or very close) to 1. In particular, the adjacency matrices of expander graphs do have this property.10 By earlier arguments, such matrices have Oðklogðn=kÞÞ rows, which translates into Oðklogðn=kÞÞ sketch length bound.

Lemma 4: Consider any m  n matrix A that is the adjacency matrix of an ðk; d; Þ-unbalanced expander G 1/4 ðU; V; EÞ. Then, the scaled matrix A=d satisfies the RIPð1; k; Þ property for  1/4 2.

Proof: Let x 2 Rn be a k-sparse vector. Without loss of generality, we assume that the coordinates of x are ordered such that jx1j  ...  jxnj. We order the edges et 1/4 ðit; jtÞ, t 1/4 1; ... ; dn of G in a lexicographic manner. It is helpful to imagine that the edges e1; e2;... of E are being

9The original paper [12] employed a slightly different notation using "double sided error," i.e., requiring that kxk2ð1  0Þ  kAxk2  kxk2ð1 þ 0Þ. The two definitions can be seen to be equivalent, by scaling A and setting ð1 þ Þ 1/4 ð1þ0Þ=ð1  0Þ.

10In fact, for some range of parameters, the opposite holds, i.e., 0–1 matrices that satisfy RIP(1) are adjacency matrices of expander graphs. See [2] and [8] for more details.

 

added to the (initially empty) graph. An edge et 1/4 ðit; jtÞ causes a collision if there exists an earlier edge es 1/4 ðis; jsÞ, s < t, such that jt 1/4 js. We define E0 to be the set of edges which do not cause collisions, and E00 1/4 E  E0.

Claim 5: We have

jxij  dkxk1:

ði;jÞ2E00

Proof: For each t 1/4 1;...; dn, we use an indicator variable rt 2 f0; 1g, such that rt 1/4 1 iff et 2 E00. Define a vector z 2 Rdn such that zt 1/4 jxitj. Observe that

 jxij 1/4 rt jxitj 1/4 r  z

ði;jÞ2E00 et1/4ðit ;jtÞ2E

To upper bound the latter quantity, observe that the vectors satisfy the following constraints:

• the vector z is nonnegative;

• the coordinates of z are monotonically nonincreas-ing, and at most kd of them are nonzero;

• for each prefix set Pi 1/4 f1; ... ; dig, i  k, we have krjPik1  diVthis follows from the expansion properties of the graph G;

• rjP1 1/4 0, since the graph is simple.

It follows that for any r; z satisfying the above

constraints, we have r  z  kzk1. Since kzk1 1/4 dkxk1,

the lemma follows. 

Since

kAxk1  jxit j  jxitj

et1/4ðit ;jtÞ2E0 et1/4ðit;jtÞ2E00

then Claim 5 immediately implies that kAxk1  dkxk1ð1  2Þ. Since for any x we have kAxk1  dkxk1, it

follows that A=d satisfies the RIPð1; k; 2Þ property. 

We now need to show that the RIP(1) property of the matrix A is sufficient to guarantee that the ‘1-minimization works. First, we show that any vector from the kernel of an adjacency matrix A of an expander graph (i.e., such that Ax 1/4 0) is "smooth," i.e., the ‘1-norm of the vector cannot be concentrated on a small subset of its coordinates. An analogous result for RIP(2) matrices and with respect to the ‘2-norm has been used before to show guarantees for LP-based recovery procedures.

Lemma 6: Consider any y 2 Rn such that Ay 1/4 0, and let S be any set of k coordinates of y. Then, we have

kySk1  ðÞkyk1

where ðÞ 1/4 ð2Þ=ð1  2Þ.

Vol. 98, No. 6, June 2010 | PROCEEDINGS OF THE IEEE 943

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

The proof proceeds by showing that any vector y whose Q1-norm is concentrated on a small set S of coordinates cannot satisfy Ay = 0. This is because [by the RIP(1) property] the Q1-norm of the vector A(yS) is "large," and (from the expansion property of the underlying graph) the contribution of the coordinates in the complement of S is not sufficient to reduce Ay to 0. See [2] for the formal proof.

The "smooth kernel" property is then used, as in prior work, to provide recovery guarantees for the e1-minimization. This is achieved by the following lemma, by setting u = x and v = ^x.

Lemma 7: Consider any two vectors u, v, such that for y = v - u, we have Ay = 0, and llvll1 < llull1. Let S be the set of k largest (in magnitude) coefficients of u, then

llv - ull1 < 2/(1- 2a(e)) • llu - uSll1.

The following theorem summarizes the discussion.

Theorem 8: There exists an m x n (expander) matrix A, m = O(k log(n/k)/e2), such that for any signal x, given Ax, we can recover x^ such that

llx - ^xll1 < c(c)Errk1

where c(c) -*2 as c -* 0. The column sparsity of A is O(log(n)/e2), and the recovery involves solving a linear program with O(n) variables and O(m + n) constraints.

This concludes the overview of the results of [2]. Further studies of e1-minimization algorithms for sparse matrices have been done in [43] and [29], where the authors obtained tight estimates for the number of measurements needed to recover signals of given sparsity. The papers consider somewhat different setups: in [43], one allows arbitrary sparse signals x and measurements contaminated by random Gaussian noise; in [29], the authors consider sparse nonnegative signals.

B. EMP, SMP, and Other Near-Linear

Time Algorithms

In this section, we describe a family of iterative algorithms for performing sparse recovery. Their key feature is that they enable performing sparse recovery in near-linear time while still using O(klog(n/k)) measure¬ments. The algorithms do not use linear programming; instead, they exploit various forms of voting mechanisms to converge to a solution. The specific algorithms covered are: expander matching pursuit (EMP) [26] and sparse matching pursuit (SMP) [5].11

11There is a very recent variant of SMP called sequential sparse matching pursuit [3]. We do not cover it in this survey due to lack of space.

944 PROCEEDINGS OF THE IEEE | Vol. 98, No. 6, June 2010

 

To describe the algorithms we need some notation. For a set S of nodes of a graph G, the ordered set of its neighbors in G is denoted by PG(S). The subscript G will be omitted when it is clear from the context, and we write P(u) as a shorthand for P({u}).

Both EMP and SMP proceed in a sequence of steps, where each step is similar to the median estimation process of the count-median algorithm. A minor technical difference is that the algorithms are constructed for general sparse matrices A, as opposed to block-structured ones used by count-median. Therefore, for a given sketch vector b = Ax, the median estimation vector Emed(b) is defined as

Emed(b)i = medianjEr(i)bj.

That is, each vertex i selects the entries bj where j is a neighbor of i in G, and then computes the median of those entries. One can observe that for the matrices used by count-median, the new and old estimators are identical. The basic intuitions behind the choice of the estimator transfer as well.

There is, however, one important difference: unlike in the for-each setup, here we cannot guarantee that each coordinate Emed(b)i differs from xi by only a small term. In fact, due to the deterministic nature of the process, it might be possible that, for some coordinate i, all sketch coordinates bj, j E P(i), could be highly "contaminated" by other entries of x. Thus, the algorithms do not provide guarantees for the lo. error of the recovered approxima¬tion. However, it is nevertheless possible to directly give guarantees for the l1 approximation error.

1) EMP: The first algorithm that achieved the O(klog(n/k)) sketch length bound and recovery time near-linear in n was the EMP. The algorithm and its analysis are somewhat complicated, so instead of a detailed description we provide only an overview.

EMP consists of two phases. In the first phase, the algorithm identifies a set I of coordinates of x that 1) contains "most" of the k largest (in magnitude) coeffi-cients of x and 2) for all nodes i E6 I the neighborhood sets P(i) and P(I) have "small" intersection. The first constraint ensures that we can set the coordinates ^xi of the approximation to zero for all i E6 I. The second constraint ensures that the values of sketch coordinates br(I) are not too contaminated by entries xi for i E6 I. Together, this implies that we can focus on decoding ^xI from br(I). This is accomplished during the second phase, which proceeds in a sequence of iterations. In each iteration, the algorithm identifies coordinates i E I such that most of elements of P(i) do not have any other neighbors in P(I). The algorithm then estimates the values ^xi of such coordinates (using the median estimator), eliminates them from the I, and subtracts their

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

contribution to the sketch. The process is continued until the set I becomes empty.

Since each coordinate of the approximation is estimat¬ed only once, and is never revised again, the EMP algorithm is very efficient: it runs in time proportional to the number of edges in the graph G, which is O(nlog(n/k)). The recovered vector x^ provides an approximation in the 21-norm, i.e., we have the following.

Theorem 9: There exists an m x n (expander) matrix A, m = O(k log(n/k)/a2), such that for any signal x, given Ax, we can recover x^ such that

IIx - ^xII1 < (1 + a)Errk1.

The column sparsity of A is O(log(n)/a2), and the recovery algorithm (EMP) has O(nlog(n/k)/a2) running time.

Although EMP offers excellent asymptotic guarantees, its empirical performance is not so great. Specifically, the number of measurements required by the algorithm to achieve correct recovery is suboptimal. For example, our recovery experiments on random signed k-sparse signals of length n, for k = 50 and n = 20 000, show that one typically needs at least 5000 measurements to recover the signal correctly using the EMP algorithm. In comparison, the linear-programming-based recovery algorithm for sparse matrices described earlier requires only about 450 measurements to perform the same task.12

2) SMP: The SMP borrows some of the ideas present in EMP, but it has been also influenced by the recent iterative algorithms for sparse recovery using dense matrices, such as [35]. The running time of the new algorithm is slightly higher (by a logarithmic factor) than of EMP. However, empirically, the algorithm performs successful recovery from a significantly smaller number of measurements. In particular, for the instances described above, SMP typically needs about 2000 measurements. The asymptotic bound on the number of required measurements is still O(klog(n/k)).

The recovery algorithm is iterative, in the spirit of matching pursuit [40]. In each iteration, the algorithm estimates the difference between the current approxima¬tion ^xj and the signal x from the sketch A^xj - b. The estimation, denoted by u*, is obtained by using the median estimator as in EMP. The approximation ^xj is updated by u, and the process is repeated.

Let Hl[y] be a "thresholding operator," which zeros out all but the l largest in magnitude coefficients of the argument y. Also, let C > 0 be some constant. The details of the algorithm, together with remarks about the properties used in the analysis, are depicted in Fig. 1.

The remarks rely on the following trick, borrowed from [35]: we can decompose the input signal x into the "head"

12For both algorithms, we used randomly generated 0–1 matrices with column sparsity equal to 20.

 

 

Fig. 1. Sparse matching pursuit algorithm: pseudocode and remarks on the analysis.

x' (containing the k most significant components of x) and the "tail" x - x'. Then, we can interpret the "sketch of the tail" term A(x - x') as measurement noise. That is, we can assume that the sketch b is equal to Ax' + µ', where µ' = A(x - x') and x' is k-sparse. Note that the RIP(1) property of A implies that IIA(x - x')II1 < dIIx - x'II1 = dErrk1. We define T' = IIµ'II1/d < Errk1.

From the remarks in the algorithm description, we conclude that for any j = 1, 2, ... , T, we have

II^xj - x'II1 < IIx'II1/2j + O(T').

Thus, setting the number of iterations to T = log(IIx'II1/T') guarantees that

II^xT - x'II1 = O(T') = O Errk

( ).

1

The following theorem summarizes the discussion.

Theorem 10: There exists an m x n (expander) matrix A, m = O(klog(n/k)), such that for any signal x, given Ax, we can recover x^ such that

IIx - ^xII1 < c Errk1

for an absolute constant c > 0. The column sparsity of A is O(logn), and the recovery algorithm (SMP) has O(nlog(n/k)T) running time, for T defined as above.

3) Connections to Message-Passing Algorithms: The SMP algorithm described above, as well as the aforementioned algorithms from [25], [27], and [44], can be interpreted in

Vol. 98, No. 6, June 2010 | PROCEEDINGS OF THE IEEE 945

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

a general framework of message-passing algorithms. Such algorithms structure their operations based on the bipartite graph G underlying the matrix A. Specifically, each node of the graph can be viewed as a separate processing unit, and the algorithm proceeds by the units sending messages to each other along the edges of the graph. Message-passing algorithms have numerous advan¬tages over the "centralized" ones: their computational complexity is low (if the underlying graph is sparse); they also can be easily implemented in a parallel or distributed manner.

There have been several papers on message-passing algorithms for sparse recovery problems using sparse random matrices. In [37] and [39], the authors introduced the belief propagation approach to compressive sensing, and applied it to the recovery of random signals, modeled by a two-state mixture of Gaussians. In a more recent paper [1], the authors used belief propagation on signals modeled as Gaussian-scale mixtures to obtain algorithms with an excellent empirical performance.

Message passing framework has been also used to design randomized algorithms that work in the worst case. In particular, the paper [32] introduced and analyzed such algorithms that work for arbitrary k-sparse signals. That algorithm can be viewed as an iterative generalization of the count-min algorithms described in earlier sections.

C. HHS and Sublinear Algorithms

As in Section II, there are versions of the above algorithms with sublinear running times. The main example is heavy hitters on steroids (HHS) [22]. The output of the

V

HHS algorithm is x^ where 11x - ^x112 < C(Err2 + 1/ kErr1) and its running time is k2(log n)O(1).It retains the same overall architecture as the iterative algorithms: within each 

 

step, it isolates significant entries by hashing, estimates their values, and then updates the measurements accordingly. It shares a "voting" procedure for determining significant signal entries with the EMP and SMP algorithms; however, these votes are derived from the bit tests rather than from the signal estimates directly. HHS differs from the simple sublinear algorithm we sketched in Section II in three major parts. First, in order to obtain a strong guarantee for all signals, we must hash k significant entries into O(k) measurements repeatedly, for O(logn) repetitions. The adjacency matrix of a (s, d, e) expander with s = O(k) is a way to achieve this. Second, because we use a simple bit tester B1 to identify the significant entries, we must ensure that it is applied to a signal that is sufficiently filtered; the contribution of the insignificant entries must be small enough not to pollute our estimates of the significant entry (recall that because the algorithm is iterative, estimation errors at one stage can accumulate at further iterations). Furthermore, we must carefully balance the ‘1 and ‘2 errors. To this end, we employ a second hash matrix that reduces the noise in each measurement after the first hash. In each iteration j, we keep a list of signal positions for which we have



at least k/j log klog(n/j)log(n) votes. Third, we use a separate matrix to estimate the values of the identified signal positions with the desired mixed norm error guarantee. Finally, in each iteration, we prune the list of signal positions to retain the top O(k) positions. h

Acknowledgment

The authors would like to thank J. Nelson, G. Cormode, and the anonymous reviewers for very helpful and insightful comments.

 

 

REFERENCES

[1] M. Akcakaya, J. Park, and V. Tarokh. (2009). Compressive sensing using low density frames. [Online]. Available: http://arxiv.org/ abs/0903.0650

[2] R. Berinde, A. Gilbert, P. Indyk, H. Karloff, and M. Strauss, "Combining geometry and combinatorics: A unified approach to sparse signal recovery," in Proc. 46th Annu. Allerton Conf. Commun. Control Comput., 2008, pp. 798–805.

[3] R. Berinde and P. Indyk, "Sequential sparse matching pursuit," in Proc. 47th Annu. Allerton Conf. Commun. Control Comput., 2009, pp. 36–43.

[4] K. Do Ba, P. Indyk, E. Price, and D. Woodruff, "Lower bounds for sparse recovery," in Proc. 19th Annu. ACM-SIAM Symp. Discrete Algorithms, 2010.

[5] R. Berinde, P. Indyk, and M. Ruzic, "Practical near-optimal sparse recovery in the l1 norm," in Proc. 46th Annu. Allerton Conf. Commun. Control Comput., 2008, pp. 198–205.

[6] M. Charikar, K. Chen, and M. Farach-Colton, "Finding frequent items in data streams," in 

 

Proc. Int. Coll. Autom. Lang. Programm., pp. 693–703, 2002.

[7] G. Cormode and M. Hadjieleftheriou, "Finding the frequent items in streams of data," Commun. ACM, vol. 52, pp. 97–105, 2009.

[8] V. Chandar, "A negative result concerning explicit matrices with the restricted isometry property,”2008, preprint.

[9] G. Cormode and S. Muthukrishnan, "Improved data stream summaries: The count-min sketch and its applications,”J. Algorithms, vol. 55, no. 1, pp. 58–75, 2005.

[10] G. Cormode and S. Muthukrishnan, "Summarizing and mining skewed data streams," in Proc. SIAM Int. Data Mining Conf., 2005.

[11] G. Cormode and S. Muthukrishnan, "Combinatorial algorithms for compressed sensing," in Proc. 40th Annu. Conf. Inf. Sci. Syst., Princeton, NJ, Mar. 2006, pp. 198–201.

[12] E. Cande`s, J. Romberg, and T. Tao, "Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information," IEEE Trans Inf. 

 

Theory, vol. 52, no. 2, pp. 489–509, Feb. 2006.

[13] M. Duarte, M. Davenport, D. Takhar, J. Laska, T. Sun, K. Kelly, and R. Baraniuk, "Single-pixel imaging via compressive sampling," IEEE Signal Process. Mag., vol. 25, no. 2, pp. 83–91, Mar. 2008.

[14] R. DeVore, "Deterministic constructions of compressed sensing matrices,”2007, preprint.

[15] D.-Z. Du and F. K. Hwang, Combinatorial Group Testing and Its Applications. Singapore: World Scientific, 1993.

[16] D. L. Donoho, "Compressed sensing," IEEE

Trans. Inf. Theory, vol. 52, no. 4,

pp. 1289–1306, Apr. 2006.

[17] Y. Erlich, N. Shental, A. Amir, and O. Zuk, "Compressed sensing approach for high throughput carrier screen," in Proc. 47th Annu. Allerton Conf. Commun. Control Comput., 2009, pp. 539–544.

[18] C. Estan and G. Varghese, "New directions in traffic measurement and accounting: Focusing on the elephants, ignoring the mice," ACM Trans. Comput. Syst., vol. 21, no. 3, pp. 270–313, Aug. 2003.

 

946 PROCEEDINGS OF THE IEEE | Vol. 98, No. 6, June 2010

 

Gilbert and Indyk: Sparse Recovery Using Sparse Matrices

 

[19] A. C. Gilbert, S. Guha, P. Indyk, Y. Kotidis, S. Muthukrishnan, and M. J. Strauss, "Fast, small-space algorithms for approximate histogram maintenance," in Proc. 34th Annu. ACM Symp. Theory Comput., 2002, pp. 389–398.

[20] A. C. Gilbert, Y. Kotidis, S Muthukrishnan, and M. Strauss, "One-pass wavelet decompositions of data streams," IEEE Trans. Knowl. Data Eng., vol. 15, no. 3, pp. 541–554, May/Jun. 2003.

[21] A. Gilbert, Y. Li, E. Porat, and M. Strauss. (2009). Approximate sparse recovery: Optimizing time and measurements. [Online]. Available: http://arxiv.org/abs/0912.0229

[22] A. C. Gilbert, M. J. Strauss, J. A. Tropp, and R. Vershynin, "One sketch for all: Fast algorithms for compressed sensing," in Proc. 39th Annu. ACM Symp. Theory Comput., 2007, pp. 237–246.

[23] V. Guruswami, C. Umans, and S. P. Vadhan, "Unbalanced expanders and randomness extractors from Parvaresh-Vardy codes," in Proc. IEEE Conf. Comput. Complex., 2007, pp. 96–108.

[24] P. Indyk. (2007). Sketching, streaming and sublinear-space algorithms. graduate course notes. [Online]. Available: http://stellar.mit. edu/S/course/6/fa07/6.895/

[25] P. Indyk, "Explicit constructions for compressed sensing of sparse signals," in Proc. 19th Annu. ACM-SIAM Symp. Discrete Algorithms, 2008, pp. 30–33.

[26] P. Indyk and M. Ruzic, "Near-optimal sparse recovery in the l1 norm," in Proc. 49th Annu. Symp. Found. Comput. Sci., 2008, pp. 199–207.

[27] S. Jafarpour, W. Xu, B. Hassibi, and A. R. Calderbank. (2008). Efficient and robust compressed sensing using high-quality expander graphs. [Online]. Available: http:// arxiv.org/PS_cache/arxiv/pdf/0806/0806. 3802v1.pdf

 

[28] R. Kainkaryam, A. Bruex, A. Gilbert, P. Woolf, and J. Schiefelbein, “Poolmc: Smart pooling of

MRNA samples in microarray experiments,”2010, Manuscript.

[29] M. A. Khajehnejad, A. G. Dimakis, W. Xu, and B. Hassibi. (2008). Sparse recovery of positive signals with minimal expansion. [Online]. Available: http://arxiv.org/abs/0902.4045

[30] B. Krishnamurthy, S. Sen, Y. Zhang, and Y. Chen, "Sketch-based change detection," in Proc. 3rd ACM SIGCOMM Conf. Internet Meas., 2003, pp. 234–247.

[31] J. Laska, S. Kirolos, Y. Massoud, R. Baraniuk, A. Gilbert, M. Iwen, and M. Strauss, "Random sampling for analog-to-information conversion of wideband signals," in Proc. IEEE Dallas

CAS Workshop Design Appl. Integr. Softw., 2006, pp. 119–122.

[32] Y. Lu, A. Montanari, B. Prabhakar,

S. Dharmapurikar, and A. Kabbani, "Counter braids: A novel counter architecture for per-flow measurement," in Proc. ACM SIGMETRICS Int. Conf. Meas. Model. Comput. Syst., 2008, pp. 121–132.

[33] S. Muthukrishnan. (2003). Data streams: Algorithms and applications, Invited Talk at the 14th Annu. ACM-IAM Symp. Discrete Algorithms. [Online]. Available: http://athos. rutgers.edu/~muthu/stream-1-1.ps

[34] S. Muthukrishnan, "Some algorithmic problems and results in compressed sensing," in Proc. Annu. Allerton Conf. Commun. Control Comput., 2006.

[35] D. Needell and J. A. Tropp, "CoSaMP: Iterative signal recovery from incomplete and inaccurate samples," Appl. Comput. Harmonic Anal., vol. 26, no. 3, pp. 301–321, 2009.

[36] N. Shental, A. Amir, and

O. Zuk. (2009). Rare-Allele detection using compressed se(que)nsing. [Online]. Available: arXiv:0909.0400

 

[37] S. Sarvotham, D. Baron, and R. G. Baraniuk, "Compressed sensing reconstruction via belief propagation," Electr. Comput. Eng. Dept., Rice Univ., Houston, TX, Tech. Rep. ECE-0601, 2006.

[38] S. Sarvotham, D. Baron, and R. G. Baraniuk, "SudocodesVFast measurement and reconstruction of sparse signals," in Proc. IEEE Int. Symp. Inf. Theory, 2006.

[39] S. Sarvotham, D. Baron, and

R. G. Baraniuk. (2008). Bayesian compressive

sensing via belief propagation. [Online].

Available: http://arxiv.org/abs/0812.4627

[40] J. A. Tropp and A. C. Gilbert, "Signal recovery from random measurements via orthogonal matching pursuit," IEEE Trans. Inf. Theory, vol. 53, no. 12, pp. 4655–4666, Dec. 2007.

[41] J. Tropp, M. Laska, M. Duarte, J. Romberg, and R. Baraniuk, "Beyond Nyquist: Efficient sampling of sparse bandlimited signals," IEEE Trans. Inf. Theory, vol. 56, no. 1, pp. 520–544, Jan. 2010.

[42] D. Takhar, J. Laska, M. B. Wakin,

M. F. Duarte, D. Baron, S. Sarvotham, K. Kelly, and R. G. Baraniuk, "A new compressive imaging camera architecture using optical-domain compression," in Proc. IS&T/SPIE Symp. Electron. Imaging, 2006, vol. 6065, DOI:10.1117/12.659602606509.

[43] W. Wang, M. J. Wainwright, and K. Ramchandran. (2008). Information-theoretic limits on sparse signal recovery: Dense versus sparse measurement matrices. [Online]. Available: http://arxiv.org/abs/0806.0604

[44] W. Xu and B. Hassibi, "Efficient compressive sensing with deterministic guarantees using expander graphs," in Proc. IEEE Inf. Theory Workshop, 2007, pp. 414–419.

 

 

ABOUT THE AUTHORS

Anna Gilbert received the S.B. degree from the University of Chicago, Chicago, IL and the Ph.D. degree from Princeton University, Princeton, NJ, in 1997, both in mathematics.

In 1997, she was a Postdoctoral Fellow at Yale University and AT&T Labs-Research. From 1998 to 2004, she was a member of the technical staff at AT&T Labs-Research, Florham Park, NJ. In 2004, she joined the Department of Mathematics, University of Michigan, Ann Arbor, where she is now an Associate Professor. Her research interests include analysis, probability, networking, and algorithms. She is especially interested in randomized algorithms with applications to harmonic analysis, signal and image processing, networking, and massive data sets.

Dr. Gilbert received a National Science Foundation (NSF) CAREER Award, a Sloan Research Fellowship, and a NAS Award for Initiatives in Research. 

 

Piotr Indyk received the M.S. degree from Uniwersytet Warszawski, Warsaw, Poland, in 1995 and the Ph.D. degree from Stanford University, Stanford, CA, in 2000, both in computer science.

Currently, he is an Associate Professor of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology (MIT), Cambridge. His research interests include high-dimensional computa¬tional geometry, sketching and streaming algorithms, sparse recovery and compressive sensing.

Dr. Indyk received a National Science Foundation (NSF) CAREER Award, a Sloan Fellowship, and a Packard Fellowship.

 

Vol. 98, No. 6, June 2010 | PROCEEDINGS OF THE IEEE 947

 

DEVELOPMENTS IN COMPUTER TECHNOLOGY IN THE

UNIVERSITY OF THE PHILIPPINES: USERS' EXPECTATIONS AND NEEDS

Evangeline C. Capuli

Marine Science Institute, University of the Philippines

UPPO Box 1, Diliman, 1101 Quezon City, Philippines

ABSTRACT: The widespread use of computer technology has spawned the expectations of users for speedy, efficient, and useable information because they are no longer satisfied with the traditional flow of the current surplus of information materials. In the University of the Philippines system, this need for access to useable information has materialized from recent subscriptions of CD-ROMs and from linkage with the Internet through the Engineering and Science Education Project (ESEP) of the Department of Science and Technology (DO ST). However, before the potential benefits brought about by these innovations can be fully realized, a number of barriers have to be overcome, among which are the misguided perception of library managers on the role of the network as information providers and the lack of handling skills by concerned personnel.

Introduction

In response to a dynamic information environment, users are now aware of the availability of varied information and communication technology suited to their individual requirements. These technological advances, which include computerized databases, CD-ROMs, telecommunication networks, electronic publishing, fax machines and photocopiers, have a great impact on library services, most notably in the areas of document delivery and interlibrary loan.

Background: electronic document delivery is a hot topic

In libraries of academic institutions such as the University of the Philippines, users rely heavily on serial articles to such a degree that the current percentage of requests for interlibrary loan and delivery of journal articles is much higher than in any other type of library. In a study conducted in the mid- 80's, a great majority of requests for serial articles was met with the provision of photocopies of the requested papers (Walhart 1985, cited in Kinnucan 1993). Before the introduction of the photocopying machine, it could thus be surmised how helpless libraries were in the midst of a deluge of interlibrary loan requests for journal articles since most were not, and still are not up to this present day and age, inclined to lend entire journal issues, especially bound volumes.

Telecommunication networks and fax machines have similarly accelerated the provision of requested documents for libraries that use them (Medina 1992, cited in Kinnucan 1993). The

 

51

 

introduction of computerized databases has made it possible for information users to learn of information they might have otherwise overlooked and has made it easier for librarians as well to locate which library has the needed document.

The above-mentioned innovations have undoubtedly changed the way interlibrary loan is accomplished and have made document delivery a reality (Kinnucan 1993).

Current advances

Worldwide initiatives in the information field are conceived to facilitate easy access for researchers and practitioners to their required information. The introduction of the CD-ROM in the mid-80s has made possible the access and dissemination of a large amount of bibliographic, full text and other data. CD-ROM titles are now available in the market covering a wide range of scientific disciplines. Simultaneously, the development of new interfaces and their continued update aid in extending CD-ROM feature services to a wider spectrum of the information community. This has therefore caused a change in the management of electronic searching. Librarians who were, to a certain point, considered intermediaries of information now had to focus their attention on training inexperienced users to search electronically.

As CD-ROM's popularity grew, so did the numbers of users and their different searching needs. As a result, experienced users requested features that would allow them to navigate and find exactly the information they needed from the available databases while those not as familiar with the new technology wanted the search kept simple.

However, while acknowledging the definite advantages of computer technologies in which CD-ROMs operate, Keylard (1993) mentions a few caveats that may lead to the under-utilization of CD-ROMs:

a. the misconception of end-users regarding a particular computer technology such as the CD-ROM;

b. the assumption that users and intermediaries require no training;

c. lack of understanding of the technology involved;

d. lack of integration of said technologies in library services; and

e. lack of long-term provision.

These so-called 'pitfalls' can easily be avoided or overcome by information campaigns initiated by either intermediaries or unit heads, appropriate training and, an assurance of funding support. A sound implementation plan needs to be conceived prior to the employment of such technologies. The parent organization should also make a commitment to integrate such technologies within the various library services to ensure their maximum use.

 

52

 

Information and the Internet The UPIDOST Engineering and Science Education Project (ESEP)

Dubbed as the "mother of all networks", Internet is the world's largest computer network linking other networks in various countries. According to Steven Goldstein, Internet Program Director, more than 30,000 autonomous networks registered with the National Science Foundation Network (NSFNET) have linked to Internet in 1993 (Magno 1994).

The Philippines' link to Internet was finalized in December 1993 following an allocation of P1 2.4M from the Department of Science and Technology (DOST) (Hilotin 1994, Magno 1994). This amount, equivalent to roughly half a million dollars, was for the installation and financing of PhilNet, the node which would serve as the central computer station through which remote computers can access the Internet's resources.

Institutions or individuals from the academic, research and commercial sectors connect to PhilNet on a subscription basis. The connection rates depend on the classification of the user, type of connection and speed of lines. For ESEP, presently in place are the access nodes in the University of the Philippines (UP) in Diliman, Ateneo de Manila University (ADMU), De La Salle University (DLSU), University of Santo Tomas (UST), UP Manila, UP at Los Banos, Saint Louis University in Baguio City, University of San Carlos (USC) in Cebu City, Xavier University in Cagayan de Oro City, Mindanao State University (MSU) in lligan, and the DOST.

It is widely believed that such direct link to Internet will hasten the transfer of scientific discoveries to and from the Philippines, since it takes months for such data to be printed in a scientific journal, stored in a library, and disseminated to the scientific communities of developing countries. Outside of this network, however, very few librarians possess the adequate know-how on automated library systems, Internet, and other applications of information technologies. To ensure the creation of a truly nationwide network, training courses are being proposed to update the technological know-how of these librarians.

Status of library networks in the University of the Philippines

Library networks leave their greatest impact on the librarians and users alike. Librarians have been offered more challenging opportunities for better service as new technology replaced their usual manual routines, i.e., roles as negotiators, facilitators, educators, and information brokers (SMeart, 1982). These roles, nevertheless, require training and skills development for effective database handling and better appreciation of the technology, to ensure successful integration of the technology into the traditional library services being offered.

On the other hand, users are able to access remote information in so short a time. Users are assumed, however, to understand different communication media. But since network systems are

 

53

 

designed for self-help, users are faced with either of the following obstacles: non-readiness, lack of computer skills, and worse, alienation. Users would then have to seek assistance from either the information managers or intermediaries. Hands-on seminars regularly organized, e.g., each semester or year, may become a necessity for a growing number of users. User-friendly manuals and computer routines may be produced since these can be avenues of self-training. The latter two, for our case, remain to be realized in the near future.

Conclusion and Recommendations

With the coming of what is known as a global information infrastructure (Valentin 1995) one can foresee a scenario for the world of information that is "high-tech" and highly profitable. However, information users in the University of the Philippines perceive that the librarians/information providers still need further training in order to fully understand the potential of such technologies and increase their competence in manipulating these sophisticated tools. Training should augment their experience and should specifically develop their expertise in information retrieval.

The problems we encounter, as with other Third World libraries, can likewise be alleviated if a budget can be allocated to sustain continuous on-the-job training for the librarians/information providers, because the pace of technological change is at such rate that computers have an average life span of three to five years and the electronic environment changes rapidly.

 

54

 

References

Hilotin, C.I. 1994. Computer networking need not be expensive. TODAY, 22 Apr. 1994:13.

Keylard, M. 1993. CD-ROM implementation in developing countries: impacts and pitfalls. IFLA Journal 19(1):35-49.

Kinnucan, M.T. 1993. Demand for document delivery and interlibrary loan in academic settings. Library and Information Science Research 15(4):355-374.

Magno, L 1994. Telecoms link to Internet that will flood RP with trade info being finalized. TODAY, 8 Apr. 1994:13

Stueart, RD. 1982. Libraries: a new role. In: Books, libraries and electronics: essays on the future of written communication (ed. by E. Sigel, et al.). New York: Knowledge Industry Publications, Inc.

Valentin, R. 1995. Information: a global resource. IDRC Reports 23(4):4-6.

 

55

 

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 4, APRIL 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/

Plagiarism Detection Using Graph-Based

Representation

1Ahmed Hamza Osman, 2Naomie Salim, 3Mohammed Salem Binwahlan

1Faculty of Computer Science, International University of Africa, Sudan

2Faculty of Computer Science and Information Systems, Universiti Teknologi Malaysia

3Faculty of Applied Sciences, Hadhramout University of Science & Technology, Yemen

1, 2, 3 81310, Skudai, Johor, Malaysia

Tel +60147747409/+607 5532208 Fax +607 5532210

Abstract—Plagiarism of material from the Internet is a widespread and growing problem. Several methods used to detect the plagiarism and similarity between the source document and suspected documents such as fingerprint based on character or n-gram. In this paper, we discussed a new method to detect the plagiarism based on graph re-presentation; however, Preprocessing for each document is required such as breaking down the document into its con¬stituent sentences. Segmentation of each sentence into separated terms and stop word removal. We build the graph by grouping each sentence terms in one node, the resulted nodes are connected to each other based on order of sentence within the document, all nodes in graph are also connected to top level node” Topic Signature “. Topic signature node is formed by extracting the concepts of each sentence terms and grouping them in such node. The main advantage of the proposed method is the topic signature which is main entry for the graph is used as quick guide to the relevant nodes. which should be considered for the comparison between source documents and suspected one. We believe the proposed method can achieve a good performance in terms of effectiveness and efficiency. 

— Plagiarism detection, graph representation, concept extraction, topic signature

- - - - - - - - - - 

 

P

LAGIRAISM is one of the forms of misuse of academ¬ic activities has increased rapidly in the quick and easy access to data and information through electron¬ic documents and the Internet, and when we talk about plagiarism, we mean the text written by others where they are re-adjust the text to format by adding or deleting without any citation or reference.

There are many types of plagiarism, such as copy and paste, which is the most common, redrafting or paraph-rasing of the text, plagiarism of the idea, plagiarism through translation from one language to another and many other methods that use plagiarism. Plagiarism is a serious problem in computer science. This is partly due to the ease with which electronic assignments may be co-pied, and to the difficulty in detecting similar as-signments in a sufficiently large class. In addition, students are becoming more comfortable with cheating. A recent study found that 70% of students admit to some plagiarism, with about half being guilty of a se 

• Ahmed Hamza Osman. is with the Faculty of Computer Science, Interna-tional University of Africa, Sudan.

• Naomie Salim. Faculty of Computer Science and Information Systems, Universiti Teknologi Malaysia,malsysia.

• Mohammed Salem Binwahlan. is with the Faculty of Applied Sciences, Hadhramout University of Science & Technology, Yemen.

 

rious cheating offense on a written assignment. In addition, 40% of students admit to using the “cut-and-paste” approach when completing their assign¬ments [1]. The key and main issue in plagiarism detec¬tion field is how to differentiate between plagiarized doc¬ument and non-plagiarized document in effective and efficient way.

The current methods of plagiarism detection relay on the comparison of small text unit such as character, n-gram, chunk or terms. Suppose we have a document contents ten sentence, each sentence contains five terms and each term consists of at least one character. The consideration of small text unit (character) for detecting of similarity between the original document and suspected document lead to a huge number of comparisons. In this paper, we propose a new method for plagiarism detection. The pro-posed method is graph-based, where each document is represented as graph. One node represents one sentence. Top level node is different node where it contains the concepts of terms in the document. Such node is called topic signature. The main advantage of proposed method is the topic signature which is the main entry for the graph is used as quick guide to the relevant nodes, which should be considered for the comparison between source documents and suspected one.

 

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 4, APRIL 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 37

 

This paper is structured as follows. Section 2 introduces the current state of the studies on plagiarism detection. While Section 3 presents an overview of the graph based representation, Section 4 describes the concepts extraction for the sentence. Contribution and comparison methods are reported in Section 5. Finally, Section 6 concludes this paper.

In plagiarism detection, a correct selection of text features in order to discriminate plagiarised from non-plagiarised documents is a key aspect. [2] Has delimited a set of fea¬tures which can be used in order to find plagiarism cases such as changes in the vocabulary, amount of similarity among texts or frequency of words. This type of features has produced diferent approaches to this task. Substan¬tive plagiarism analysis [3] is a different task from pla¬giarism detection with reference. It captures the style across a suspected document in order to find fragments that are plagiarism candidates. This approach saves the cost of the comparison process, but it does not give any hint about the possible source of the potentially plagia¬rised text fragments. In those cases where a reference cor¬pus is considered, the search process has been based on different features. [4] Considers text comparison based on word n-grams. The reference, as well as the suspected text, is split into trigrams, composing two sets which are compared. The amount of common trigrams is considered in order to detect potential plagiarism cases. [5] Considers the sentence as the comparison unit in order to compare local similarity. It differentiates among exact copy of sen-tences, word insertion, word removal and rewording on the basis of a Wordnet-based word expansion process.

some authors [6] define plagiarism as “unacknow¬ledged copying of documents or programs” that can “oc¬cur in many contexts: in industry a company may seek competitive advantage; in academia academics may seek to publish their research in advance of their colleagues.” Most empirical study and analysis has been undertaken by the academic community to deal with student plagiarism, although methods of detection have found their way into the commercial world, e.g. Measuring software reuse and identifying reused code (see, e.g. [7]).

There are several schemes to characterize documents be-fore applying one of the plagiarism detection techniques. Some document descriptors such as Character-based representation, the simplest form, in which documents are represented as a sequence of characters with ignoring spaces between words, periods (full stops) between statements and lines. Also Word-based representation, in which documents are represented as a collection of words with ignoring periods (full stops) between statements and lines. Moreover Phrase-based representation, in which a phrase (part of a statement) is used as a unit of compari¬son. For example, 3-word phrase or so-called trigrams can be used as a comparison unit. and Sentence-based 

 

representation, in which documents are segmented into statements using periods (full stop) as a statement-end indicator. Although Paragraphed-based representa¬tion, in which documents are described as a collection of paragraphs or passages.

There are several techniques have been developed or adapted for plagiarism detection in natural language documents. They can be classified into four main ap-proaches. The first technique is Fingerprint Matching [8][9][10] which involves the process of scanning and examining the fingerprints of two documents in order to detect plagiarism. Then, Clustering [11][10] that uses specific words (or keywords) to find similar clusters between documents Fingerprinting techniques mostly rely on the use of K-grams [12] because the process of fingerprinting divides the document into grams of certain length k. Then, the fingerprints of two documents can be compared in order to detect plagiarism. It can, therefore, be classified fingerprints into three categories: character-based fingerprints, phrase-based fingerprints and statement-based fingerprints. The early fingerprint-ing technique uses sequence of characters to form the fingerprint for the whole document.

Some authors refer about the tools in plagiarism detection [13] which are currently particularly popular and describe their main features in what follows.

www.plagiarism.org

Turnitin: This is a product from iParadigms. It is a web based service. Detection and processing is done remotely. The user uploads the suspected document to the system database. The system creates a complete fingerprint of the document and stores it. Proprietary algorithms are used to query the three main sources: one is the cur¬rent and extensively indexed archive of Internet with approximately 4.5 billion pages, books and journals in the ProQuestTM database; and 10 million documents already submitted to the Turnitin database.

www.urkund.com

Urkund: Another server based plagiarism detection web service which offers an integrated and automated solution for plagiarism detection. It utilizes standard email systems for submission of documents and viewing results. This tool also claims to search through all available online sources giving priority to educational and scandinavian origin. This system claims to process 300 different types of document submissions.

www.copycatchgold.com

Copycatch: A client based tool used to compare lo¬cally available databases of documents. It offers ‘gold’ and ‘campus versions’ , giving comparison capabilities for large number of local r sources. It also offers a web version which extends the capabilities of plagiar 

 

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 4, APRIL 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 38

 

ism detection across the internet using the Goggle API. www.plagiarism.phys.virginia.edu

WCopyfind: An open source tool for detecting words or phrases of defined length within a local repository of documents . The product is being modified to extend searching capabilities across the internet net using the Google API at ACT labs10.

www.canexus.com

Eve2 (Essay Verification Engine): This tool works at the client side and uses it own internet search mechanism to find out about plagiarized contents in a suspected document.

http://www.plagiarism.com

GPSP - Glatt Plagiarism Screening Program: This soft-ware works locally and uses an approach to plagiarism detection that differs from previously mentioned ser-vices. GPSP detection is based on writing styles and pa terns. The author of a suspected submission has to go through a test of filling blank spaces in the writing. The number of correctly filled spaces and the time taken for completion of the test provides the hypothesis of pla-giarism guilt or innocence.

www.cs.berkeley.edu

MOSS - a Measure of Software Similarity: MOSS In¬ternet service “accepts batches of documents and re¬turns a set of HTML pages showing where significant sections of a pair of documents are very similar ” [14]. The service specializes in detecting plagiarism in C, C++, Java, Pascal, Ada, ML, Lisp, or Scheme programs.

www.ipd.uni‐karlsruhe.de

JPlag: Another internet based service which is used to detect similarities among program source codes. Users upload the files to be compared and the system presents a report identifying matches. JPlag does programming lan¬guage syntax and structure aware analysis to find results.

Graph-based method is introduced, designed especial¬ly for web document representation [15]. “ The main advantage of graph-based techniques is that they allow keeping the inherent structural information of the original document. Before describing the graph-based methodol¬ogy, the definition of a graph, subgraph and graph iso¬morphism should be given. A graph G is a 4-tuple: G= (V,E,α,β), where V is a set of nodes (vertices), E ك V×V is a

set of edges connecting the nodes, á : V  v is a

function labeling the nodes, and â : V×V  e is a

function labeling the edges (v and e being the sets of labels that can appear on the nodes and edges, re¬ 

 

spectively). For brevity, we may refer to G as G= (V,

E) by omitting the labeling functions. A graph

G1=(V1,E1,α1,β1) is a subgraph of a graph G2=(V2,E2,α2,β2), denoted G1 ك G2, if V1 ك V2, E1 ك E2 0 (V1 × V1), á1(x) = á2(x) ׊ xאV1 and â1(x, y) = â2(x, y) ׊ (x, y) א E1. Conversely, graph G2 is also called a supergraph of G1. All graph representations proposed in [ 15] are based on the adjacency of terms in an HTML document. Under the standard method[16] each unique term (word) appearing in the document, except for stop words such as “the”, “of”, and “and” which convey little information, becomes a vertex in the graph representing that document. Each node is labeled with the term it represents. Note that we create only a single vertex for each word even if a word appears more than once in the text to build the terms graph in the sentence, and we create also a single vertex for each sentence this vertex involved graph of terms. Also Under the n-distance representation, there is a user-provided parame-ter.Instead of considering only terms immediately follow-ing a given term in a web document. For example[16], if we had the following text on a web page, “AAABBBCCCDDD”, then we would have an edge from term AAA to term BBB labeled with a 1, an edge from term AAA to term CCC labeled 2, and so on. Similar to n-distance, we also have the fourth graph representation, n-simple distance. This is identical to n-distance, but the edges are not labeled, which means we only know that the distance between two connected terms is not more than n. frequency representation model is a type of graph representation too. Each node and edge are labeled with an additional frequency measure. For nodes this indicates how many times the associated term appeared in the web document; for edges, this indicates the number of times the two connected terms appeared adjacent to each other in the specified order”. We discussed the re-presentation of graphs in this paper in section 5.

Concept identification is a common to applications such as ontology learning, glossary extraction and keyword extraction. These applications have different definitions for concept, hence different methods. Pre¬vious methods start from the idea that concepts can be found as word or phrases contained in sentences, which are then divided into smaller phrases in one of two ways: Using grammatical or syntactical information. The former can be found in ontology learning [17], glossary extraction [18] and information retrieval systems [19]. Using a shallow grammar parser, an entire sentence is parsed into a grammatical tree, which classifies sub-phrases as noun or verb phrases, noun phrases are selected as concepts. The syntactical information division of sentences uses punctuation or conjunctions to separate phrases within a sentence, all these phrases are concepts. This approach can be found in keyword extraction systems [20].

The idea that was used in this paper refers to extrac-tion the general concepts from the sentences by any me 

 

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 4, APRIL 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 39

 

 

Fig. 1. Represent sentences in the nodes and extract the concepts from the sentences and the weight between the nodes refer to similari-ty between the sentences, T1, T2,... ,Tn represent the terms.

thod used to extraction. We suggest the same ideas used in the extraction of concepts to be applied in the method that we have proposed.

In this section we discuss the new method for plagiarism detection based on the graph representation. The method relies on a number of steps, first we break down the doc-ument into its constituent sentences. Preprocessing for each document is required such as segmentation of each sentence into separated terms and stop word removal and. the stemming process is applied on the sentence, then we represent the sentences in the form of a nodes related to edge on the order of the sentence within the document. Where each node contains a one sentence of the document. To represent the terms of each sentence as graph we use the method mentioned in section 3. Where each node in the graph contains one term. The node are connected to each other according to order of term posi-tion in the sentence. The whole document consists of a number of nodes determined by the number of sentence in the document. Each node is a graph which represents a sentence. The concepts of node terms are extracted and used for calculating the similarity between each pair of nodes using e.q. (1).

Where Si is sentence 1 and Si+1 is sentence following the sentence1 ,Csi is a number of concepts in sentence I and Csi+1 is a number of concepts in sentence i+1, W is a weight or similarity between si and si+1.

After getting all the concepts of sentences those concepts 

 

grouped in one node called the topic signature, this node inked to each node in the graph and then the similarity between the topic signature and each node is computes separately, We calculate the similarity between the node of the topic signature and the other nodes based on shared concepts using the following equation:

W

Topic. Signature

s

Where Csi is a number of concepts in the sentence I, C is the number of concepts in the topic signature node.

Topic signature node is formed by extracting the concepts of each sentence terms and grouping them in such node. The main advantage of the proposed method is the topic signature which is main entry for the graph is used as quick guide to the relevant nodes. which should be con¬sidered for the comparison between source documents and suspected one. For example, if there is a matching between concept1 on the suspected document and con¬cept 1 in the original document, we go directly to the nodes that containing of concept1 in both original docu¬ment and the suspected document and we ignore all the remaining sentences. In the case of full matching of topic signature of original document with topic signature of suspected document, we will face the problem of the huge number of comparisons which is taken as disadvan¬tage in current methods. To avoid such problem, only we compare the most important nodes. To determine those important nodes we link each node with all the nodes, However, each node has a number of in-links and out-links then compute the similarity of each sentence with the rest of the sentences in the document. Based on the similarities, we extract the highest degree of similarity between the nodes which define the most important nodes and ignore the rest. To calculate the degree of simi¬larity of the node with other nodes using the following equations

c c

W  (  )  si n sk out link si sk

csk

c nc

W ( si sk 

i si sk )

si

Where is a sentence I, Csi is a number of concepts in the sentence I, Csk is a number of concepts in each sentence in the graph.

 

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 4, APRIL 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 40

In this paper we have considered the problem of pla-giarism detection, one of the most publicized forms of text reuse around us today. In particular, we have focused on plagiarism detection using graph based document representation. We have discussed various approaches of plagiarism detection. To date there are few resources which specifically address the task of plagiarism detection.

The proposed method to detect the plagiarism based on graph representation required break down the document into its constituent sentences and the graph building by grouping each sentence terms in one node, the resulted nodes are connected to each other based on order of sen¬tence within the document, all nodes in graph are also connected to top level node” Topic Signature “.  

Fig. 2. Topic Signature formulation for concepts Sentence as Graph

 

Topic signature is a main entry for the graph is used as quick guide to the relevant nodes. Which should be considered for the comparison between source documents and suspected one.

The proposed method based on graph representations contributed by increase the efficiency and reduce the huge a number of matching process.

[1] D. McCabe. Levels of Cheating and Plagiarism Remain High. Center for Academic Integrity, Duke University, 2005. Website: http://academicintegrity.org 

[2] Clough. P.: Plagiarism in Natural and Programming Lan-guages: an Overview of Current Tools and Technologies. Research Memoranda: CS-00-05, Department of Computer Science. University of Sheffield, UK (2000)

[3] Meyer zu Eissen, S., Stein, B.: Intrinsic plagiarism detec-tion. In: Lalmas, M., Mac- Farlane, A., R¨uger, S.M., Tom-bros, A., Tsikrika, T., Yavlinsky, A. (eds.) ECIR 2006. LNCS, vol. 3936, pp. 565–569. Springer, Heidelberg (2006)

[4] Lyon, C., Barrett, R., Malcolm, J.: A Theoretical Basis to the

Automated Detection of Copying Between Texts, and its Practical Implementation in the Ferret Plagiarism and Col¬lusion Detector. In: Plagiarism: Prevention, Practice and Policies Conference, Newcastle, UK (2004)

[5] Kang, N., Gelbukh, A., Han, S.-Y.: PPChecker: Plagiarism pattern checker in document copy detection. In: Sojka, P., Kopeˇcek, I., Pala, K. (eds.) TSD 2006. LNCS (LNAI), vol. 4188, pp. 661–667. Springer, Heidelberg (2006)

[6] Joy, M. and Luck, M. (1999), Plagiarism in Programming Assignments, IEEE Transactions of Education, Vol. 42(2), 129-133.

[7] Hislop, G. W. (1998), Analyzing existing software for soft-ware reuse, Journal of Systems and Software, Vol. 41, 33-40.

[8] Heintze, N. (1996). Scalable document fingerprinting. Pa-per presented at the Second USENIX Workshop on Elec-tronic Commerce.

 

[9] Lyon, C., Malcolm, J. A., & Dickerson, R. G. (2001). Detect¬ing short passages of similar text in large document collec¬tions. Paper presented at the Conference on Empirical Me¬thods in Natural Language Processing.

[10] Yerra, R., & Ng, Y.-K. (2005). ASentence-Based Copy Detec¬tion Approach for Web Documents. In Fuzzy Systems and Knowledge Discovery (pp. 557-570).

[11] Antonio, S., Hong Va, L., & Rynson, W. H. L. (1997). CHECK: a document plagiarism detection system. Paper presented at the Proceedings of the 1997 ACM symposium on applied computing.

[12] Manuel, Z., Marco, F., Massimo, M., & Alessandro, P.

(2006). Plagiarism Detection through Multilevel Text

Comparison. Paper presented at the Second Interna¬tional Conference on Automated Production of Cross Me¬dia Content for Multi-Channel Distribution.

[13] Maurer, H., F. Kappe, B. Zaka. Plagiarism – A Survey. Journal of Universal Computer Sciences, vol. 12, no. 8, pp. 1050 – 1084, 2006.

[14] S. Schleimer, D. S. Wilkerson, and A. Aiken. Winnowing: local algorithms for document fingerprinting. In SIGMOD: Proceedings of the 2003

[15] A. Schenker, H. Bunke, M. Last, and A. Kandel, "Graph-Theoretic Techniques for Web Content Mining", Series in Machine Perception and Artificial Intelligence, 62, World Scientific, 2005.

[16] A.SCHENKER, M. LAST, H. BUNKE3, AND

A.KANDEL,“CLASSIFICATION OF WEB DOCUMENTS USING GRAPH MATCHING”

[17] R. Navigli and P. Velardi, “Learning Domain Ontologies from Document Warehouses and Dedicated Web Sites," Computational Linguistics, vol. 30, pp. 151--179, 2004.

[18] D. Bourigault and C. Jacquemin, "Term extraction + term clustering: An integrated platform for computer-aided terminology," in EACL, 1999.

[19] I. Bichindaritz and S. Akkineni, “Concept Mining for In-dexing Medical Literature," Lecture Notes in Computer Science, vol. 3587, pp. 682--692, 2005.

[20] I. H. Witten, G. W. Paynter, E. Frank, C. Gutwin, and C. G.

 

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 4, APRIL 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/ 41

Nevill-Manning, "KEA: practical automatic key phrase extraction,” in Fourth ACM conference on Digital libraries, 1999.

is a PHD student in Universiti Teknologi Malay-sia. Presently working as a lecturer in the Faculty of Computer Science in International University of Africa(IUA) in Sudan .I have bachelor degree in Computer Science from IUA in 2004. I received master degree in Computer Science from Sudan University of science and technology in 2008. I current research interest includes Information Retrieval, Database, and Data mining.

is an Associate Professor presently working as a Deputy Dean of Research & Postgraduate Studies in the Faculty of Computer Science and Information System in Universiti Teknologi Malaysia. She received her bachelor degree in Computer Science from Universiti Teknologi Malaysia in 1989. She received her master degree in Computer Science from University of West Michigan in 1992. In 2002, she received her Ph.D (Computational Informatics) from University of Sheffield, United Kingdom. Her current research interest includes Information Retrieval, Distributed Database and Chemoinformatic.

received his B.Sc. dgree in

Computer Science from Hadhramout University of Science and

Technology, Yemen in 2000. He received his Master degree

from Universiti Teknologi Malaysia in 2006. He is currently

with Hadhramout University of Science and Technology as lecturer and pursuing Ph.D degree in the Faculty of Computer Science and Information System, Universiti Teknologi Malay¬sia. His current research interest includes Information Retriev¬al, Text Summarization and Soft Co puting.

 

Roman Business Law

Jean-Jacques Aubert, Université de Neuchâtel

Historians of the Roman economy seem to agree that during the period 200 BC – AD 200 the Gross Domestic Product of the overall empire grew, however moderately. They explain this phenomenon partly as the result of a reduction of transportation costs and of ‘transaction costs,’ namely the sum of the costs of looking for opportunities for exchanging goods and services, of reaching agreements between parties through contracts, and of

enforcing such transactions.1 In other words, this reduction was due to, among other causes, the development of a common legal system, ‘especially in the field of commercial

law.’2 This statement begs the question: What is ‘Roman commercial law’? Where does it come from? How did it develop? And how does it fit in the wider field of Roman law? This chapter will attempt to provide some answers.

Law of commerce, commercial law, business law

A standard handbook of Roman law unambiguously states at the outset: ‘Ein

besonderes Handelsrecht haben die Römer daneben nicht ausgebildet.’3 However, there is no dearth of books and articles entitled, or explicitly dealing with, Diritto commerciale romano

or L’histoire du droit commercial romain.4 Behind this apparent contradiction lurk both a question of definition and a recurrent and on-going scholarly debate concerning the way to

approach Roman legal institutions governing trade and other economic activities.5 ‘Commercial law’ (Handelsrecht), with its alleged specificity (‘besonderes’) and independent status (‘daneben’), and its modern scion, ‘business law’ (droit des affaires), will serve as heuristic instruments to evaluate the usefulness, sophistication, and shortcomings of the Roman law of commerce.6

According to a strict definition, ‘commercial law’ is a set of legal rules originating with merchants, designed for merchants, and enforced – partly at least – by merchants. Scholars looking for Roman ‘commercial law’ focus on sources of law, legal interpretation, practical

1 As suggested by the New Institutional Economics (NIE) school of thought, cf. Thomas, cf. North – Thomas (1973); North (1981) and (1991); and Malanima (2009), 7-8.

2 Lo Cascio (2007), 619 and 626 (quotation); Harris (2003), 282 and 285; Silver (2007), 192 (quoting A. Wacke and H.W. Pleket), 210-11; and 217-20; and Terpstra (2008).

3 Kaser (1971) 474 (‘The Romans did not develop a specific, independent commercial law.’)

4 Cf. most recently Cerami – Petrucci (2010). Cf. also Di Porto (1997), citing Carnazza (1891) (non vidi); and Bianchini (1989) and (2007), with some caution. Labruna (1994) coins the term ‘diritto mercantile’. Others speak only of ‘istituti commerciali del/nel diritto romano’: cf. Fadda (1903/1987); Cogliolo (1922); and Földi (2001), 85.

5 The debate started anew in Italy in the late 1980s, cf. Bianchini (1989) and (2007), and Di Porto (1997). It was subsequently carried on at various conferences, e.g. SIDA in Rotterdam (2001), cf. Gerkens (2003); Circolo Toscano ‘Ugo Coli’ at the Certosa di Pontignano (12-14 Jan. 2006); and the Fourth OxREP conference, 1-4 October 2009, cf. Sirks (forthcoming).

6 For surveys in English, cf. Crook (1967), ch. 7, 206-49; and Johnston (1999), ch. 5, 77-111.

 

application, and jurisdiction. The former two are quite accessible through extant sources, the latter two mostly blurred for a lack of them.

Historically, the definition proposed above applies to the law developed in Western Europe, mostly Italy and France, during the Middle Ages. Thus, ‘commercial law’ (ius mercatorum, lex mercatura, Law Merchant) was meant to be more pragmatic and flexible, less bookish, and less dominated by scholars than Roman and Canon laws. Its purpose was to satisfy the needs of commerce, facilitating transactions, expediting proceedings through separate jurisdictions and procedures, and transcending the limits of national legal systems as an early form of international law. Its provisions unavoidably reflect the concerns and interests of an identifiable socio-economic class of people, i.e. traders. ‘Commercial law’ was a part of private law, and its status with regard to civil law has been fluctuating throughout history: distinct at first, it tended to merge and be absorbed by the latter, as is the case in Swiss or Italian law, eventually becoming a subfield of the law of obligations. Alternatively, ‘commercial law’ opened up to other fields of law and became more inclusive in terms of both people ruled by it and types of issues and transactions dealt with it, thus evolving into ‘business law’ (cf. below).7

In the Middle Ages, cases of ‘commercial law’ were heard in special courts. As litigation frequently involved people of various national and social origins, judges based their decision on a mix of mercantile codes and usages, while paying particular attention to good faith. The formalism of legal procedure was somewhat relaxed, and judicial decisions

were rendered on the basis of the consideration of facts rather than legal technicalities.8

Originally, ‘commercial law’ was mostly maritime law, with an emphasis on contracts related to sales, transportation, and money-lending. With time, the nature of trade became more diverse and more complex. ‘Commercial law’ also applied to land-based trade (in the context of fairs and markets), including production, storage, and distribution, and to people involved in any economic activity. A final development even took consumption and consumer protection into consideration. A standard modern treatise of (British) commercial law deals with property, contract (mostly sales and partnerships), agency, payment instruments and systems, financing, insolvency (in connection with company law),

and the resolution of commercial disputes through litigation or arbitration.9

While ‘commercial law’ was exclusively concerned with identifiable (i.e. registered) businesses, and exclusively applied to business transactions performed by well-defined groups of people, traders and professional businessmen, such restrictions came to be seen as counterproductive, and called for adjustment. According to recent trends ‘commercial law’ evolved into ‘business law,’ disregarding the specific status of both structures and people, and combining elements of both private and public law. Legal practitioners thus

7 Merryman (1985), 12-13, 98-100, and 146-47.

8 Goode (2004), 3-6.

9 Goode (2004).

2

 

acquired a transversal competence allowing them to deal with all kinds of issues relating to the multifarious aspects of commercial life.10

Because ‘business law’ is more inclusive than ‘commercial law,’ and because of the reciprocal ‘civilizing’ of commercial law and ‘commercializing’ of civil law which resulted in the development of ‘business law,’ it seems relevant to look at Roman legal institutions, both private and public, in their historical development from the earliest time until the period of classical law in order to evaluate their commercial relevance, usefulness, and adequacy, in comparison with the trappings of both later ‘commercial law’ and ‘business law.’ The following survey will show that while the appellation of Roman ‘commercial law’ is unsustainable, the Roman law of commerce shares many features with modern ‘business law,’ features which extend far beyond the scope and limits of the Roman law of obligations.

Toward a history of Roman business law

‘Buying and selling originate with exchange or barter.’ These are the opening words of the title on sale in Justinian’s Digest, excerpted from Paul’s commentary on the praetorian edict (33 ad ed., D. 18.1.1 pr.). Trade certainly existed before extant sources reveal how Roman law dealt with it. In a pre-monetized society, the exchange of goods and services was assumedly based on barter or exchange (gr. amoibè/lat. permutatio).11 In spite of the

symmetrical feature of barter,12 both parties being equal with regard to the uncertainty about the quality of goods to be exchanged, barter was viewed by classical jurists as an impediment to trade, because of the difficulty to have demand meet supply. To be sure, some commodities (such as cattle, metals, slaves, staples, etc.) were deemed universally desirable and were used as monetary instruments at a very early date. The consensus on a constant medium of exchange eventually took the form of coinage. While relying on bronze bullion (aes rude/grave), the Romans started using Greek coinage by the fifth or fourth century BC, and coined their own by the early third. Barter undoubtedly gave rise to disputes since Rome’s earlier period, but there is no trace of any litigation connected with it. It is possible that barter was not legally recognized before a much later period and that social control was sufficient to settle such disputes. Besides, non-monetary commercial exchanges must have existed throughout Roman history, especially wherever and whenever currency was a scarce commodity, and monetization an unfamiliar abstraction. Even though the geographer Strabo, active in the Augustan period, associates barter with backwardness and uncivilized ways of life typical of marginal, unassimilated tribes,13 it must have been a Roman reality all along.

10 Champaud (1994); Legeais (2003); Lucas (2005); Kelly – Holmes – Hayward (2005). For the combination of private and public laws, cf. Goode (2004), 10; and Lucas (2005), 8.

11 Meylan (1959). Kudlien (2001) rightly points out that the word permutatio has various meanings.

12 Morley (2007) 59, as opposed to sale, where the seller had the advantage of knowing the quality of the goods on sale.

13 Strabo 3.3.7 (about Lusitanian mountain-dwellers, trading through exchange/amoibè or bullion); and 7.5.5 (about Dalmatians). Cf. de Churruca (2001).

 

3

 

Classical Roman jurists knew of the practice of barter and dealt with it rather marginally. In the first century AD Sabinus and Cassius thought of it as equivalent to sale, while Nerva and Proculus disagreed with them (Paul [33 ad ed.] D. 18.1.1.1). The mid-second-century jurist Gaius (Inst. 3.141), following Sabinus, underlines its antiquity by citing Homer (Il. 7.472-475) and reports earlier disputes concerning its contractual status. Around the time of Trajan, Sextus Pedius and Aristo, both cited and followed by Paul a century later ([33 ad ed.] D. 19.4.1.3 and [5 ad Plautium] 19.4.2), address marginal issues by analogy with the consensual contract of sale (emptio-venditio). Paul – the only classical jurist whose works are excerpted in the title of the Digest dealing with barter (D. 19.4, De rerum permutatione) – wonders about the nature of the obligation (re) arising from such transaction, buyer (emptor) and seller (venditor), price (pretium) and good (merx) being undistinguishable from one another, with negative consequences in case of non-delivery or eviction.14 It is only in the late classical period that barter was promoted from the status of ‘unenforceable pactum’ to that of so-called ‘innominate real contract.’ The Justinianic Code preserves several imperial constitutions dated to the mid- and late third century AD on the subject (CI. 4.64.1-8), suggesting that as payments in kind may have increased when the Roman monetary system was in shamble, the need to regulate this type of commercial exchanges may have become more urgent.

The history of barter in Roman law serves as a reminder that all economic transactions were not necessarily sanctioned by law.15 Looking at commerce in general, the first (i.e. both earliest and most prominent) problem to be dealt with is legal recognition and, consequently, jurisdiction.

Ius comm ercii

Among the first policies enacted by the Republican state after the revolution of 509 BC, an important step consisted in establishing and defining (commercial) contacts with neighbors, both immediate (Latins and other Italic people in the region) and farther ones (Etruscans, Greeks, and Carthaginians, to name only the most important ones). Polybius (3.22-27) records the content of a series of treaties between Romans and Carthaginians. In the first treaty (ca. 509-507), it is agreed upon that trade carried out by the Romans in Sardinia, Sicily, and Africa – all Carthaginian territories at the time – should be strictly controlled and guaranteed by the state. Transactions must be concluded through an auctioneer (kèrux) and a scribe (grammateus), both of whom engage the Carthaginian state’s good faith (pistis) toward sellers (3.22). Reciprocity is not mentioned. This provision sounds like a protection against piracy, ransoming, or extortion, as the distinction between such practices and trade is sometimes blurred. The second treaty (ca. 348, cf. Liv. 7.27) explicitly forbids piracy and opens up trade in Sicily and Africa for the Romans and at Rome for the Carthaginians, where traders of each nation had the same rights as the natives (3.24). This means that the exchange of goods between Romans and Carthaginians had the same legal

14 D. 19.4.1-2 (with Mommsen’s correction of the provenance). Cf. also id. (ibid.) D. 18.1.1 pr.-1; and Ulp. (1 ad ed. aed. cur.) D. 21.1.19.5, possibly itp. Cf. Zimmermann (1990), 250-52; 532-37; Johnston (1999), 78-79.

15 Aubert (2007b), for such a situation in fourth-century Athens.

 

4

 

validity in either place as between fellow countrymen. A third (or fourth?, Liv. 9.43) treaty, dated ca. 278, reasserts these dispositions. Subsequent treaties, while redefining areas of respective power and influence, do not question the basic trade agreement, the text of which, interestingly, was engraved on bronze tablets and preserved in the aediles’ office (3.26).16

Whereas Polybius does not enter into detail about the legal aspects of such arrangements for international trade, Dionysius of Halicarnassus reports (6.95), in the context of the so-called foedus Cassianum, a treaty concluded ca. 493 with neighboring Latin communities, that contracts between Romans and Latins would be enforced in courts, within ten days, wherever they had been concluded. This provision implies that in any court of law Romans and Latins would enjoy an identical legal standing, with equal protection from the law recognized and enforced by the respective courts. This would have excluded the potential contradictions attached to competing legal systems and inaugurated a form of international law (ius gentium).17 Accessorily, plaintiffs are guaranteed a speedy trial, facilitated by the fact that discrepancies between legal systems should be irrelevant. This arrangement will be known subsequently to the Romans as commercium/ius commercii (Tit. Ulp. 19.5), namely the right to make formal contracts, to acquire property, and to resort to courts, according to Roman law and procedure.18 It was, or became, part of a larger package (isopoliteia) including the right to intermarry (conubium) and to participate in civic life (suffragium). This interpretation is based on the terms of the settlement of 338 BC, whereby Latins would be deprived of various rights they previously enjoyed (Liv. 8.14.10). It is quite possible that commercium allowing Latins to benefit from the protection of Roman law was more than the rights secured through the Romano-Carthaginian treaties.19 It is also likely that the scope of commercium was limited in comparison with Roman citizenship and did not come close to extending to those privileged foreigners (peregrini) a legal protection equal to that enjoyed by Roman citizens.20

The issue of legal recognition of commercial transactions concluded with foreigners was taken up in the first Roman ‘codification’ of law known as the Law of the Twelve Tables, dated ca. 451-450 BC. Foreigners (hostes, cf. Cic., Off. 1.37), assumedly with commercium, can have their day in court (2.2, Crawford), possibly with some degree of priority over other cases with respect to international treaties,21 but transfer of ownership cannot occur through long-term prescription (usucapio), as the retention of eminent title (auctoritas) with regards to, or against the claim of, foreigners is not limited to one or two

16 Ferenczy (1969); Scardigli (1991); Nörr (2005), esp. 171-77, where the author sees in Pol. 3.22.8 a reference to two common forms of sales (auctioneering and written contract), and suggests a possible interpretation for the term telos (auctoritas) as the effect of the contract.

17 Nörr (2005) 183-84, with reference to Wolff (1979).

18 Mayer-Maly (2003); Nörr (2005); and Minaud (2011), ch. 1 (§§ 19-70, esp. 35-36).

19 Capogrossi Colognesi (1994); Kremer (2005).

20 Kaser (1953), suggesting that mancipatio, not in iure cessio, was made available to foreigners as a formal way to convey property.

21 Kremer (2005), 197-203.

 

5

 

years (6.3), but should be everlasting (6.4).22 This means that foreigners can only acquire ownership through formal conveyance, such as mancipatio (Tit. Ulp. 19.4), and that the seller/transferor would have to protect such a buyer/transferee against eviction by a third party for an unlimited period of time.

In this regard, the Law of the Twelve Tables contains some dispositions offering potential for innovation. Take, for example, the law of contracts (6.1): ‘When someone shall perform a nexum or a mancipatio, the ius will be defined by what the tongue has pronounced.’ Nexum is an early form of loan (Varro, LL 7.105), performed, like mancipatio, ‘by means of bronze and scale’

(per aes et libram) and guaranteed by the pledge of the very person of the debtor. Regarded as unduly risky and anti-social in its consequences, it was abolished in the late fourth century by a lex Poetelia Papiria (326 or 313 BC). Mancipatio proved a more durable institution. The provision of the Twelve Tables introduces a verbal dimension to the formal act, allowing the parties to specify the terms of the contract to be concluded. Originally the solemn utterance before witnesses called nuncupatio may have been more or less fixed, prescribed words being imposed on the parties. The sheer fact that the same ritual per aes et libram was performed in widely different contexts, such as the making of a will or a donation, the conveyance of property, the constitution of a dowry or servitude, the emancipation of a dependent, or the contracting of a loan, points to a large spectrum of required statements. In pre-classical Roman law, contractual obligations became overwhelmingly verbal, the so-called stipulatio (and its variants, sponsio, promissio, cautio, etc.) being so flexible – and liable to become increasingly so – as to adequately address most social and economic needs. Nuncupatio and stipulatio share the faculty of clarifying intentions in any legal situation. In both cases only one party’s intention is clarified.

It is remarkable that the Twelve Tables, for all of their preserved or reconstructed provisions (between 88 and 109), have little to say about commerce.23 Provisions 3.5 and 3.6 incidentally allude to periodic market-days (nundinae). Some other clauses are pregnant with important features of later legal developments in the law of commerce, such as the civil liability of masters for the (wrong)doings of their dependents (8.2 and 12.2, noxa), deceit (8.10, fraus), and malice aforethought (8.9, dolus malus). Much of the law, however, is concerned with criminal law, police regulation, and civil procedure. This latter field, with its reliance on the role of magistrates and despite its formalism, would prove instrumental in the development of much of the Roman law of commerce over the next three centuries or so (mid-fifth to late-second c. BC).

In the archaic and mid-republican periods, civil litigation falls within the scope of one of five actions of the law (legis actiones), or general remedies granted by a magistrate endowed with jurisdiction based on imperium. These remedies ranged from the taking of a pledge (pignoris capio) or personal execution (manus iniectio) to the initiating of a judicial procedure in front of the magistrate (iudicis arbitrive postulatio), on the basis of a specific

22 Pace Kremer (2005), 203-06; and Humbert (2005), 393-97.

23 Aubert (2004), 164-65. As W.V. Harris suggested to me, those archaic rules may have been completely superseded by the time the law found its way into our extant literary sources.

 

6

 

claim (condictio), which was at times stated under oath (sacramentum). Plaintiffs had to resort to one of these remedies (actiones) according to the nature of each claim. If none of these remedies fitted the case, there was no claim. Some remedies were more flexible than others. However, because of its excessive formalism, this archaic system of civil procedure underwent a natural evolution while keeping some of its key features: the two-step procedure, first in front of the magistrate (in iure), then before a judge (apud iudicem); and the turning point of the joinder of issue (litis contestatio), whereby the parties agreed on the legal framework, sanctioned by the magistrate, within which the appointed judge(s) would have to evaluate the facts. This legal framework was eventually described in a written formula instructing the judge(s) about the path to follow and leading to two opposite outcomes of the forthcoming trial (‘if it occurs that ...., then condemn; if not, then absolve’). Whatever happened thereafter, the joinder of issue extinguished the plaintiff’s claim.24

Commercial transactions could often, but not always, be enforced through the existing set of original remedies. Some situations however called for new solutions. Thus, additional remedies came to be created over the next centuries by statutes (e lege), such as the lex Marcia against usury in 104 BC (Gai., Inst. 4.23), or by some magistrate’s edicts. Therefore, remedies were divided between actiones civiles and honorariae (more specifically, actiones aediliciae or praetoriae). As a result, remedies available to parties tended to mushroom, since magistrates were free to deliver – or not – a legal remedy upon one party’s request (petitio actionis). Since the edict was valid for the duration of the magistrate’s tenure of office, an existing actio could technically be denied (denegatio actionis), although social pressure may have played a part in the quest for consistency. Adventurous magistrates would be expected to create new actiones if and when they considered that the situation and/or their sense of equity allowed them to do so. Creativity could take various forms. When a remedy existed (actio directa) but did not exactly match the situation, the magistrate could make adjustment by issuing an actio utilis that extended the scope of the original remedy. If the unsatisfactory existing remedy was an actio civilis (i.e. based on statute), the magistrate could introduce a fictitious element into its intentio to have it fit the situation, such as Roman citizen’s status for one of the parties even though he was an alien (peregrinus). The adjusted remedy would then be regarded as an actio ficticia. If Roman law provided no previously existing remedy to address the issue even remotely, the magistrate could create one based on his perception of the situation (actio in factum). Faced with an actual legal problem, the magistrate was allowed – and expected – to devise a legal solution, leaving it to the judge to decide whether the facts that had led to the solution were correct. As will be seen later on, this instrument proved to be most efficient in dealing with issues related to commercial life.25

It is obvious that such a system, introduced between the fourth and second century BC and called ordo or formulary system, gave great power to magistrates in charge of delivering actiones. To alleviate the suspicion of arbitrariness and to give a sense of coherence to the administration of justice, magistrates with iurisdictio were required to

24 Gaius, Inst. 11-30; and Borkowski – Du Plessis (2005), 63-83.

25 Gai., Inst. 4.30-47; Gaudemet (1982), 615-20.

 

7

 

announce ahead of time when and in what circumstances they would grant a remedy (actio). Magistrates did so through their (yearly) edict. We know next to nothing about the circumstances in which they devised it, but we can imagine that outside inspiration or pressure may not have been totally foreign to their decisions. Whose pressure remains debatable, but jurists and professionals involved in all kinds of business transactions

unavoidably come to mind (e.g. Cic., 2Verr. 1.119).26 Edictal law should be looked at as the ad hoc answer to foreseen or encountered legal problems and situations. In this sense, it can be said that the Roman law of commerce, at least for its private law components, was essentially a law originating with traders and adapted to the requirements of traders, though not exclusively them. The ius honorarium introduced by some magistrates in order to abet, supplement, or correct the existing (civil) law was geared toward protecting and promoting the interests of the community.27 Economic interests must have ranked high on the list. Unfortunately, the details mostly escape us.

During the fifth and early fourth centuries BC, justice was administered by the consuls. Admittedly, plebeian aediles, two in number, may have had some jurisdiction, the extent of which is unclear. In 367 BC, if not before, new magistracies reserved to patricians were created: both the praetorship and the curule aedileship were devised to compensate for the loss of power resulting from sharing the consulship with plebeians. The new magistracies distracted some of the consuls’ powers, thus allowing patricians to retain control over them. As holders of potestas, including the right to issue edicts (ius edicendi) and to enforce their authority (coercitio), and imperium, the basis of their judicial power (iurisdictio),28 these new magistrates, with the addition of more praetors in and after 241 BC who dealt with foreigners (peregrini) and took charge of provincial governorships, were responsible for legal matters and for the supervision of markets. Their part in the development and implementation of the Roman law of commerce is attested by what remains of the edicts they promulgated over several centuries. I would contend that Roman business law is first and foremost edictal law. However, individual edicts were necessarily phrased as briefly as possible and therefore left much room for interpretation. This is where the jurists stepped in, and edictal law must be approached through the juristic writings of the classical period, often the very source from which edictal law can be reconstructed. In the next sections, both sources of law will be jointly examined in order to assess the making and refining of Roman business law. One should remember however that several centuries may have elapsed from the time a remedy was created until the time the legal texts commenting on it were written, before finding their way into the Digest.

Aedilician, praetorian, and gubernatorial edicts

a) edictum aedilium curulium

26 The evidence is clearer for a later period, cf. Nov. Iust. 136 (AD 535); 106 (540) and 110 (541); Ed. Iust. 7 (542) and 9 (date unknown) for the influence of moneylenders on imperial legislation. Cf. Jones (1964), 350 and 1139, n. 63.

27 Papinianus (2 definitionum) D. 1.1.7 and Marcianus (1 inst.) D. 1.1.8.

28 The question of the curule aediles’ imperium is debated, cf. Impallomeni (1955), 109-21; and Reduzzi Merola (2001), 325, n. 18.

 

8

 

In their capacity of city magistrates, the curule aediles were in charge of supervising markets. What remains of their edicts (FIRA I2 66) goes back at least to the first half of the second century BC and is known through a couple of citations preserved in the Digest (D. 21.1, De aedilicio edicto et redhibitione et quanti minoris) and in the work of the second-century

AD antiquarian Aulus Gellius (Noctes Atticae 4.2.1).29 These citations can be divided into two subjects: the sale of slaves (mancipia) and the sale of animals (iumenta, ferae). The edict stipulates a) that buyers of defective slaves will be granted a remedy for rescission (actio redhibitoria) if the seller has concealed a specific defect, whether or not he was aware of it; and b) that a remedy for diminution of the price (actio aestimatoria or quanti minoris) would be available to buyers who had been cheated of what was owed to them, including advertised qualities that turn out to be lacking. A claim of the former kind was valid for sixty days or six months, of the latter for a whole year.30

The aedilician edict was traditional (tralaticium): aediles were in charge during one year only and they usually did not have the legal training necessary to be innovative. Like their better-known senior colleagues, the praetors, aediles had the tendency to borrow most or all of their edicts from their predecessors. This habit resulted with time into a rather static document, eventually codified by Salvius Iulianus in the age of Hadrian, perhaps as an appendix to the praetorian edict (edictum perpetuum). We do not know when the aedilician edict stopped being modified, but some first-century AD documentary evidence suggests that it was still viewed as dynamic under Nero, if not later.31

The content of the aedilician edict is mostly lost, perhaps due to its eventual merging with the praetorian edict. However, the reconstructed text gives several examples of what a defective slave may be. The aediles are said to have done everything possible to avoid ambiguity (Ulp. [1 ad ed. aed. cur.] D. 21.1.1.7), but there was ample room for elaboration. For all its flexibility and pragmatism, edictal law would be inadequate without the jurists’ interpretation. Faced with provisions phrased briefly and with a rather general scope in mind, the jurists took on themselves, over several centuries, the task to adjust these provisions to the requirements of social and economic life, within a logical framework. Title 21.1 of the Digest includes sixty-two excerpts from classical juristic writings on the subject. The earliest authority quoted in them is Cato (presumably the Elder, in [ibid.] D. 21.1.10.1), followed by several Republican jurists. It provides precious evidence about ancient slavery and the slave trade. The late Republican jurist C. Trebatius Testa, for instance, downplayed bad breath as the result of poor oral hygiene (D. 21.1.12.4 [ibid.]), a condition that Apuleius (Apol. 6) could have treated adequately two centuries later. The Flavian jurist Sextus Pedius discusses the case of the bed wetter, and distinguishes between slaves suffering from a bladder condition and those who are too drunk or too lazy to get up at night (D. 21.1.14.4, cited with approval by Ulpian [ibid.]). Mental addictions are taken

29 Impallomeni (1955), 90-136; Pugsley (1974).

30 Gaius (1 ad ed. aed. cur.) D. 21.1.28; Ulp. (1 ad ed. aed. cur.) D. 21.1.19.6 and (2 ad ed. aed. cur.) D. 21.1.38 pr. (about iumenta) .

31 Reduzzi-Merola (2001), based on TPSulp. 43 (21 Aug. 38) (possibly also TPSulp 42 and 44); TH 59-62 (between AD 47 and 63); and Petronius, Sat. 53.9-10. Cf. de la Hoz Montoya (2008), 376-80.

 

9

 

seriously: slaves hooked on games or art work are considered defective by the late second-century AD jurist Venuleius Saturninus ([5 act.] D. 21.1.65 pr.). At stake were the smoothness of the slave trade, consumer protection, and the expected productivity of slave labor.

The animal trade presented similar problems, but seemingly triggered less discussion on the part of the jurists (D. 21.1.38, 40, and 41): dangerous animals, including dogs, should be kept away or chained, so that they do not attack people, thus causing damages calling for monetary penalties. Animals should be sold with the trappings (ornamenta) they wore at the time of sale, lest the sale be rescinded or a diminution of the price be granted. Interestingly, the aediles consider sales (and returns) in bulk, for instance for a pair of mule. This is valid for slaves as well, be it a company of actors or, merely, siblings (Ulp. [ibid.] D. 21.1.38.1 and 14; Paul [1 ad ed. aed. cur.] D. 21.1.39).

Diminution of the price could be obtained through the actio quanti minoris/aestimatoria, which the aediles mentioned only in connection with animals (D. 21.1.38 pr. and 13), but which the jurists extended to slaves.32 This is typical of the work of classical jurists. We know of provisions dealing with companies of slave dealers (societates venaliciariorum, Paul [2 ad ed. aed. cur.] D. 21.1.44.1), the castration of young slaves (Ulp. [18 ad ed.] D. 9.2.27.28), or the definition of trappings (ornamenta, Paul [2 ad ed. aed. cur.] D. 50.16.74). Consumer protection, however, was not limited to the sale of slaves or animals (at least by Diocletian’s time, CI. 4.58.4 [286-293]). Besides, Ulpian points out that the aedilician edict applies to sales only, not to other categories of contracts such as hire and lease (locatio conductio), because such contracts were never under aedilician jurisdiction or – and the introduction of an alternate explanation would be telling if the whole passage were not interpolated – because both contracts are different (‘non similiter,’ Ulp. [1 ad ed. aed. cur.] D. 21.1.63).

Even though aediles were traditionally in charge of supervising local markets and their edicts were recognized as valid not only in Rome, but also in the rest of Italy and apparently across the Empire by the second century AD,33 the scope of aedilician law was dwarfed by praetorian law. This is confirmed by the fact that only few classical jurists are known to have commented on the aedilician edict: Ofilius and possibly Labeo in the late Republican and Augustan periods, Caelius Sabinus (cos. 69) and Sextus Pedius in the first century AD, Pomponius and Gaius in the second, Paulus and Ulpian in the early third. It is likely that the aedilician edict eventually became an appendix to the praetorian edict, since

32 Gaius (1 ad ed. aed. cur.) D. 21.1.18 (false advertisement); Ulp. (1 ad ed. aed. cur.) 21.1.19.6 (respective deadlines, six months for a. redhibitoria, one year for a. quanti minoris); (ibid.) 21.1.31.5 (citing Pomponiius, collective purchase), 10 (idem), and 16 (successive claims by buyer); Pomponius (23 ad Sab.) 21.1.36 (about bulk price); Paul (1 ad ed. aed. cur.) 21.1.43.6 (about the complementariness of both remedies); Paul (11 ad Sab.) 21.1.47 pr. (extinction of the buyer’s claim after the slave’s manumission); and Ulp. (80 ad ed.) 21.1.61 (for an undeclared servitude).

33 Cf. above n. 27, and FIRA III2 87-88 (AD 139 et 142, Dacia); FIRA III2 133 (AD 151, AD 151) and 132 (Seleucia in Pieria, AD 166).

 

10

 

the curule aediles’ competences were progressively absorbed by other magistrates and imperial officials.34

b) edictum praetoris

Both urban and peregrine praetors issued edicts, general rules and specific remedies, that were valid and possibly binding during their – one-year – term in office, at the end of which individual edicts could be either dropped or renewed by transfer into the successor’s edict.35 Dio Cassius reports that in 67 BC a plebeian tribune named C. Cornelius introduced a plebiscite (Lex Cornelia de edictis/de iurisdictione praetoris) compelling all praetors – possibly governors as well – to abide by their own edicts, which contained the basic principles (dikaia) according to which they would administer justice, but not all remedies (dikaiomata) which were required in order to enforce contracts. This piece of legislation was part of a larger package aimed at curbing corruption on the part of the senatorial class and ensuring legal consistency. It may have reenacted an earlier custom or law that had been neglected or broken during the 80s and 70s BC.36

The making of praetorian edicts during the Republic and early Empire is somewhat of

a mystery.37 Of all preserved or reconstructed edicts, none can be dated precisely, and few only roughly. Vague termini ante quos are provided by quotations, mentions, or allusions in the commentaries to the edict by Republican and Augustan jurists, such as Servius Sulpicius Rufus, Ofilius, or Labeo. The sum of those edicts that had been made permanent over the years was codified ca. AD 130 into the Edictum Perpetuum, again by Salvius Iulianus. What remains of it (FIRA I2 65) is known as the edict of the urban praetor (EP) and has been reconstructed by O. Lenel in the nineteenth century on the basis of the organization of Justinian’s Code and Digest and the numerous quotations preserved in the latter, especially excerpted from large commentaries by Paul (80 books), and Ulpian (81 books). What is left is tantamount to 292 entries (rubricae) distributed in 45 titles (tituli) and five parts (partes). For many of the 292 entries, nothing but the title is preserved or can be reconstructed. The listing is not necessarily representative of the relative chronology of their introduction into the edict. Actually, there are reasons to believe that the arrangement was revisited even shortly before, if not at the time of, its final codification. On the basis of a comparison of the space devoted to various parts of the Edictum Perpetuum in Paul’s and Ulpian’s commentaries as opposed to Sextus Pedius’ commentary on the first-century-AD edict, it appears that some dispositions regarding business law may have been shifted to a different section of the edict. Thus, sometimes in the late first or early second century AD, possibly at the time of the codification of the Edictum Perpetuum by Salvius Iulianus, the law of

34 Dio 53.2.2 and 54.2.3. Cf. Giachi (2005), 65-70, esp. 67, n. 147.

35 Brennan 2000. On the peregrine praetor, cf. Daube (1951); and Serrao (1954).

36 Dio Cassius 36.40.1-2; Asconius, Ad Cic. Pro Cornelio de maiestate 59.8-9 (Clark) = 48 (Stangl); and Cic., 2Verr. 1.46.119 about Verres deciding against his own edict. Cf. Rotondi (1912/1990), 371; Griffin (1973), 209; Pinna Parpaglia (1987) and (1992); and Palazzolo (1991).

37 Guarino (1980), esp. 68-76 for the distinction between edictum perpetuum (promulgated at the beginning to the year of office) and edictum repentinum (promulgated during the year of office), as opposed to denegatio actionis (denial of a remedy); and Mantovani (2000).

11

 

indirect agency (cf. below) seems to have been severed from its original context, i.e. the special liability of seamen, innkeepers, and stablekeepers (= managers of relay-stations) for what has been entrusted to them in connection with the practice of their trade, to be eventually linked with banking and financing on the one hand, and with consensual (good faith) contracts on the other.38

The urban praetor’s edict contributes a lot to our knowledge of Roman business law, as nine titles out of forty-five, and more than forty entries out of 292, deal with legal issues concerning commerce.39 Unsurprisingly, this means maritime law, banking, agency, contracts among private individuals, and between private individuals or companies and the Roman government, securities, and procedure. The question is whether any of the edicts concern traders to the exclusion of other actors participating in legal transactions. The answer is unambiguously positive, although such cases are rather limited.

EP 2.19.106-112 introduced good-faith remedies (De bonae fidei iudiciis) that are central to economic activities and will provide a starting point. Deposit (D. 16.3) as a real contract, trust (Gai., Inst. 2.60) or fiduciary agreement as a form of real security, mandate (D. 17.1), partnership (D. 17.2), sale (D. 19.1), and hire (19.2) as consensual contracts, were all devised during the mid-Republican period, and became standard legal instruments. They advantageously supplemented the older, flexible though formal contract by (verbal) stipulatio in that they considered the will (voluntas), permission (patientia), awareness (scientia), or lack of it (ignorantia) of the parties to a contract. By allowing transactions to be carried out despite the absence of one or both parties, they certainly opened the door to a major innovation in Roman business law, the law of indirect agency (cf. below). The consensual contract of sale (emptio venditio) was introduced by the second century BC. At the minimum, buyer and seller had to reach a specific agreement on both the object of sale and its price. The usual terms of the contract could be modified or specified through stipulatio or pactum. In practice, the parties’ agreement was often made explicit – and strengthened – by the written record (chirographum) of what sounds like a stipulatio. The seller had to warrant for his title to the object of sale and for its quality (i.e. lack of defects). The buyer, who originally bore all the risks of the transaction (“caveat emptor”), came to enjoy the protection of the law, enforced by aediles, praetors, and later on prefects, in accordance with established standards of good faith.40 The Roman law of sale, for all its sophistication and prominent importance in commercial life, was not specific to the business

38 Giachi (2001), 14, n. 23, about the transfer of EP 2.18.101-105 (Quod cum magistro navis, institore eove qui in aliena potestate est, negotium gestum erit) from its hypothetic, original place just after EP 2.15.78 (De his quae cuiusque in bonis sunt – In factum adversus nautas caupones stabularios) toward a place between EP 2.17.95-100 (De rebus creditis) and EP 2.19.106-112 (De bonae fidei iudiciis).

39 EP 1.3.9 (De edendo – argentariae mensae exercitores); EP 1.11.49-50 (De receptis); EP 2.15.78 (De his quae cuiusque in bonis sunt – In factum adversus nautas caupones stabularios); EP 2.17.95-100 (De rebus creditis); EP 2.18.101-105 (Quod cum magistro navis, institore eove qui in aliena potestate est, negotium gestum erit); EP 2.19.106-112 (De bonae fidei iudiciis); EP 3.32.183-185 (De publicanis); EP 4.40.218-223 (Quemadmodum a bonorum emptore vel contra eum agatur); EP 5.44.269-279 (De exceptionibus).

40 TPSulp. 42 (Puteoli, AD 26); 43 (Puteoli, AD 38); and 44 (Volturnum, first c. AD); and FIRA III2 132-142 (all documents of early and late imperial date). On sale, cf. Gai., Inst. 3.139-141; and D. 18.1-19.1; Crook (1967), 215-21; Johnston (1999), 79-84; and Zimmermann (1990), 230-337.

 

12

 

community.41 Other areas of law, such as transportation or agency are more likely to have been the preserve of merchants.

The law of carriage, for both land- or sea-transport of people and goods, combines several areas of law, including the consensual contract of hire (locatio conductio, D. 19.2; C. 4.65), whose object could be either the means of transportation (res in the form of pack- or draft-animals, mounts, porters, wagons, ships, storage room), the task (opus) or the services provided by professionals (operae); specific modes of financing (maritime loans or faenus nauticum, D. 22.2; CI. 4.33) combined with early forms of insurance against the enormous risks connected with navigation (lex Rhodia de iactu, D. 14.2) (cf. below); and safekeeping (custodia) in storage (horrea) or in transit (naves, cauponiae, stabula, D. 4.9 and 47.5). EP 1.11.49 (De receptis) and 2.15.78 (In factum adversus nautas caupones stabularios) call for a higher standard of liability on the part of shippers and inn-/stablekeepers for the goods entrusted to their care in the context of their professional activities. The operators (exercitores) of the ship, inn, or stable are responsible for the wrongdoings (delicta, including theft/furtum) of their employees, slaves or free, because they chose their staff and had a chance to vet them. By analogy, innkeepers are also responsible for the wrongdoings of their guests, at least if they live there on a permanent basis, unlike passing travelers and passengers on a ship (Ulp. [38 ad ed.] D. 47.5.1.6).42 The owner of the stolen good(s) can sue either the thief iure civili or the operator honorario iure (ibid. D. 47.5.1.3). If however the operator has guaranteed his safekeeping of the goods, he can sue the thief himself (D. 47.5.1.4).

The duty of safekeeping was conditioned by the operators’ free choice (arbitrium) in accepting the goods to be watched. This is what distinguishes traders from thieves: the former, unlike the latter, must show good faith (fides) and refrain from fraud. In that sense, the praetor’s edict is considered “most useful” by Ulpian ([14 ad ed.] D. 4.9.1.1). The phrasing is however most economical, and the terminology is to be elucidated by jurists: ‘seaman’ (nauta) designates the operator of a ship (exercitor) or his agent (magister navis), not the crew. Along the same line, the ‘inn-’ or ‘stablekeeper’ (caupo and stabularius) is the person in charge of the facilities, either as operator (exercitor) or as manager (institor) (ibid. D. 4.9.1.2-3 and 5); ‘ship’ is understood as sealiner, freighter, river-boat, or raft (Labeo, cited ibid. D. 4.9.1.4); ‘goods’ (res, merces) mean not only merchandises (mercedes), but also personal belongings transported as luggage or clothing (Vivianus and Pomponius, ibid. D. 4.9.1.6-8); etc. The praetor provides a general ruling, the jurists determine its scope, and the judge sees to its application.

In principle, the remedy brought against the operator/manager is based on the initial contract of hire and lease or deposit that binds him to the plaintiff, and consequently calls for different standards of liability, fault (culpa) in the former case, fraud (dolus) in the latter (Ulp. [14 ad ed.] D. 4.9.3.1). Cases not covered by the edict, for instance for lack of fault or fraud on the part of the operator or because no price was paid for the service, gave rise to

41 With the possible exception of the sale of wine, cf. Frier (1983), 292.

42 A. Petrucci kindly pointed out to me that Gaius ([5 ad ed. prov.] D. 4.9.2) includes the innkeeper’s liability for theft committed by travelers (viatores) in case of receptum. Ulpian follows suit for passengers (vectores) in similar condition ([14 ad ed.] D. 4.9.1.8).

 

13

 

an actio in factum (ibid. and Paul [22 ad ed.] D. 4.9.6). If the damage was caused by one of the sailors’ slave who was not a sailor himself, the operator would nevertheless be liable to an actio utilis (Ulp. [18 ad ed.] D. 4.9.7.3). Conversely, operators could avail themselves of special legal protection (exceptio) in case of an act of God (vis maior), such as shipwreck or pirate/bandit attack (Labeo, cited by Ulp. [14 ad ed.] D. 4.9.3.1). Thus the law took into account the reality of commercial life by striking a balance between the interests of customers and those of business people.43

The praetorian edict includes two dispositions concerning professional bankers (argentarii, argentariae mensae exercitores). The so-called receptum argentarii/argentariorum (EP 1.11.50) was, like the receptum nautarum cauponum stabulariorum discussed above, an informal promise or guarantee (pactum praetorium), in this case, to pay a client’s debt on an agreed day. Surprisingly, this arrangement binds banker and creditor, and leaves out the customer/debtor. In addition, the nature and very existence of the debt are irrelevant. Despite its obvious usefulness for banking, the receptum argentariorum is little attested in the legal sources, because its was merged in late Antiquity with the wider ranging, less specific constitutum debiti.44

The money that the banker agrees to pay probably comes from the customer’s account. The praetor rightly compels bankers (argentarii, argentariae mensae exercitores) to produce accounts upon request from judicial authorities (EP 1.3.9a and b, De edendo). The reasoning behind the edict is considered most fair by Ulpian ([4 ad ed.] D. 2.13.4 pr.-1). The praetor arranged for reciprocity, the banker being entitled to ask for the production of accounts on the part of an opponent, unless the banker had the means to achieve the same result on the basis of documents readily accessible to him because of his occupation (ibid.

D. 2.13.6.8-9).45 Gaius, in the first book of his commentary to the provincial edict (D. 2.13.10.1), explains that bankers (argentarii) have a special obligation to produce their accounts because their trade has a publica causa, which means that the Roman people had a vested interest in regulating the profession.46 For that reason women were banned from it

according to the third-century jurist Callistratus.47

EP 2.17.95-100 (De rebus creditis) deal with various forms of loans, for consumption (mutuum, D. 12.1; CI. 4.1-2) or for use (commodatum, D. 13.6; 4.23), pledge (pignus, D. 13.7;

43 Huvelin (1929), 115-59; Zimmermann (1990), 514-26 (receptum nautarum cauponum stabulariorum); Gröschler (2002), 70-79.

44 D. 13.5 and CI. 4.18 (De pecunia constituta). Cf. Crook (1967), 232-33 and 243; Zimmermann (1990), 511-14; Andreau (1987), 597-602; id. (1999), 43-44 and 58, with reference to TP 151 (= FIRA III2 131e = CIL IV 3340.91, AD 62) as a possible specimen from legal practice; Petrucci (1991), 378-83; id. (2002), 57-65; and id., in Cerami et al. (2004), 129-35.

45 EP 1.3.9 (De edendo – argentariae mensae exercitores); cf. Andreau (1987), 551 and passim; Andreau (1999), 30-49, esp. 46; Petrucci (1991), 141-71; id. (2002), 23-27 and 140-53; id., in Cerami et al. (2004), 175-86.

46 Petrucci (2002), 18 and 123, n. 30. On causa as a ground for legal action, cf. Borkowski – du Plessis (2005), 258-59, with reference to Ulp. (4 ad ed.) D. 2.14.7.4.

47 ‘Opera virilis’: Callistratus (1 edicti monitorii) D. 2.13.12. Cf. Andreau (1987), 497. The nature of the edictum monitorium is uncertain.

 

14

 

CI. 4.24), and setoff (compensatio, D. 16.2; CI. 4.31). There is no questioning the usefulness of these legal institutions for commercial life, in terms of credit, security, and payment, but were by no means restricted to the activities of professional traders. It is however tempting to explain the sophistication and sometimes paradoxical and adventurous nature of some arrangement as dictated by the requirement of specialized trading.48 It is not entirely surprising that deposit (depositum) is dealt with in what could have been the very next entry of the edict (EP 2.19.106), before the shift of the title dealing with agency assumedly occurred.49

Agency is undoubtedly one of the most significant areas of progress in the field of Roman business law.50 As a matter of principle Roman law was adverse to the idea that a person’s action could engage another’s liability, while there was no problem with benefiting from it. This somewhat contradictory position precludes the concept of agency, so important in economic life, even more so given the negative attitude of the Roman elite towards trade and commerce (Cic., Off. 1.150-151). The patriarchal nature of Roman society offered a way to overcome this obstacle: persons in power (in potestate, alieni iuris), such as sons, daughters, other descendants, and mostly slaves had no legal capacity of their own, and therefore participated in that of their pater familias/dominus. Slaves’ wrongdoings (delicta) originally gave rise to noxal liability on the part of their master.51 The praetor, aware of this asymetrical state of affairs, had only to extend the liability of the principal to the contracts of his agent.52 To that effect a set of six remedies was created over the course of time, presumably between the late third and early first centuries BC: EP 2.18.100¬105 (Quod cum magistro navis institore eove qui in aliena potestate est negotium gestum erit - On the dealings of the ship’s captain, business manager, and person in power) and contains one of the boldest and most ingenious creations of praetorian law,53 the early history of which is unfortunately blurred, but can be reconstructed on the basis of a few classical – therefore much later – legal texts.

Ulpian ([29 ad ed.], D. 15.1.1 pr.) records that the praetor attended first to remedies given for the full amount (in solidum) on the basis of contracts concluded with persons in power. Remedies giving rise to a limited liability (dumtaxat de peculio aut de in rem verso) and those based on the contracts of non-dependent persons (sui iuris, extranei) should therefore be regarded as later additions or extensions. Gaius (Inst. 4.70-71) explicitly says that the praetor started (‘inprimis’) with the grant of a remedy on authorized transactions carried by a dependent (actio quod iussu) and added by analogy (‘eadem ratione’) two further remedies

48 Sirks (forthcoming), with reference to Ulp. (26 ad ed.) D. 12.1.9.8; Celsus (5 dig.) D. 12.1.32; and Ulp. (30 ad ed.) D. 16.3.7.2 (about the insolvency of nummularii).

49 Cf. above, with Giachi (2001).

50 Serrao (1989); Zimmermann (1990), 34-67, esp. 45-58; Aubert (1994); Cerami, in Cerami et al. (2010), 34-63.

51 Cf. above, XII Tables 8.2 and 12.2 (Crawford); Ulp. (14 ad ed.) D. 4.9.3.3; and id. (18 ad ed.) D. 4.9.7.4.

52 Ulp. (28 ad ed.) D. 14.3.1. The blurring of criminal and civil liability is touched upon by Ulp. (1 ad ed. aed. cur.) D. 21.1.23.4-5.

53 The latest work on this part of the edict (formulae) is Micelli (2001), esp. 185-228.

 

15

 

for the transactions concluded by dependent ship’s captains (actio exercitoria) or business managers (actio institoria). The liability of the father/master was based on his willingness (voluntas) to allow contracts to be made with his dependent. This willingness was expressed through the appointment (praepositio) of the ship’s captain (magister navis) to a ship (navis) and of the business manager (institor) to a business (taberna aut quaelibet alia negotiatio). It is possible though not sure that originally the actio institoria was available only in the context of

the management of a shop (taberna).54 Alternatively, a rural context, such as the Catonian villa, may have provided the original Sitz im Leben of the actio institoria, since non-legal sources refer to institores as vilici and actores.55 The extension to other negotiationes, such as workshops (officinae, fabricae), credit institutions (mensae), etc., would only be natural.

Whether ships should be regarded as a later extension is a controversial question. Gaius (Inst. 4.71) and Justinian’s Digest (D. 14.1 and 3) and Code (CI. 4.25) invariably present the actio exercitoria before the actio institoria. This suggests that by the time of the composition of the Edictum Perpetuum (ca. AD 130) the order of presentation reflected the prominence – not necessarily the priority – of the actio exercitoria over the actio institoria. On the other hand, both Gaius (4.71) and Ulpian ([28 ad ed.] D. 14.1.1 pr.) stress the specificity of the conditions in which a ship’s captain works in comparison with a business manager: the distance separating agent and principal from one another in the context of seaborne trade makes it more difficult for third parties to check the agent’s legal status and sphere of competence. Subappointments are more readily acceptable in the case of ship’s captains than business managers, ‘for practical reasons’ (‘propter utilitatem’, ibid. D. 14.1.1.5). It also seems that, by contrast with business managers, ship’s captains are presented as non-dependents, which I believe to be the result of a later development: third parties could elect whether to sue the ship’s operator (exercitor) or the ship’s captain; the operator had no remedy against third parties contracting with his captain, because he allegedly did not need any, since he could sue the captain on the contract of employment or mandate which defined the relationship between principal and agent. Ulpian notes however that ‘in practice’ (‘plane’) the prefects in charge of the corn supply (annona) and provincial governors helped them in an informal way.56 The specificity of agency in the context of seaborne trade explains why two distinct but related remedies were necessary, and may suggest that the actio exercitoria developed from a more general actio institoria into a ground¬breaking legal instrument, both remedies eventually applying to the contracts of non¬dependent agents.57

The main legal issue discussed by the jurists in connection with both remedies concerns the scope of the appointment (praepositio). In order to give rise to the principal’s full liability, the contract concluded by the agent must pertain to the business he or she –

54 Földi (2001), 78-84.

55 Aubert (1994), 117-200.

56 Ulp. (28 ad ed.) D. 14.1.1.17-18 (‘extra ordinem’) to be compared with Ulp. (28 ad ed.) D. 14.3.1, citing the late second-century jurist Marcellus, and Gaius (9 ad ed. prov.) D. 14.3.2. Cf. Sirks (2002), 139.

57 Aubert (1999a). On the actio exercitoria, cf. Gaurier (2004), 79-95.

 

16

 

women and children could be appointed58 – is in charge of. Republican and Augustan jurists, such as Servius and Labeo, and their successors list various types of facilities in connection with which specific transactions may give rise to an actio institoria/exercitoria, thus contributing to the definition of business and enterprise in the Roman world.59 The scope of the appointment was implicitly conditioned by the nature of the business, and could be explicitly spelled out in a charter (lex praepositionis) used as a job description.60 Any extension (iussum) or limitation (proscriptio) had to be publicly posted and advertised. Double appointments – combining two (or more) types of activities under the responsibility of the same manager – are attested.61

Ulpian also discusses cases of subappointment ([28 ad ed.] D. 14.1.1.5), of joint managers working for a single operator (ibid., D. 14.1.1.13-14 and 14.3.11.5), and of joint operators appointing a single manager, either within a partnership (societas) or as joint owners (ibid., D. 14.3.13.2). The latter configuration was identified as the possible ancestor of company law, so prominent in modern commercial/business law.62 The Roman law of indirect agency also applied to collectivities, both private and public, such as companies of publicans (societates publicanorum), professional and religious associations (collegia), and towns (municipia and coloniae).63

The second part of the praetorian edict on indirect agency deals with those remedies that strictly apply to transactions carried out by dependents. The edictum triplex (Ulp. [29 ad ed.] D. 15.1.1.1) includes the actio quod iussu (D. 15.4), mentioned above as the likely original remedy for full liability of the principal, and two additional, eventually intertwined remedies (actio de peculio aut de in rem verso, D. 15.1-3), which called for a limited liability on the part of the principal on account of the contracts concluded by and with his dependent.64 Connected with the actio de peculio was a sixth remedy (actio tributoria, D. 14.4), whereby the principal was treated like any of the creditors when the agent’s insolvency gave rise to an

action on his peculium.65 The sheer order of presentation of the various remedies in the Digest (D. 14 and 15) and assumedly in the Edictum Perpetuum (EP 2.18.101-105) shows that

58 Ulp. (28 ad ed.) D. 14.1.1.21 and 14.3.7.1; Gaius (9 ad prov. ed.) D. 14.3.8. Cf. Aubert (1994), 43, 56, 140-41, 193, 372, 224-26, 292-93, and 419-20. The impact of the SC Velleianum (ca. AD 46) (EP 2.18.105; D. 16.1; CI: 4.29) on the ability of women to act as agents is unclear to me.

59 Ulp. (28 ad ed.) D. 14.3.5.1-15 and 14.3.13 pr.; Paul (29 ad ed.) D. 14.3.16 and id. (30 ad ed.) D. 14.3.17 pr., etc.

60 For lex praepositionis as a kind of lex contractus, cf. Aubert (2003); id. (2005a); id. (2005b); and id. (2007a); Aubert – Raepsaet (2011).

61 Ulp. (28 ad ed.) D. 14.1.1.12 (certa lex); ibid. D. 14.3.11.2-6 (proscriptio); ibid. 14.3.13 pr. (double appointment); and Gaius (9 ad ed. prov.) D. 14.5.1 (iussum). Cf. Aubert (1993); id. (1994), 6-14, 50-52, and 335; id. (2003); id. (2005a); id. (2005b); id. (2007a); Jakab (2008); and Aubert – Raepsaet (2011) (mandate).

62 Di Porto (1984), 169-204; Aubert (1994), 54-57 and 62-63, with additional references, to which Paul (4 ad Plaut.) D. 14.3.14 should be added. For Roman company law, cf. Crook (1967), 229-36.

63 Aubert (1994), 325-47; id. (1999b). On publicans, cf. EP 3.32.183-185; D. 39.4, with Gaius’ commentary ad edictum praetoris titulo de publicanis; and Maganzani (2002).

64 Actio quod iussu: Schleppinghoff (1996); and Coppola Bisazza (2003). Actio de peculio: Aubert (2010), with earlier bibliography. Actio de in rem verso: Chiusi (2001).

65 Actio tributoria: Chiusi (1993).

17

 

it cannot reflect the chronology of their respective creation. The transfer of the actia quad iussu to the very end of the series (D. 15.4) suggests that the order of presentation reflects the relative importance of each individual remedy in comparison with one another, the actia quad iussu being at best subsidiary in the classical period. Therefore it is telling that both commercial remedies (actianes exercitaria and institaria) applying to the contracts concluded by both dependents and non-dependents had taken precedence over all the others by the time of redaction of the Edictum Perpetuum in the second century AD. Accessorily, the reconstructed relative chronology of the creation of the so-called actianes adiecticiae qualitatis (with the addition of the actia tributaria) and their respective order in the Edictum Perpetuum indicate that the praetor favored the interests of third parties contracting with dependent agents over those of principals whose liability was engaged by their dependents’ transactions.

Gaius, in the ninth book of his commentary on the provincial edict, to which we will move shortly, reports that governors (pracansules) ensured that those who contracted with persons in power obtain their due: if the commercial remedies (a. exercitaria, institaria, and, paradoxically, tributaria) did not apply, the governor would grant a remedy for full liability on authorized transaction (a. quad iussu), provided such authorization existed, or on the enrichment of the family estate (a. de in rem verso), or, in the last resort, on the peculium (D. 14.5.1). The order of preference is altogether clear, and reflects the sense of equity (‘ex bono et aequo’) of both the third party and the political/judicial authority. In a separate case, the master of a runaway slave appointed to lend money and to accept security (pignus) fought off a suit from barley traders who had been promised payment on behalf of customers. Interestingly, the prefect of the corn supply stepped in on behalf of the traders and decided to hold the master liable in full, pointing out that the slave was notoriously in the habit of being involved in various businesses, such as renting warehouses (harrea). The early-third-century jurist Paul, who reports the case in which he seemed to have acted as the master’s lawyer, unsuccessfully argued for considering the slave’s payment as a type of guarantee (fideiussia), as opposed, presumably, to a receptum argentarii (cf. above). On appeal, the prefect’s decision was upheld by the emperor (Paul [1 decretarum] D. 14.5.8).

c) edictum pravinciale

Ancient legal commentaries leave no doubt that the bulk of edictal law pertaining to commercial life was established through the praetorian edict, before being developed by the jurists. As the case discussed in the last paragraph shows, provincial governors and prefects also used their ius edicendi to intervene when needed. Gaius’ lone commentary on the provincial edict (in thirty books vs. only ten for his commentary on the praetorian edict)66 suggests that a general provincial edict, distinct from individual edicts applying to distinct provinces, existed by the second century, but that it was not so different from the praetorian edict to justify a separate treatment by any other classical jurist. Like the edict of the curule aediles, it may have been absorbed in the Edictum Perpetuum at one point.

66 Martini (1969), 103-28; Santaluccia (1975).

18

 

The situation may have been different in the Republican period. We happen to know a bit about Cicero’s own edict as governor of Cilicia in 51/50 BC through his letters to his friend Atticus. In the winter of 50, Cicero was alternatively approached by the Salaminians of Cyprus and by M. Scaptius and P. Matinius, M. Iunius Brutus’ agents, for the recovery of a debt owed by the former to the latter. The disagreement bore on the interest to be paid. Cicero argued that he had promised (in his edictum translaticium) that he would not allow for more than twelve percent yearly compound interest (Att. 5.21.10: ‘cum anatocismo anniversario’), whereas Scaptius was asking for 48% simple interest. Scaptius opposed to Cicero’s edict a senatorial decree passed a few years before (56 BC) compelling the governor of Cilicia to honor the bond (syngrapha) as such, in blatant contradiction with the lex Gabinia of 68/67 BC forbidding Romans to lend money to provincial communities. The details of the story and its outcome should not concern us here. Cicero obviously felt constrained by the terms of his own edict (described in a later letter, Att. 6.1.15-16), though not by those of his predecessor in the position of governor of Cilicia. Cicero’s provincial edict, kept intentionally short, was derived from the Asiatic edict of Q. Mucius Scaevola (cos. 86 BC) and divided into two parts (genera): one was considered exclusively provincial and dealt with civic accounts (rationes civitatum), debt (aes alienum), rate of interest (usura), contracts (syngraphae), and regulations applying to publicans; the other part, dealing with inheritance, possession and sale of goods, etc., contained usual edictal material. Some rules remained unwritten, but Cicero boasts that he would let provincials use their own laws in their own courts, thus maintaining the fiction of restored autonomy. In disputes between publicans and provincials, the governor’s protection of the latter amounted to no more than a temporary measure, advertised as such to pressure the Greeks to settle as quickly as possible: after a fixed deadline the provision of the provincial edict regarding the rate of interest would give way to the terms of the agreements (pactiones). Provincial governors, like aediles and praetors, used their ius edicendi both before and during their term of office, a double practice allowing them to react to unexpected circumstances and problems while exposing them to undue pressure on the part of groups and individuals, and to the risk of self-contradiction.67 Accessorily, local customs (mos regionis) and laws could be taken into account (Ulp. [24 ad ed.] D. 25.4.1.15, admittedly a different context).

The combined creativity of curule aediles, urban and peregrine praetors, and provincial governors provided ample material for the jurists to work on to adjust the law to the needs of the business community in a growing and increasingly interconnected Mediterranean world. The flexibility of law-making through temporary or permanent edicts combined with a constant, recurring, and diversified exposure to neighboring legal systems, especially in the Greek East, resulted in the blending of the old Roman legal system. Cicero’s Salaminian affair brought him into contact with the Greek institution of the syngrapha, which, like the better attested chirographum, eventually sifted into Roman law as a form of written contract in addition to real, oral, and consensual contracts.68

67 Cic., Att. 5.21.10-13 (13 Febr. 50); 6.1.15-16 (22 Febr. 50); and 6.2.7-10 (early May 50). Rotondi (1912/1990), 373-74 (lex Gabinia). Cf. Pugliese (1964/1985); Martini (1969), 11-102; Peppe (1991); and Maganzani (2007).

68 Gaius, Inst. 3.134; Meyer (2004), 12-19 and 125-68.

 

19

 

Legal borrowing

Unsurprisingly, the Greeks were mostly influential in the field of maritime law. In spite of what the Romano-Carthaginian treaties of the early Republican period (cf. above) may suggest, the Romans were late – in relation to the Greeks and Carthaginians – in developing seaborne trade and their own sea power. The Mediterranean world, especially in the East, was already bursting with commercial activities in the classical and Hellenistic periods. Navigation was regarded as – and to some extent was – a dangerous activity, and the Greeks had devised some legal institutions aimed at minimizing the financial risks attached to it: bottomry loan (foenus nauticum) and compensation (contributio) for jettison (iactus) are two forms of maritime insurance which the Roman jurists elaborated on the basis of allegedly preexisting arrangements familiar to the trading community. How these arrangements passed into Roman law – provided they were actually borrowed and not simply reinvented as the logical solution to a universal problem – is a mystery, but it is fairly clear that the edict was not the way.69

Maritime loans are attested in the Roman world as early as the second century BC.70 The loan (pecunia traiecticia) was made out by private investors or professional bankers – acting as middlemen – to a shipper or group of shippers in a partnership, and had to be paid back only if and when ship and cargo reached the agreed destination. Ship and cargo were considered securities for the loan, whose duration was limited in time (e.g., 200 days in Callimachus’ case, for a trip from Beirut to Brindisi).71 In case of shipwreck, pirate attack, or other acts of God, the creditor would bear the loss. On the other hand, if the navigation was successful, the creditor was entitled to collect interests at a much higher rate than the law permitted for regular loans (12% yearly, cf. above). Many points remain problematic, and unfortunately neither the nine excerpts in the Digest (D. 22.2, from Servius to Ulpian) nor the four Diocletianic constitutions from the Code (CI. 4.33) provide more than hints. It is not clear, for instance, on which ground the legal limit of the rate of interest could be exceeded. We know of no statute, senatorial decree, plebiscite, or edict lifting it, although no less than four laws on usury have been proposed or passed between 217 and 192 BC.72 The jurists and the drafters of imperial constitutions insist on the notion of periculum creditoris (moneylender’s risk), thus revealing a favorable bias toward shippers, who had to show nothing but good faith.73 Maritime loans sound like a necessary evil akin to both speculation and insurance against disasters.

69 Crook (1967), 223-25. Cf. in general Huvelin (1929), 184-218; Rougé (1966); Gaurier (2004), 97-133. On the Greek daneion nautikon, cf. Schuster (2005).

70 Plut., Cato Maior 21.6. Cf. D. 22.2; and CI. 4.33, with Litewski (1973); Biscardi (1974); Casson (1986); id. (1990); Zimmermann (1990), 181-86; and Ankum (2000). This type of loan is occasionally represented in the papyri, cf. for instance SB III 7169 (BL, second cent. BC) for a trip to the Somali coast (with Roman names: ll. 12, 19, 21!); SB III 7170 (id.); SB XIV 11850 (BL, Theadelphia? Febr. 13, AD 149); and SB XVIII 13167 (BL, so-called Muziris papyrus, mid-second cent. AD) for a trip from Alexandria to India and back.

71 Scaevola (28 dig.) D. 45.1.122.1; Johnston (1999), 95-96; and Sirks (2002), 142-49.

72 Rotondi (1912/1990), 99.

73 Moneylenders had a say in legislation in a later period, cf. Nov. Iust. 106 (540) and 110 (541); cf. Jones (1964), 350.

20

 

For disasters did strike, though unevenly. Greek and Roman novels, including Paul’s report in the Acts of the Apostle (27:14-20), betray both fascination for and familiarity with standard procedure during storms on sea. To rescue the ship, part or all of the cargo sometimes had to be jettisoned. Those whose goods had been saved thanks to the others’ – sometimes unwilling – sacrifice were called to contribute to the loss. The Roman jurists acknowledge the routine resort to some foreign usage,74 the so-called lex Rhodia de iactu, the Rhodian origin of which is little more than hypothetical. From what can be reconstructed on the basis of D. 14.2 and Pseudo-Paul’s Sententiae (2.7), it seems that a contribution was expected not only in case of jettison, but also for ransoming from pirates, according to late Republican and Augustan jurists like Servius, Ofilius, and Labeo (Paul [34 ad ed.] D. 14.2.2.3). Those and later classical jurists devised a very sophisticated system of calculating the estimated, respective market value of the goods, either lost, damaged, or saved. Even though the lex Rhodia was not included in any edict and should be regarded as a mere appendix to the law of hire and lease (locatio conductio), the title dedicated to it was placed between the actio exercitoria (D. 14.1) and the actio institoria (D. 14.3) by the compilers of the Digest, because the owner(s) of the jettisoned goods (vectores) would have a remedy against the ship’s captain, and because the lex Rhodia was akin to a lex contractus or lex praepositionis, in that it expounded the terms of the contract of hire between shipper and merchants.75

The legal status of such a usage is uncertain, but somewhat illuminated by a very controversial text by the late second-century jurist Volusius Maecianus, the alleged author of a monograph on the lex Rhodia (D. 14.2.9). A petitioner writes to the emperor to complain about being robbed by islanders after a shipwreck. The emperor’s answer, based on an earlier ruling by Augustus, specifies that the lex Rhodia applies whenever it is not in conflict with Roman law. Taken at face value, the text indicates that legal vacuum could – or had to – be filled with existing usage, whatever they are or come from: foreign customs are better than nothing, and Roman lawmakers could not be expected to cover all situations. Maritime law was obviously permeable to external input.

Public law: regulation and exploitation

The combination of edictal law and jurisprudence, and the occasional adoption of international/local norms (laws and customs) certainly facilitated the development of commerce by providing the business community with adequate legal instruments and protection. However, there is another side to the coin, reflecting social and political concerns and fiscal necessities. This is where public law steps in.

Starting in the early Republican period, a series of leges fenebres tended to limit the rate of interest before banning interest altogether, though unsuccessfully (cf. above). Other public laws bearing on commercial activities regulated the food supply (leges

74 In the seventh cent., Isidorus of Seville (Etym. 5.17) speaks of Rhodian laws of maritime commerce as ‘antiquitus mercatorum usus.’

75 Zimmermann (1990), 406-12; Chevreau (2005); Aubert (2007a); Badoud (forthcoming).

 

21

 

annonariae/frumentariae), luxury consumption (leges sumptuariae), the occupation of agricultural land (leges agrariae), and taxation.76 The point was to preserve the social order and the political power of the elite. One such law, the plebiscitum Claudianum of 219/218 BC, reiterated in slightly different form by the lex Iulia repetundarum of 59 BC, barred senators and their sons from owning – though not operating – ships of large capacity (over 300 amphoras), excluding them from lucrative public contracts connected with the food supply and hampering the marketing of the produce of ever-growing agricultural estates.77 The ban may have been instrumental in developing the actio exercitoria and its extension to non¬dependent ship’s captains (cf. above). Alternatively, the activities of shippers (navicularii) drew the attention of imperial government officials and gave rise to an abundant legislation down to late Antiquity. At stake was the steadiness of the food supply of Rome and, since the fourth century, that of Constantinople. Organized in associations (collegia, corpora), they enjoyed at first privileges such as exemption from compulsory public services, and ended

up fulfilling a public service (onus publicum) even against their will.78

State control over economic activities was not limited to shipping and became a general phenomenon in the fourth and fifth centuries. Unsurprisingly, it affected trades connected with the food industry (bakers, meat sellers, etc.), but extended to other

commercial activities.79 It was mostly exercised through taxation: in the Republican and early Imperial periods, trade was subjected to all kinds of taxes (vectigalia), above all tolls and custom duties (portoria) at both municipal and imperial levels.80 It is difficult to estimate the impact of taxation on the volume of trade, but it is clear that the burden increased with time: from the reign of Constantine until 498, a special tax called collatio lustralis or chrysarguron was collected in gold and silver on behalf of the imperial treasury (sacrae largitiones) from merchants (negotiatores, mercatores) who therefore had to be registered

(negotiatorum matricula).81 By then, the time of laisser-faire and empiric promotion of commercial activities on the part of public authorities was long gone.

Traders were not only considered a fiscal golden hen by a needy government. The attitude of lawmakers towards them had changed. The preamble of Diocletian’s Price Edict (Nov./Dec. 301) accuses them in no uncertain terms of greed and selfishness, the cause of uncontrollable inflation, and threatens them and their agents (institores) with capital punishment unless they desist from speculating and abide by the law setting maximum

76 Rotondi (1912(1990), 92-100.

77 Liv. 21.63; Cic., 2Verr. 5.17.44-18.45; and Ps.-Paul, Sent. (Leiden frg. published in 1956) 3 (p. 5), ll. 7¬11; Aubert (2004), 166-68 and 178, n. 17; Tchernia (2007), with a telling calculation of the capacity of the ship.

78 Sirks (1991); de Salvo (1992).

79 Pistores, suarii, pecuarii, boarii, vinarii, etc. CI., Book 11; CTh., Books 10 and 14; Sirks (1991), 307-413; Földi (2001), 85-87.

80 France (2001), for Gaul; Cottier et al. (2008), for Asia Minor; and Delmaire (1989), 275-312, for late Antiquity.

81 CTh. 13.1; CI. 11.1 (abolition in the East); Jones (1964), 351 and 431-32; Delmaire (1989), 254-74, esp. 367, n. 41, with reference (for the matricula) to CTh. 16.2.15.1 (359 or 360); and P. Oxy. L 3577 (28 Jan. 342). Cf. also Minaud (2011), §298. Other taxes on trade are attested: siliquaticum, canon telonei and transmarinorum, cf. Jones (1964), ss.vv.

 

22

 

prices, possibly disconnected from market prices. A few years later, Lactantius claimed that the ill-advised imperial policy resulted in both slaughter and scarcity of goods.82 In spite of its failure and eventual repealing, the measure shows that soldiers, not traders, have the emperors’ ear.

Conclusion

To the question, was Roman business law designed for traders, we can propose a qualified answer: looking at the time of its development (Republic and Principate), there is no doubt that edictal law and jurisprudence jointly produced legal institutions of unprecedented efficiency, offering pragmatic solutions to practical problems, occasionally borrowed from subjected communities. This is true of private law. As for public law, it was concerned with social stability and fiscal necessity, not economic growth, and should be viewed as a permanent hindrance.

To the question, was Roman business law inspired by traders, the answer is less clear-cut. Roman magistrates and jurists invariably belonged to the elite. The dominant ideology would have liked us to think of it as a landed aristocracy, but the evidence suggests that senatorial, equestrian, and curial families were heavily –for senators, perhaps indirectly – involved in commercial activities, in spite of legal prohibition and social pressure. Because of the pragmatic nature of edictal law and the flexibility in its application warranted in juristic writings, it is difficult to exclude close and recurrent contacts between the business community and lawmakers, as Cicero’s indiscretion occasionally reveals. The sheer quantity and strong rhetorical taste of legal opinions preserved in the Digest cannot be allowed to blur the fact that Roman business law is mostly edictal law, explained and extended by the jurists.

To the question, was Roman business law enforced by traders or, at least, in special courts reserved for traders, the answer can only be a non liquet: there is not much evidence

for a Roman equivalent to the Greek emporikai dikai.83 Roman courts were composed of recuperatores, sometimes limited to a iudex unus or arbiter, and selected from a list of respected people, some of whom could have been negotiatores (Cic., 2Verr. 2.13.32-34). In the imperial period, jurisdiction passed to civil servants (praefecti, praesides, iudices) whose interests can hardly be confused with those of traders, and whose technical competence rested less on their personal legal expertise and practical experience than on the services of their staff. The evident worsening of the condition of traders in late Antiquity does not point toward the establishment of judicial privileges.

82 Edictum Diocletiani et collegarum de pretiis rerum venalium (ed. M. Giacchero, 1974), 1: 134-37, esp. ll. 64-136; Lact., De mortibus persecutorum 7.6-7. Cf. Corcoran (2000), 205-33.

83 Cohen (1973); and id. (2005), 300-2. The Roman evidence collected by Minaud (2011), §365 (Cic., 2Verr. 2.13.34; CIL XIV 2630; SHA, Alex. 33.2; CTh. 14.7.1 [397]; CI. 3.13.7 pr. [502]; Isid., Etym. 5.17) is inconclusive, with the possible exception of the passage from the Historia Augusta.

23

 

Roman Business Law: Bibliography

Andreau, J. (1987) La vie financière dans le monde romain. Les métiers de manieurs d’argent (IVe siècle av. J.-C. –IIIe siècle ap. J.-C.). Rome.

Andreau, J. (1999) Banking and Business in the Roman World. Cambridge.

Ankum, H. (2000) ‘Some aspects of maritime loans in old-Greek and in Roman law,’ in Velissaropoulou-Karakosta, I et al. (eds.), Mélanges I. Triantaphyllopoulos. Komotini, 293-306.

Aubert, J.-J. (1993) ‘Workshop managers,’ in Harris, W.V. (ed.), The Inscribed Economy. Production and Distribution in the Roman Empire in the Light of instrumentum domesticum. Ann Arbor, MI, 171-89.

Aubert, J.-J. (1994) Business Managers in Ancient Rome. A Social and Economic Study of Institores, 200 BC – AD 250. Leiden/New York/Cologne.

Aubert, J.-J. (1999a) ‘Les institores et le commerce maritime dans l’empire romain,’ Topoi 9: 145¬64.

Aubert, J.-J. (1999b) ‘La gestion des collegia: aspects juridiques, économiques et sociaux,’ CCG 10: 49-69.

Aubert, J.-J. (2003) ‘En guise d’introduction: contrats publics et cahiers des charges,’ in Aubert, J.-J. (ed.) Tâches publiques et entreprise privée. Neuchâtel/Genève, 1-25.

Aubert, J.-J. (2004) ‘The Republican economy and Roman law: regulation, promotion, or reflection?’ in Flower, H. I., The Cambridge Companion to the Roman Republic. Cambridge, 160-78.

Aubert, J.-J. (2005a) ‘Corpse disposal in the Roman colony of Puteoli: public concern and private enterprise,’ in Harris, W.V. – Lo Cascio, E. (eds.) Noctes Campanae. Studi di storia antica ed archeologia dell’Italia preromana e romana in memoria di Martin W. Frederiksen. Napoli, 141-57.

Aubert, J.-J. (2005b) ‘L’estampillage des briques et des tuiles: une explication juridique fondée sur une approche globale,’ in Bruun, C. (ed.), Interpretare i bolli laterizi di Roma e della Valle del Tevere: Produzione, storia economica e topografia. Rome, 53-59.

Aubert, J.-J. (2007a) ‘Dealing with the abyss: the nature and purpose of the Rhodian sea-law on jettison (Lex Rhodia de iactu, D 14.2) and the making of Justinian’s Digest,’ in Cairns, J. W. – du Plessis, P. (eds.), Beyond Dogmatics. Law and Society in the Roman World. Edinburgh, 157-72.

Aubert, J.-J. (2007b) ‘L’économie romaine et le droit de la représentation indirecte sous la République romaine,’ in Fides, Humanitas, Ius. Studii in onore di Luigi Labruna. Napoli, 215-30.

Aubert, J.-J. (2010) ‘Productive investment in agriculture: instrumentum fundi and peculium in the later Roman Republic,’ in Carlsen, J. – Lo Cascio, E. (eds.), Agricoltura e scambi nell’Italia tardo-repubblicana. Bari, 167-85.

Aubert, J.-J. (2011) ‘Vitia animi: tares mentales, psychologiques, caractérielles et intellectuelles des esclaves en droit romain,’ in Maffi, A. – Gagliardi, L. (eds.), I diritti degli altri in Grecia e a Roma. Sankt Augustin, 236-48. Aubert, J.-J. – Raepsaet, G. (2011), ‘Un mandat inscrit sur une sigillée argonnaise à Liberchies-Geminicacum,’ in L’Antiquité classique 80,139-56.Badoud, N. (forthcoming), La lex Rhodia de iactu, to be published in the Schweizerische Beiträge zur Altertumswissenschaft, based on his unpublished 2008 Neuchâtel doctoral dissertation, La cité de Rhodes. De la chronologie à l’histoire (Second part, 443-569).

Bianchini, M. (1989) ‘Diritto commerciale nel diritto romano,’ in Digesto delle Discipline privatistiche. Sezione commerciale IV. Torino, 320-33.

24

 

Bianchini, M. (2007) ‘Attività commerciali fra privato e pubblico in età imperiale,’ in Fides,

Humanitas, Ius. Studii in onore di Luigi Labruna. Naples, 423-38.

Biscardi, A. (1974) Actio pecuniae traiecticiae. Contributo alla dottrina delle clausole penali. Sec. ed.,

Torino.

Borkowski, A. – Du Plessis, P. (2005) Textbook on Roman law. Third ed., Oxford.

Brennan, T. C. (2000) The Praetorship in the Roman Republic. Oxford.

Capogrossi Colognesi, L. (1994) ‘Ius commercii, conubium, civitas sine suffragio. Le origini del

diritto internazionale privato e la romanizzazione delle comunità latino-campane,’ in

Corbino, A. (ed.), Le strade del potere. Catania, 3-64.

Carnazza, G. (1891) Il diritto commerciale dei Romani. Catania (non vidi).

Casson L. (1986) ‘New light on maritime loans: P. Vindob. G. 19792 (= SB VI 9571),’ in

Bagnall, R.S. – Harris, W.V. (eds.), Studies in Roman Law in Memory of A. Arthur Schiller.

Leiden/New York, 11-17.

Casson L. (1990) ‘New light on maritime loans: P. Vindob. G 40822,’ ZPE 84: 195-206.

Cerami, P. – Petrucci, A. (2010) Diritto commerciale romano. Profilo storico. Third ed., Torino.

Champaud, C. (1994) Le droit des affaires. Fifth ed., Paris.

Chevreau, E. (2005) ‘La lex Rhodia de iactu: un exemple de la réception d’une institution

étrangère dans le droit romain,’ TvR 73: 67-80.

Chiusi, T. J. (2001) Die actio de in rem verso im römischen Recht. Munich.

Chiusi, T. J. (2003) Contributo allo studio dell’editto ‘de tributoria actione’. Rome.

Churruca, J. de (2001) ‘Le commerce comme élément de civilisation dans la Géographie de

Strabon,’ RIDA 48: 41-56.

Cogliolo, P. (1922) Gli istituti commerciali nel diritto romano. Corso compilato dall’Avv. E. Fin!Zi e dallo

studente G.G. Traverso, Anno accademico 1921-1922. Genova.

Cohen, E. E. (1973) Ancient Athenian Maritime Courts. Princeton.

Cohen, E. E. (2005) ‘Commercial law,’ in Gagarin, M. – Cohen, D. (eds.), The Cambridge

Companion to Ancient Greek Law. Cambridge, 290-302.

Coppola Bisazza, G. (2003) Lo iussum domini e la sostitu!Zione nego!Ziale nell’esperien!Za romana.

Milano.

Corcoran, S. (2000), The Empire of the Tetrarchs: Imperial Pronouncements and Government, AD 284 

324. Rev. ed., Oxford.

Cottier, M. et al. (eds.) (2008) The Customs Law of Asia. Oxford.

Crook, J. A. (1967) Law and Life of Rome, 90 B.C. – A.D. 212. Ithaca, NY.

Daube, D. (1951) ‘The peregrine praetor,’ JRS 41: 66-70.

Delmaire, R. (1989) Largesses sacrées et res privata. L’aerarium impérial et son administration du IVe

au VIe siècle. Rome.

De Salvo, L. (1992) I corpora naviculariorum: economia privata e pubblici servi!Zi nell’impero romano.

Messina.

Di Porto, A. (1984) Impresa collettiva e schiavo ‘manager’ in Roma antica (II sec. a.C. – II sec. d.C.).

Milano.

Di Porto, A. (1997) ‘Il diritto commerciale romano. Una ‘zona d’ombra’ nella storiografia

romanistica e nelle riflessioni storico-comparative dei commercialisti,’ in Marrone, M. (ed.),

No!Zione, forma!Zione e interpreta!Zione del diritto dall’età romana alle esperien!Ze moderne. Ricerche dedicate

al Prof. Filippo Gallo. Napoli, 3: 413-52.

Fadda, C. (1903/1987) Istituti commerciali del diritto romani. Le!Zioni dettate nella R. Universitä di

Napoli nell’anno scolastico 1902-1903. Naples. (Repr. con una nota di lettura di L. Bove).

 

25

 

Ferenczy, E. (1969) ‘Die römisch-punischen Verträge und die Protohistorie des commercium,’ RIDA 16 (1969) 259-82.

Földi, A. (2001) ‘Eine alternative Annäherungsweise: Gedanken zum Problem des Handelsrechts in der römischen Welt,’ RIDA 48: 65-90.

France, J. (2001) Quadragesima Galliarum. L’organisation douanière des provinces alpestres, gauloises et germaniques de l’empire romain. Rome.

Frier, B. W. (1980) ‘Roman law and the wine trade: the problem of ‘vinegar’ sold as wine,’ ZRG 100: 257-95.

Gaudemet, J. (1982). Institutions de l’Antiquité. 2nd ed., Paris.

Gaurier, D. (2004) Le droit maritime romain. Rennes.

Gerkens, J.-F. (2003) ‘Chronique de la 55e session de la Société internationale Fernand de Visscher pour l’histoire des droits de l’Antiquité à Rotterdam (18 au 22 septembre 2001),’ RIDA 50: 489-516.

Giachi, C. (2001) ‘Storia dell’editto e struttura del processo in età pre-adrianea. Un’ipotesi di lavoro,’ in Atti del Convegno Processo civile e processo penale nell’esperienza giuridica del mondo antico in memoria di Arnaldo Biscardi, Siena, Certosa di Pontignano, 93-95 dicembre 2009. Rivista di diritto romano http://www.ledonline.it/rivistadirittoromano/attipontignano.html (12/01/09).

Giachi, C. (2005) Studi su Sesto Pedio. La tradizione, l’editto. Milano.

Goode, R. (2004) Commercial Law. Third ed. London.

Griffin, M. (1973) ‘The tribune C. Cornelius,’ JRS 63: 196-213.

Gröschler, P. (2002) Actiones in factum. Eine Untersuchung zur Klage-Neuschöpfung im nichtvertraglichen Bereich. Berlin.

Guarino, A. (1980) ‘La formazione dell’editto perpetuo,’ ANRW 2.13: 62-102.

Harris, W. V. (2003) ‘Roman governments and commerce, 300 B.C. – A.D. 300,’ in Zaccagnini, C. (ed.), Mercanti e politica nel mondo antico. Rome, 275-305.

Hoz Montoya, J. de la (2008), ‘Neron y el impuesto sobre la venta de esclavos,’ SDHI 74: 367¬395.

Humbert, M. (2005) ‘Il valore semantico e giuridico di vsvs nelle Dodici Tavole,’ in Humbert, M. (ed.), Le Dodici Tavole. Dai Decemviri agli Umanisti. Pavia, 377-400.

Huvelin, P. (1929) Etudes d’histoire du droit commercial romain (Histoire externe – Droit maritime). Paris.

Impallomeni, G. (1955) L’editto degli edili curuli. Padova.

Jakab, E. (2008) ‘Vertragspraxis und Bankgeschäfte im antiken Puteoli: TPSulp. 48 neu interpretiert,’ in Verboven, K. – Vandorpe, K. – Chankowski, V. (eds.), Pistoi dia tèn technèn. Bankers, Loans and Archives in the Ancient World. Studies in Honour of Raymond Bogaert. Leuven, 321-44.

Johnston, D. (1999) Roman Law in Context. Cambridge.

Jones, A. H. M. (1964) The Later Roman Empire, 284-602. A Social, Economic, and Administrative Survey. London/Norman, OK (repr. Baltimore 1986).

Jones, D. (2006) The Bankers of Puteoli. Finance, Trade and Industry in the Roman World. Stroud. Kaser, M. (1953) ‘Vom Begriff des commercium,’ in Studi in onore di V. Arangio-Ruiz. Naples, 2: 131-67.

Kaser, M. (1971-1975) Das römische Privatrecht. Second ed. Munich.

Kelly, D. – Holmes, A. – Hayward, R. (2005), Business Law. Fifth ed., Abingdon/New York. Kremer, D. (2005) ‘Trattato internazionale e legge delle Dodici Tavole,’ in Humbert, M. (ed.), Le Dodici Tavole. Dai Decemviri agli Umanisti. Pavia, 191-207.

 

26

 

Kudlien, F. (2001) ‘Mutator und permutatio als finanz- und handelstechnische Termini,’ MBAH

20.1: 48-54.

Labruna, L. (1994) ‘Il diritto mercantile dei Romani e l’espansionismo,’ in Corbino, A. (ed.), Le

strade del potere. Catania, 115-37.

Legeais, D. (2003) Droit commercial et des affaires. Fifteenth ed., Paris.

Litewski, W. (1973) ‘Römisches Seedarlehen,’ Iura 24: 112-83.

Lo Cascio, E. (2007) ‘The early Roman empire: the state and the economy,’ in Scheidel, W. –

Morris, I. – Saller, R. (eds.), The Cambridge Economic History of the Greco-Roman World.

Cambridge, 619-47.

Lucas, F.-X. (2005) Le droit des affaires. Paris.

Malanima, P. (2009) ‘Progresso o stabilità? il mercato nelle economie preindustriali,’ Studi storici

50: 5-23.

Maganzani, L. (2002) Pubblicani e debitori d’imposta. Ricerche sul titolo edittale de publicanis. Torino.

Maganzani, L. (2007) ‘L’editto provinciale alla luce delle Verrine: profili strutturali, criteri

applicativi,’ in Dubouloz, J. – Pittia, S. (eds.), La Sicile de Cicéron. Lectures des Verrines.

Besançon, 127-46.

Mantovani, D. (2000) ‘L’édit comme code,’ in Lévy, E. (ed.), La codification des lois dans

l’Antiquité. Actes du colloque de Strasbourg, 27-29 novembre 1997. Paris, 257-72.

Mayer-Maly, Th. (2003) ‘Commercium,’ TR 71: 1-6.

Meyer, E. A. (2004) Legitimacy and Law in the Roman World. Tabulae in Roman Belief and Practice.

Cambridge.

Meylan, Ph. (1959) ‘Permutatio rerum,’ in Ius et lex. Festgabe zum 70. Geburtstag von M. Gutzwiller.

Basel, 45-64 (repr. 1977).

Merryman, J. H. (1985) The Civil Law Tradition. An Introduction to the Legal Systems of Western

Europe and Latin America. Second ed. Stanford.

Miceli, M. (2001) Sulla struttura formulare delle actiones adiecticiae qualitatis. Torino.

Minaud, G. (2011), Les gens de commerce et le droit à Rome. Aix-en-Provence.

Nörr, D. (2005) ‘Osservazioni in tema di terminologia giuridica predecemvirale e di ius

mercatorum mediterraneo: il primo trattato cartaginese-romano,’ in Humbert, M. (ed.), Le

Dodici Tavole. Dai Decemviri agli Umanisti. Pavia, 147-89.

North, D. (1981) Structure and Change in Economic History. New York/London.

North, D. (1991) Institutions, Institutional Change and Economic Performance. Cambridge.

North, D. – Thomas, R.P. (1973) The Rise of the Western World: A New Economic History.

Cambridge.

Palazzolo, N. (1991) ‘Sulla lex Cornelia de edictis,’ Labeo 37.2: 242-45.

Peppe, L. (1991) ‘Note sull’editto di Cicerone in Cilicia,’ Labeo 37: 14-93.

Petrucci, A. (1991) Mensam exercere. Studi sull’impresa finanziaria romana (II secolo a.C. – metà del

III secolo d.C.). Napoli.

Petrucci, A. (2002) Profili giuridici delle attività e dell’organizzazione delle banche romane. Torino.

Pinna Parpaglia, P. (1987) Per una interpretazione della ‘lex Cornelia de edictis praetorum’ del 67 A.C.

Sassari.

Pinna Parpaglia, P. (1992) ‘Lex Cornelia de edictis, mutui feneratizi, certezza del diritto,’ Labeo

38.2: 372-78.

Pugliese, G. (1964) ‘Riflessioni sull’editto di Cicerone in Cilicia,’ in Synteleia V. Arangio-Ruiz.

Napoli, 2: 972-96 (= [1985] Scritti giuridici scelti. Napoli, 99-115).

 

27

 

Pugsley, D. (1974) ‘The Aedilician Edict,’ in Watson, A. (ed.), Daube Noster. Essays in Legal History for David Daube. Edinburgh/London, 253-64.

Reduzzi-Merola, F. (2001) ‘Ventes d’esclaves sur les marchés de Campanie,’ in Routes et marchés d’esclaves. 26e colloque du GIREA. Besançon, 321-25.

Rotondi, G. (1912) Leges publicae populi romani. Milano.

Rougé, J. (1966) Recherches sur l’organisation du commerce maritime en Méditerranée sous l’empire romain. Paris.

Santalucia, B. (1975) L’opera di Gaio ‘ad edictum praetoris urbani’. Milano.

Scardigli, B. (1991) I trattati Romano-Cartaginesi. Pisa.

Schleppinkhoff, A. M. M. (1996) Actio quod iussu. Die Geheissklage (und ihre Bedeutung für die Entwicklung des Stellvertretungsgedanken im 19. Jahrhundert. Diss. Köln.

Schuster, S. (2005) Das Seedarlehen in den Gerichtsreden des Demosthenes: mit einem Ausblick auf die weitere historische Entwicklung des Rechtsinstitutes: dáneion nautikón, fenus nauticum und Bodmerei. Berlin.

Serrao, F. (1954) La ‘iurisdictio’ del pretore peregrino. Milano.

Serrao, F. (1989) Impresa e responsabilità a Roma nell’età commerciale: forme giuridiche di un’economia-mondo. Ospedaletto.

Serrao, F. (2000) ‘Impresa, mercato, diritto. Riflessioni minime,’ in Lo Cascio, E. (ed.), Mercati permanenti e mercati periodici nel mondo romano. Atti degli Incontri capresi di storia dell’economia antica (Capri 13-15 ottobre 1997). Bari, 31-67.

Silver, M. (2007) ‘Roman economic growth and living standards: perceptions versus evidence,’ Ancient Society 37: 191-252.

Sirks, B. (1991) Food for Rome. Amsterdam.

Sirks, B. (2002) ‘Sailing in the off-season with reduced financial risk,’ in Aubert, J.-J. – Sirks, B. (eds.), Speculum iuris. Roman Law as a Reflection of Social and Economic Life in Antiquity. Ann Arbor, MI, 134-50.

Sirks, B. (forthcoming) ‘Law, commerce and finance in the Roman empire,’ unpublished paper presented at the Fourth OxREP conference, 1-4 October 2009.

Tchernia, A. (2007) ‘Le plebiscitum Claudianum,’ in Andreau, J. – Chankowski, V. (eds.), Vocabulaire et expression de l’économie dans le monde antique. Bordeaux, 253-78.

Terpstra, T. (2008) ‘Roman law, transaction costs and the Roman economy: evidence from the Sulpicii archive,’ in Verboven, K. – Vandorpe, K. – Chankowski, V. (eds.), Pistoi dia tèn technèn. Bankers, Loans and Archives in the Ancient World. Studies in Honour of Raymond Bogaert. Leuven, 345-69.

Wolff, H.-J. (1979) Das Problem der Konkurrenz von Rechtsordnungen in der Antike. Heidelberg. Zimmermann, R. (1990) The Law of Obligations. Roman Foundations of the Civilian Tradition. Oxford.

 

28

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

LAKIREDDY BALIREDDY COLLEGE OF ENGINEERING

(AUTONOMOUS)

(Approved by AICTE, Accredited by NBA,

Affiliated to JNTUK, Kakinada and ISO 9001: 2008 Certified)

ACADEMIC REGULATIONS,

COURSE STRUCTURE

AND

DETAILED SYLLABUS

M.TECH – SYSTEMS AND SIGNAL PROCESSING 

DEPARTMENT OF ELECTRONICS AND COMMUNICATION

ENGINEERING 

L.B.Reddy Nagar, MYLAVARAM – 521 230

Krishna District, Andhra Pradesh State

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 0

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

INDEX

Contents

1. Introduction

2. Programme Offered Post raduation Page No.

2

2

3. Eligibility Criteria for dmission 2

4. ard of M. ec Degree 3

5. Duration of t e Programme 3

6. Semester- ise distribution of credits 3

7. Distribution and eig tage of Mar s 3

8. ttendance Regulations Condonation 5

9. Minimum cademic Re uirements 6

10. Course Pattern 6

11. ard of rade 7

12. Minimum Instruction days 8

13. eneral 8

14. ransitory Regulations 8

15. Course Code and Course umbering Sc eme 9

16. Medium of Instruction 10

17. mendments to Regulations 10

18. rade Card 10

19. Conduct and Disci line 10

20. Mal ractices 12

21. ard of Ran 12

22. Course structure 14

23. Syllabus 16-42

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 1

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

1. INTRODUCTION

cademic Programmes of t e institute are governed by rules and regulations as

a roved by t e cademic Council, ic is t e ig est cademic body of t e

Institute.

ese academic rules and regulations are a licable to t e students admitted during academic year 2010-11 into first year of t o year Postgraduate rogramme offered by t e college leading to Master of ec nology M. ec degree.

1.1 a ireddy alireddy College of Engineering, Mylavaram, an autonomous

institution, follo s Semester attern for all t o years of its Postgraduate M. ec rogramme it internal and e ternal evaluation.

1.2 Semester Pattern : Eac academic year s all be divided into t o semesters, eac of 20 ee s duration, including instruction, evaluation, etc. Eac semester consists of a minimum of 90 instruction days it at least 35 to 40 contact eriods er ee .

2. PROGRAMME OFFERED (POST GRADUATE)

Master of ec nology M. ec .

3. ELIGIBILITY CRITERIA FOR ADMISSION

e eligibility criteria for admission into irst year M. ec rogramme s all be as

mentioned belo :

i. dmissions to t e above rogram s all be made sub ect to t e eligibility, ualifications and s eciali ation rescribed by t e IC E from time to time.

ii. dmissions s all be made on t e basis of merit ran obtained by t e

ualifying candidate at E e amination or an entrance test conducted by

t e university sub ect to reservations rescribed by t e niversity/State government from time to time.

4. AWARD OF M.Tech DEGREE

student ill be declared eligible for t e a ard of t e M. ec Degree if e/s e fulfills t e follo ing academic regulations:

i. Pursued a course of study for not less t an t o academic years and not more t an four academic years.

ii. e student as to fulfill all t e academic re uirements i.e. Registered for 88 credits and as to secure all t e 88 credits it minimum grade oints.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 2

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

5. DURATION OF THE PROGRAMME

Students, o fail to fulfill all t e academic re uirements for t e a ard of t e degree it in four academic years from t e year of t eir admission, s all forfeit t eir seat in M. ec course.

6. SEMESTER –WISE DISTRIBUTION OF CREDITS:

Year First Semester

Credits Second Semester

Credits Total Credits

irst ear 3028

Mini Pro ect 2 Credits 3028

erm Pa er 2 Credits 60

Second ear Dissertation or 20

ec nical seminar-8 28

TOTAL 88

Table .1 Semester wise Credits Distribution 

7. DISTRIBUTION AND WEIGHTAGE OF MARKS:

i In I-Semester and II- Semesters, t e course of study consists of 6 t eory sub ects 2

laboratories or 6 t eory sub ects 1 aboratory. Ho ever, t e final year ill be on dissertation or and ec nical seminar only.

ii e erformance of a student in eac semester s all be evaluated sub ect ise it a ma imum of 100 mar s for t eory and 100 mar s for aboratory Courses. In addition, Mini ro ect, ec nical seminar, and dissertation or s all be evaluated for 50, 50 and 200 mar s res ectively.

iii or t eory sub ects t e distribution s all be 40 mar s 35 for est and 5 for

attendance for Internal E aminations and 60 mar s for t e End Semester E amination.

iv or t eory sub ects, during t e semester t ere s all be t o internal mid term

e aminations, for duration of 90 minutes. irst internal mid term e amination to be conducted in first and second units and t e second internal mid term

e amination to be conducted in t ird, fourt and fift units of eac sub ect.

Ho ever,75 eig tage for t e best and 25 for t e ot er internal mid term

e amination s all be considered for a arding internal mar s.

v e internal mid term e amination uestion a er s ould be for 35 mar s. Out of five

uestions given, student as to ans er any t ree uestions.

vi or aboratory courses, t ere s all be a continuous evaluation during t e semester

for 40 sessional mar s and 60 end semester e amination mar s. Of t e 40 mar s for internal, 15 mar s s all be a arded for day-to-day or and 15 mar s to be

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 3

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

a arded by conducting Internal laboratory test and 5 mar s for record or and 5 mar s for attendance. e end semester e amination s all be conducted by an e ternal e aminer and t e concerned internal e aminer.

vii Mini ro ect s all be submitted in re ort form and s ould be resented before t e

committee, ic s all be evaluated for 50 mar s. e committee consists of t e

Head of t e de artment, t e Su ervisor of mini ro ect and a senior faculty member oft e de artment. ere s all be no internal mar s for mini ro ect. e student as to secure minimum 50 mar s to be declared successful.

viii ere s all be ec nical seminar in second year. or t e seminar, t e student s all

collect t e information on a s eciali ed to ic and re are a tec nical re ort,

s o ing is understanding over t e to ic, and submit to t e de artment, ic

s all be evaluated by t e De artment committee consisting of Head of t e de artment, Seminar su ervisor and a senior faculty member. e seminars all be evaluated for 50 mar s based on is/ er resentation and t e submitted

re ort. ere s all be no e ternal e amination for seminar. e student as to

secure minimum 50 mar s to be declared successful.

i Dissertation Work: 

a. De artmental Dissertation Revie Committee DDRC s all be constituted it

t e Head of t e De artment as t e c airman and t o senior faculty as members to su ervise t e roceedings of t e dissertation or from t e time of allotment to submission.

b. Registration of Pro ect or :

student is ermitted to register for t e ro ect or after satisfying t e

attendance re uirement of all t e courses t eory and ractical courses u to

Second Semester. candidate as to submit, in consultation it is/ er

dissertation su ervisor, t e title, ob ectives and lan of action of is/ er dissertation or to t e DDRC for its a roval. Only after obtaining t e a roval of DDRC, t e student can initiate t e dissertation or .

c. e duration of t e dissertation is for t o semesters Second ear .

d. our co ies of t e dissertation re ort, certified by t e su ervisor s all be submitted tot e Institute.

e. Out of a total 200 mar s for t e dissertation or , 50 mar s s all be for Internal Evaluation and 150 mar s for t e End Semester E amination. e End Semester E amination viva-voce or defence s all be conducted by t e committee consisting of an E ternal E aminer, Head of t e De artment and t e dissertation su ervisor. E ternal e aminer ill be selected by t e Princi al/Director out of

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 4

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

t ree member anel submitted by t e Head of t e De artment, o are eminent

in t at field of study. or ma imum of 18 students, one dissertation evaluation committee as to be formed.

f. If t e or is not satisfactory, and if t e student is failed to secure t e minimum

50 of total mar s , t e student s all revise and resubmit t e dissertation re ort

before t ree mont s. If e/s e fails to get a satisfactory re ort again, t e dissertation s all be summarily re ected.

g. e to ics for mini ro ect and dissertation or s all be different from eac

ot er. e evaluation of dissertation or s all be conducted at t e end of t e

Second year.

. e student as to clear all t e sub ects of M. ec course to attend t e Viva

Voce or Defence of is / er Dissertation or .

i. e dissertation internal evaluation s all be on t e basis of continuous revie by

t e DDRC on t e rogress of t e dissertation or .

8. ATTENDANCE REGULATIONS & CONDONATION:

i. student s all be eligible to a ear for end semester e aminations, if ac uired a minimum of 75 of attendance in aggregate of all t e sub ects.

ii. Condonation of s ortage of attendance in aggregate u to 10 on medical grounds 65 and above and belo 75 in eac semester may be granted by t e College cademic Committee C C . Ho ever, t e sub ect of granting is urely at t e discretion of t e College cademic Committee or Com etent

ut ority.

iii. Student ill not be romoted to t e ne t semester unless e/s e satisfies t e

attendance re uirement of t e resent semester as a licable. ey may see re-admission for t at semester as and en offered ne t.

iv. Due eig tage in eac of t e sub ects s all be given to t e attendance. Mar s

not e ceeding 5 s all be given to all suc candidates o satisfy t e follo ing

criteria

Table .2 Marks weightage for Attendance

Percentage of attendance Mar s

90 5

85 to 90 4

80 to 85 3

75 to 80 2

75 1

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 5

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

v. S ortage of ttendance belo 65 in aggregates all in no case be condoned.

vi. Students ose s ortage of attendance is not condoned in any semester are not eligible to ta e t eir end Semester e aminations of t at articular semester and t eir registration for e amination s all stand cancelled.

vii. sti ulated fee s all be ayable to ards condonation of s ortage of attendance.

viii. ttendance may also be condoned for t ose o artici ate in restigious

s orts, co- and e tra-curricular activities rovided t eir attendance is in t e minimum rescribed range for t e ur ose and recommended by t e concerned aut ority.

9. MINIMUM ACADEMIC REQUIREMENTS:

e follo ing academic re uirements ave to be satisfied in addition to t e attendance re uirements mentioned in Item o.8.

i students all be deemed to ave secured t e minimum academic re uirement in

a sub ect if e/s e secures a minimum of 40 of mar s e clusively in t e end semester e amination and a minimum aggregate of 50 of t e total mar s in t e end semester e amination and internal evaluation ta en toget er. 

ii student ill be romoted to second year, if e/s e secures t e minimum

attendance re uirement.

Students o fail to earn 88 credits as indicated in t e course it in

four academic years from t e year of t eir admission s all forfeit t eir seat in M. ec course and t eir admission s all stand cancelled.

10. COURSE PATTERN:

e entire course of study is of t o academic years. Eac academic year s all ave

t o semesters.

Student eligible to a ear for t e end e amination in a sub ect, but absent at it

or as failed in t e end e amination may a ear for t at sub ect only as and en it

is conducted.

ll admitted students are to study 4 electives during t eir course of t o year study at t e institute. e follo ing s all be t e rogramme of study of electives.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 6

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

ear Semester o. of electives

1 1 2

1 2 2

inal year o semesters is com letely reserved for Dissertation or and

ec nical seminar.

en a student is detained due to s ortage of attendance/any ot er reason, e

may be re-admitted en t e semester is offered after fulfillment of academic

regulations. ereas, t e academic regulations old good it t e regulations

e/s e first admitted.

11. AWARD OF GRADE:

fter a student as satisfied t e re uirement rescribed for t e com letion of t e rogramme and is eligible for t e a ard of M. ECH Degree e/s e s all be laced in one of t e follo ing four grades. e a ard of t e degree is on a grade oint of scale

4. e grade oints are a arded as follo s:

CGPA Division of Pass

3.0 irst Class it Distinction

2.0 and 3.0 irst division

1.6 and 2.0 Pass division

1.6 ail

ased on t e erformance of t e candidate, e follo ing s all be t e criteria for t e

a ard of letter grades at t e end of eac semester in t e sub ects in ic t e

candidate a eared for t e e amination

Mar s

Scored rades rade

Points

90 S 4.00

85 to 90 3.67

80 and 85 3.33

75 and 80 3.00

70 and 75 2.67

65 and 70 C2.33

60 and 65 C 2.00

55 and 60 D 1.67

50 and 55 E 1.33

50 0

Table .3. Marks,Grades and Grade Points

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 7

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

11.1 Calculation of Semester Grade Points Average (SGPA)

e erformance of eac student at t e end oft e eac semester is indicated in terms of P . e S P is calculated as belo :

 

ere CR Credits of a course

GP rade oints a arded for a course

SGPA (Semester Grade Point Average) is calculated for t e candidates

o assed all t e courses in t at semester.

11.2 Calculation of Cumulative Grade Point Average (CGPA) for Entire Programme.

e C P is calculated as belo :

 

for entire rogramme

ere CR Credits of a course

GP rade oints a arded for a course

12. MINIMUM INSTRUCTION DAYS:

e minimum instruction for eac semesters all be 90 instruction days e cluding e amination days.

13. GENERAL:

a. ere t e ords e im is , occur in t e regulations, t ey include

s e , er , ers .

b. e academic regulations s ould be read as a ole for t e ur ose of any

inter retation.

c. In t e case of any doubt or ambiguity in t e inter retation of t e above rules, t e decision of t e Director is final.

d. e Institute may c ange or amend t e academic regulations or syllabi at any time and t e c anges or amendments made s all be a licable to all t e students it effect from t e dates notified.

14. TRANSITORY REGULATIONS

14.1 candidate, o is detained or discontinued in t e year/semester, on

readmission s all be re uired to do all t e courses in t e curriculum rescribed for

suc batc of students in ic t e student oins subse uently. Ho ever, e em tion

ill be given to t ose candidates o ave already assed in suc courses, ic

e/s e ad assed in t e earlier semester s e/s e as originally admitted into.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 8

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

14.2 student o is follo ing t e , a inada curriculum, detained due to

lac of academics/attendance at t e end of a semester of an academic year, s all oin it t e autonomous batc at t e a ro riate semester. Suc candidates s all be re uired to ass in all t e courses in t e rogramme rescribed by concerned OS for suc batc of students, to be eligible for t e a ard of degree. Ho ever,

e em tion ill be given to all t ose courses of t e semesters of t e batc , ic

t e candidate oins no , ic e/s e ad assed earlier. e student as to clear

all is bac log sub ects by a earing t e su lementary e aminations, conducted by

, a inada and utonomous stream for t e a ard of Degree. e mar s

secured by t e students in - a inada attern ill be converted in to

a ro riate grade oints as er t e autonomous grading system and t e class ill be

a arded based on t e academic erformance of a student in t e entire 2 years as er t e guidelines of autonomous Pattern.

14.3 e concerned oard of Studies s all give t e guidelines regarding Course E uivalence and Course E em tions from time to time.

15. COURSE CODE AND COURSE NUMBERING SCHEME:

Course umbers are denoted by 5 digit uni ue al a numeric c aracters. irst t o digits are M , t at describes t e Course name i.e. Master of ec nology.

Third digit re resents semester of offering as mentioned in able o. 4.

THIRD DIGIT DESCRIPTION

1 irst Semester

2 Second Semester

3 ird Semester

4 ourt Semester

Table 4: Third digit description 

Fourth digit re resents course ty e, as er able o. 5

O R H DI I DESCRIP IO

0 eory course

5 ab course/ot er t an t eory sub ect

Table 5 : Course type description 

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 9

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 10

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

c e follo ing acts of omission and/or commission s all constitute gross violation of

t e code of conduct and are liable to invo e disci linary measures it regard to ragging.

i. ac of courtesy and decorum indecent be avior any ere it in or outside t e cam us.

ii. illful damage or distribution of alco olic drin s or any ind of narcotics tot e fello students/citi ens.

d Possession, consum tion or distribution of alco olic drin s or any ind of narcotics or allucinogenic drugs.

e Mutilation or unaut ori ed ossession of library boo s.

f oisy and unseemly be avior, disturbing studies of fello students.

g Hac ing in com uter systems suc as entering into ot er erson s areas it out

rior ermission, mani ulation and/or damage of com uter ard are and soft are or any ot er cyber crime etc.

sage of camera cell ones in t e cam us.

i Plagiarism of any nature.

ny ot er act of gross indisci line as decided by t e academic council from time to time.

Commensurate it t e gravity of offense, t e unis ment may be re rimand, fine, e ulsion from t e institute / ostel, debarment from an e amination, disallo ing t e use of certain facilities of t e Institute, rustication for a s ecified eriod or even outrig t e ulsion from t e Institute, or even anding over t e case to a ro riate

la enforcement aut orities or t e udiciary, as re uired by t e circumstances.

l or an offence committed in i a ostel ii a de artment or in a class room and iii else ere, t e c ief arden, t e Head of t e De artment and t e rinci al res ectively, s all ave t e aut ority to re rimand or im ose fine.

m Cases of ado tion of unfair means and/or any mal ractice in an e amination s all be re orted to t e rinci al for ta ing a ro riate action.

n ll cases of serious offence, ossibly re uiring unis ment ot er t an re rimand,

s all be re orted to t e cademic council.

o e Institute evel Standing Disci linary ction Committee constituted by t e

academic council, s all be t e aut ority to investigate t e details of t e offence, and recommend disci linary action based on t e nature and e tent of t e offence committed.

e Princi al s all deal it any academic roblem, ic is not covered under

t ese rules and regulations, in consultation it t e Programmes Committee in an a ro riate manner, and subse uently suc actions s all be laced before t e

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 11

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

academic council for ratification. ny emergency modification of regulation, a roved by t e academic council earlier, s all be re orted to t e academic council for ratification.

“Grievance and Redressal Committee” (General) constituted by t e rinci al

s all deal it all grievances ertaining to t e academic/administrative /disci linary matters.

r ll t e students must abide by t e code and conduct rules of t e college.

21. MALPRACTICES

a e Princi al s all refer t e cases of mal ractices in internal assessment tests and

Semester-End E aminations, to a Mal ractice En uiry Committee, constituted by im/ er for t e ur ose. Suc committee s all follo t e a roved scales of

unis ment. e Princi al s all ta e necessary action, against t e erring students

based on t e recommendations of t e committee.

b ny action on t e art of candidate at an e amination trying to get undue

advantage in t e erformance at e aminations or trying to el anot er, or derive

t e same t roug unfair means is unis able according to t e rovisions

contained ereunder. e involvement of t e Staff, o are in c arge of

conducting e aminations, valuing e amination a ers and re aring/ ee ing records of documents relating to t e e aminations in suc acts inclusive of

roviding incorrect or misleading information t at infringe u on t e course of natural ustice to one and all concerned at t e e amination s all be vie ed seriously and recommended for a ard of a ro riate unis ment after t oroug en uiry.

22. AWARD OF RANK

e ran s all be a arded based on t e follo ing:

22.1 Only suc candidates o ass t e inal year e amination at t e

end of t e t ird academic year after admission as regular final year

students along it t e ot ers in t eir batc and become eligible for

t e a ard of t e Degree s all be eligible for t e a ard of ran .

Candidates, o loose one or more years of study for any reason

atsoever are not eligible for t e a ard of ran .

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 12

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

22.2 Ran s s all be a arded in eac branc of study for t e to five

students a earing for t e Regular e ternal E aminations.

22.3 ard of ri es, sc olars i s, or any ot er Honors s all be based on

t e ran secured by a candidate, consistent it t e desire of t e

Donor, erever a licable.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 13

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

M.TECH(ECE – SYSTEMS AND SIGNAL PROCESSING) - COURSE STRUCTURE

I-SEMESTER

Code No. Name of the Course Scheme of Instruction Scheme of Examination Total credits

Periods per Week Maximum Marks

Lecture Tutorial Lab. Internal External

MEC101 dvanced Digital Signal Processing 4 -- 40 60 100 4

MEC102 ransform ec ni ues 4 -- 40 60 100 4

MEC103 V SI ec nology and Design 4 -- 40 60 100 4

MEC104 Microcontrollers or Embedded System Design 4 -- 40 60 100 4

MEC1051

MEC1052 ELECTIVE – I 4 - - 40 60 100 4

DSP Processers rc itecture

Image and Video Processing

MEC1061

MEC1062 ELECTIVE – II 4 - - 40 60 100 4

Radar Signal Processing

io Medical Signal Processing

MEC151 Seminar -- 3 50 -- 50 2

MEC152 dvanced Digital Signal Processing ab -- 3 40 60 100 2

TOTAL 24 - 6 330 420 750 28

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 14

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

II-SEMESTER

Code No. Name of the Course Scheme of Instruction Scheme of

Examination Total credits

Periods per Week Maximum Marks

Lecture Tutorial Lab Internal External

MEC201 da tive Signal Processing 4 -- -- 40 60 100 4

MEC202 S eec Processing 4 -- -- 40 60 100 4

MEC203 SOC rc itecture 4 -- -- 40 60 100 4

MEC204 Coding eory and ec ni ues 4 -- -- 40 60 100 4

MEC2051

MEC2052 ELECTIVE – III 4 -- -- 40 60 100 4

CP D P rc itectures nd lications

Design for estability

MEC2061

MEC2062 ELECTIVE – IV 4 -- -- 40 60 100 4

ireless Communication and et or s

V SI Signal Processing

MEC207 Seminar -- -- 3 50 -- 50 2

MEC208 dvanced EC D ab -- -- 3 40 60 100 2

TOTAL 24 -- 6 330 420 750 28

III & IV SEMESTERS

Code No. Name of the Course Scheme of Instruction Scheme of Examination Total credits

Periods per Week Maximum Marks

Lecture Tutorial Lab. Internal External

MEC351 ec nical Seminar -- -- 6 50 50 8

MEC352 Dissertation -- -- 15 50 150 200 24

TOTAL -- -- 21 100 150 250 32

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 15

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC101: ADVANCED DIGITAL SIGNAL PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

UNIT - I

Review of DFT, FFT, IIR Filters, FIR Filters, Multirate Signal Processing: Introduction, Decimation by a factor D, Inter olation by a factor I, Sam ling rate conversion by a rational factor I/D, Multistage Im lementation of Sam ling Rate Conversion, ilter design

Im lementation for sam ling rate conversion, lications of Multirate Signal Processing

UNIT - II 

Non-Parametric methods of Power Spectral Estimation: Estimation of s ectra from finite

duration observation of signals, on- arametric Met ods: artlett, elc lac man

u ey met ods, Com arison of all on-Parametric met ods

UNIT - III 

Parametric Methods of Power Spectrum Estimation: utocorrelation Its Pro erties,

Relation bet een auto correlation model arameters, R Models - ule- a er urg

Met ods, M RM models for o er s ectrum estimation.

UNIT - IV

Linear Prediction : or ard and ac ard inear Prediction or ard inear Prediction,

ac ard inear Prediction, O timum reflection coefficients for t e attice or ard and ac ard Predictors. Solution of t e ormal E uations: evinson Durbin lgorit m, Sc ur lgorit m. Pro erties of inear Prediction ilters

UNIT - V

Finite Word Length Effects: nalysis of finite ord lengt effects in i ed- oint DSP

systems i ed, loating Point rit metic DC uanti ation noise signal uality inite

ord lengt effect in IIR digital ilters inite ord-lengt effects in algorit ms.

TEXTBOOKS

1. Digital Signal Processing: Princi les, lgorit ms lications - . .Proa is

t

D. .Manolo is, 4 ed., PHI.

2. Discrete ime signal rocessing - lan V O en eim Ronald Sc affer, PHI.

3. DSP Pratical roac Emmanuel C.Ifeac er, arrie. .ervis,

2 ed., Pearson Education.

REFERENCES:

1. Modern s ectral Estimation : eory lication S. M . ay, 1988, PHI.

2. Multirate Systems and ilter an s P.P.Vaidyanat an Pearson Education

3. Digital Signal Processing S.Saliva anan, .Vallavara , C. nana riya, 2000, MH

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 16

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC102: TRANSFORM TECHNIQUES

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT – I

Review of Transforms: Signal s aces, conce t of convergence, Hilbert s aces for energy

signals, ourier basis, -failure of -need for time-fre uency analysis, s ectrogram lot 

ase s ace lot in time-fre uency lane, Continuous ,D , Discrete ourier Series

and ransforms, - ransform, relation bet een C -D , D -D S,D S-D ,

DC 1D 2D, als ,Hadamard, Haar, Slant, ,Hilbert ransforms definition,

ro erties and a lications

UNIT – II

CWT & MRA: ime-fre uency limitations, tiling of time-fre uency lane for S ,

Heisenberg uncertainty rinci le, S ort time ourier ransform S analysis, s ort

comings of S , eed for avelets- avelet asis- Conce t of Scale and its relation it fre uebcy , Continuous time avelet ransform E uation- Series E ansion using avelets 

C - eed for scaling unction- Multi resolution analysis, iling of time scale lane for

C .Im ortant avelets : Haar, Me ican Hat Meyer, S annon, Daubec ies.

UNIT – III 

Multirate Systems, Filter Banks and DWT.

asics of Decimation and Inter olation in time fre uency domains, o-c annel ilter

ban , Perfect Reconstruction Condition, Relation s i bet een ilter an s and avelet

basis, D ilter an s or Daubec ies avelet unction

UNIT – IV

Special Topics: avelet Pac et ransform Multidimensional avelets, i-ort ogonal basis¬-s lines, ifting Sc eme of avelet eneration, Multi avelets

UNIT – V

Applications of Transforms

Signal Denoising, Subband Coding of S eec and Music, Signal Com ression - se of

DC , D , , 2-D D , ractal Signal nalysis.

TEXT BOOKS

1. undamentals of avelets- eory, lgorit ms and lications , aideva C

os ami, ndre C an, o n iley Sons, Inc, Singa ore, 1999.

2. avelet ransforms-Introduction t eory and a lications-Rag uveer M.Rao and it S. o ardi ar, Pearson edu, sia, e Del i, 2003.

3. Insig t into avelets from eory to ractice , Soman. .P, Ramac andran. .I, Printice Hall India, irst Edition, 2004.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 17

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

REFERENCES

1. avelets and sub-band coding , Vetterli M. ovacevic, P I, 1995.

2. Introduction to avelets and avelet ransforms , C. Sydney urrus, PHI, irst Edition, 1997.

3. avelet our of Signal Processing , Ste en . Mallat,. cademic Press, Second Edition,

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 18

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC103: VLSI TECHNOLOGY AND DESIGN

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Review of Microelectronics and Introduction to MOS Technologies: MOS, CMOS, iCMOS ec nology, rends nd Pro ections. asic Electrical Pro erties of MOS, CMOS BiCMOS Circuits: Ids-Vds relationships, Threshold Voltage Vt, Gm, Gds and ωo, Pass

ransistor, MOS, CMOS i CMOS Inverters, u/ d, MOS ransistor circuit model,

atc -u in CMOS circuits.

UNIT - II 

LAYOUT DESIGN AND TOOLS: ransistor structures, ires and Vias, Scalable Design rules, ayout Design tools.

LOGIC GATES & LAYOUTS: Static Com lementary ates, S itc ogic, lternative ate

circuits, o o er gates, Resistive and Inductive interconnect delays.

UNIT - III 

COMBINATIONAL LOGIC NETWORKS: ayouts, Simulation, et or delay, Interconnect design, Po er o timi ation, S itc logic net or s, ate and et or testing.

UNIT - IV

SEQUENTIAL SYSTEMS: Memory cells and rrays, Cloc ing disci lines, Design, Po er o timi ation, Design validation and testing.

UNIT – V

FLOOR PLANNING & ARCHITECTURE DESIGN: loor lanning met ods, off-c i

connections, Hig -level synt esis, rc itecture for lo o er, SOCs and Embedded CP s,

rc itecture testing.

TEXT BOOKS

1. Essentials of V SI Circuits and Systems, . Es rag ian Es rag ian. D, .Puc nell, 2005, PHI.

2. Modern V SI Design - ayne olf, 3rd ed., 1997, Pearson Education.

REFERENCES

nd

1. Princi als of CMOS V SI Design .H.E este, .Es rag ian, 2 ed., disson

esley.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 19

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC104: MICROCONTROLLERS FOR EMBEDDED SYSTEM DESIGN

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT – I

Introduction to Embedded Systems Overvie of Embedded Systems, Processor Embedded into a system, Embedded Hard are nits and Devices in system, Embedded Soft are, Com le System Design, Design Process in Embedded System, ormali ation of System Design, Classification of Embedded Systems.

UNIT – II

Microcontrollers and Processor rc itecture Interfacing 8051 rc itecture, In ut/Out ut

Ports and Circuits, E ternal Memory, Counters and imers, PIC Controllers. Interfacing Processor 8051, PIC , Memory Interfacing, I/O Devices, Memory Controller and Memory arbitration Sc emes.

UNIT - III

Embedded RISC Processors Embedded System-on C i Processor PSOC

Programmable System-on-C i arc itectures, Continuous imer bloc s, S itc ed

Ca acitor bloc s, I/O bloc s, Digital bloc s, Programming of PSOC, Embedded RISC

Processor arc itecture RM Processor arc itecture, Register Set, Modes of o eration and

overvie of Instructions

UNIT - IV

Interru ts Device Drivers E ce tions and Interru t andling Sc emes Conte t Periods

for Conte t S itc ing, Deadline interru t latency. Device driver using Interru t Service

Routine, Serial ort Device Driver, Device drivers for Internal Programmable timing devices

UNIT – V

et or Protocols Serial communication rotocols, Et ernet Protocol, SDM , C annel IDM , E ternal us Interface

TEXT BOOKS

nd

1. Embedded Systems - rc itecture Programming and Design Ra amal, 2 ed., 2008, MH.

2. PIC Microcontroller and Embedded Systems Mu ammad li Ma idi, Rolin

D.Mc inaly, Danny Causy PE.

3. Designers uide to t e Cy ress PSOC Robert s y, 2005, Elsevier.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 20

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

REFERENCES

1. Embedded Microcom uter Systems, Real ime Interfacing onat an . Valvano roo es / Cole, 1999, omas earning.

2. RM Systems Develo ers uides-Design O timi ing System Soft are - ndre . Sloss, Dominic Symes, C ris rig t, 2004, Elsevier.

3. Designing it PIC Microcontrollers- o n . Peatman, 1998, PH Inc.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 21

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC1051: DIGITAL SIGNAL PROCESSORS AND ARCHITECTURE

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT – I

INTRODUCTION TO DIGITAL SIGNAL PROCESING

Introduction, Digital signal- rocessing system, e sam ling rocess, Discrete time

se uences. Discrete ourier ransform D and ast ourier ransform , inear

time-invariant systems, Digital filters, Decimation and inter olation, nalysis and Design tool

for DSP Systems M , DSP using M .

COMPUTATIONAL ACCURACY IN DSP IMPLEMENTATIONS

umber formats for signals and coefficients in DSP systems, Dynamic Range and Precision,

Sources of error in DSP im lementations, /D Conversion errors, DSP Com utational errors, D/ Conversion Errors, Com ensating filter.

UNIT - II 

ARCHITECTURES FOR PROGRAMMABLE DSP DEVICES

asic rc itectural features, DSP Com utational uilding loc s, us rc itecture and

Memory, Data ddressing Ca abilities, ddress eneration nit, Programmability and Program E ecution, S eed Issues, eatures for E ternal interfacing.

UNIT - III 

EXECUTION CONTROL AND PIPELINING

Hard are loo ing, Interru ts, Stac s, Relative ranc su ort, Pi elining and Performance, Pi eline De t , Interloc ing, ranc ing effects, Interru t effects, Pi eline Programming models.

PROGRAMMABLE DIGITAL SIGNAL PROCESSORS

Commercial Digital signal- rocessing Devices, Data ddressing modes of MS320C54

DSPs, Data ddressing modes of MS320C54 Processors, Memory s ace of

MS320C54 Processors, Program Control, MS320C54 instructions and

Programming, On-C i Peri erals, Interru ts of MS320C54 rocessors, Pi eline

O eration of MS320C54 Processors.

UNIT - IV

IMPLEMENTATIONS OF BASIC DSP ALGORITHMS

e Q-notation, IR ilters, IIR ilters, Inter olation ilters, Decimation ilters, PID

Controller, da tive ilters, 2-D Signal Processing.

IMPLEMENTATION OF FFT ALGORITHMS

n lgorit m for D Com utation, utterfly Com utation, Overflo and scaling, it 

Reversed inde generation, n 8-Point im lementation on t e MS320C54 ,

Com utation of t e signal s ectrum.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 22

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

UNIT - V

INTERFACING MEMORY AND I/O PERIPHERALS TO PROGRAMMABLE DSP DEVICES Memory s ace organi ation, E ternal bus interfacing signals, Memory interface, Parallel I/O interface, Programmed I/O, Interru ts and I/O, Direct memory access DM . Multic annel buffered serial ort Mc SP , Mc SP Programming, a CODEC interface circuit, CODEC rogramming, CODEC-DSP interface e am le.

TEXT BOOKS

1. Digital Signal Processing vtar Sing and S. Srinivasan, omson Publications, 2004.

2. DSP Processor undamentals, rc itectures eatures a sley et al. 2000, S. C and Co.

REFERENCES

1. Digital Signal Processors, rc itecture, Programming and lications . Ven ataramani and M. as ar, 2002, MH.

2. Digital Signal Processing onat am Stein, 2005, o n iley.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 23

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC1052: IMAGE AND VIDEO PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Fundamentals of Image Processing and Image Transforms

asic ste s of Image Processing System Sam ling and Quanti ation of an image asic

relations i bet een i els Image ransforms: 2 D- Discrete ourier ransform, Discrete Cosine ransform DC ,avelet ransforms: Continuous avelet ransform, Discrete avelet ransforms.

UNIT - II 

Image Processing Techniques – Image Enhancement Spatial domain methods:

Histogram rocessing, undamentals of S atial filtering, Smoot ing s atial filters,

S ar ening s atial filters. re uency domain met ods: asics of filtering in fre uency domain, image smoot ing, image s ar ening, Selective filtering.

Image Segmentation

Segmentation conce ts, Point, ine and Edge Detection, res olding, Region ased

segmentation.

UNIT - III 

Image Compression

Image com ression fundamentals - Coding Redundancy, S atial and em oral redundancy,

Com ression models: ossy ossless, Huffman coding, rit metic coding, coding,

Run lengt coding, it lane coding, ransform coding, Predictive coding, avelet coding, PE Standards.

UNIT - IV

Basic steps of Video Processing

nalog Video, Digital Video. ime-Varying Image ormation models: ree-Dimensional

Motion Models, eometric Image ormation, P otometric Image ormation, Sam ling of Video signals, iltering o erations.

UNIT - V

2-D Motion Estimation

O tical flo , eneral Met odologies, Pi el ased Motion Esimation, loc - Matc ing

lgorit m, Mes based Motion Estimation, lobal Motion Estimation, Region based Motion

Estimation, Multi resolution motion estimation, aveform based coding, loc based

transform coding, Predictive coding, lication of motion estimation in Video coding.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 24

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

TEXT BOOKS

rd

1. Digital Image Processing on ale e and oods, 3 ed., Pearson.

2. Video rocessing and communication ao ang, oemOstermann and

st

a uin ang. 1 Ed., PH Int.

REFRENCES

1. Digital Video Processing M. e al, Prentice Hall International

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 25

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC1061: RADAR SIGNAL PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I 

Introduction Radar loc Diagram, Radar E uation, Information vailable from Radar

Ec o. Revie of Radar Range Performance eneral Radar Range E uation, Radar

Detection it oise amming, eacon and Re eater E uations, istatic Radar.Matc ed

ilter Receiver Im ulse Res onse, re uency Res onse C aracteristic and its Derivation, Matc ed ilter and Correlation unction, Correlation Detection and Cross-Correlation

Receiver. Efficiency of on-Matc ed ilters, Matc ed ilter for on- ite oise.

UNIT - II 

Detection of Radar Signals in oise: Detection Criteria eyman-Pearson Observer,

i eli ood-Ratio Receiver, Inverse Probability Receiver, Se uential Observer. Detectors

Envelo e Detector, ogarit mic Detector, I/Q Detector. utomatic Detection -C R

Receiver, Cell veraging C R Receiver, C R oss, C R ses in Radar. Radar Signal Management Sc ematics, Com onent Parts, Resources and Constraints.

UNIT - III 

aveform Selection 3, 2 : Radar mbiguity unction and mbiguity Diagram Princi les

and Pro erties S ecific Cases Ideal Case, Single Pulse of Sine ave, Periodic Pulse

rain, Single inear M Pulse, oiseli e aveforms. aveform Design

Re uirements.O timum aveforms for Detection in Clutter, amily of Radar aveforms.

UNIT - IV

Pulse Com ression in Radar Signals: Introduction, Significance, y es. inear M Pulse

Com ression loc Diagram, C aracteristics, Reduction of ime Sidelobes, Stretc

ec ni ues, eneration and Decoding of M aveforms loc Sc ematic and

C aracteristics of Passive System, Digital Com ression, S Pulse Com ression.

UNIT - V

P ase Coding ec ni ues: Princi les, inary P ase Coding, ar er Codes, Ma imal engt Se uences M S/ RS/P , loc Diagram of a P ase Coded C Radar. Poly P ase Codes: ran Codes, Costas Codes, on- inear M Pulse Com ression, Do ler olerant PC aveforms S ort Pulse, inear Period Modulation PM/H M . Sidelobe Reduction for P ase Coded PC Signals.

TEXT BOOKS

nd

1. Radar Handboo - M.I. S olni, 2 ed., 1991, Mc ra Hill.

2. Radar Design Princi les : Signal Processing and e Environment - red E.

nd

at anson,2 ed., 1999, PHI.

rd

3. Introduction to Radar Systems - M.I. S olni, 3 ed., 2001, MH.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 26

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

REFERENCES

1. Radar Princi les - Peyton . Peebles, r., 2004, o n iley.

2. Radar Signal Processing and da tive Systems - R. it berg, 1999, rtec House.

st

3. Radar Design Princi les - .E. at anson, 1 ed., 1969, Mc ra Hill.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 27

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC1062: BIO-MEDICAL SIGNAL PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Discrete and continuous Random variables, Probability distribution and density functions. aussian and Rayleig density functions, Correlation bet een random variables.

Stationary random rocess, Ergodicity, Po er s ectral density and autocorrelation function of random rocesses. oise o er s ectral density analysis, oise band idt , noise figure of systems.

UNIT- II

Data Compression Techniques: ossy and ossless data reduction lgorit ms. EC data

com ression using urning oint, EC, COR ES, Huffman coding, vector uantisation,

DC and t e transform.

UNIT- III 

Cardiological Signal Processing: Pre- rocessing. QRS Detection Met ods.R yt m

analysis. rr yt mia Detection lgorit ms. utomated EC nalysis.EC Pattern

Recognition.Heart rate variability analysis. da tive oise Cancelling: Princi les of da tive oise Cancelling. da tive oise Cancelling it t e MS da tation lgorit m. oise Cancelling Met od to En ance EC Monitoring. etal EC Monitoring.

UNIT- IV

Signal Averaging, polishing mean and trend removal, Prony s met od, Prony s Met od

based on t e east S uares Estimate, inear rediction. ule al er e uations,

nalysis of Evo ed Potentials.

UNIT- V

Neurological Signal Processing: Modeling of EE Signals. Detection of s i es and s indles Detection of l a, eta and amma aves. uto Regressive .R. modeling of sei ure EE . Slee Stage analysis. Inverse iltering. east s uares and olynomial modeling.

TEXT BOOKS

t

1. Probability, Random Variables Random Signal Princi les Peyton . Peebles, 4 ed., 2009, MH.

2. iomedical Signal Processing- Princi les and ec ni ues - D.C.Reddy, 2005, MH.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 28

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

REFERENCES

1. Digital io signal Processing - eit unat R, 1991, Elsevier.

2. iomedical Signal Processing - ay M, IEEE Press.

3. iomedical Signal Processing -Vol. I ime re uency nalysis - Co en. , 1986, CRC Press.

4. iomedical digital Signal Processing : C- anguage E eriments and aboratory E eriments, illis . om ins, PHI.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 29

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC201: ADAPTIVE SIGNAL PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

UNIT – I

Introduction to Adaptive Systems Adaptive Systems: Definitions, C aracteristics,

lications, E am le of an da tive System. e da tive inear Combiner - Descri tion,

eig t Vectors, Desired Res onse Performance function - radient Mean S uare Error.

UNIT – II

Development of Adaptive Filter Theory & Searching the Performance surface:

Introduction to iltering - Smoot ing and Prediction inear O timum iltering, Problem

statement, Princi le of Ort ogonality - Minimum Mean S uare Error, iener- Ho f

e uations, Error Performance - Minimum Mean S uare Error.

Searching the performance surface Met ods Ideas of radient Searc met ods -

radient Searc ing lgorit m its Solution - Stability Rate of convergence - earning

Curves.

UNIT - III 

Steepest Descent Algorithms

radient Searc by e tons Met od, Met od of Stee est Descent, Com arison of

earning Curves.

UNIT – IV

LMS Algorithm & Applications

Overvie - MS da tation algorit ms, Stability Performance analysis of MS lgorit ms -

MS radient Stoc astic algorit ms - Convergence of MS algorit m. Applications:

oise cancellation Cancellation of Ec oes in long distance tele one circuits, da tive eam forming.

UNIT – V

Kalman filtering:

Introduction - Recursive Mean S uare Estimation Random variables, Statement of alman

filtering roblem iltering -Initial conditions - Variants of alman filtering E tend alman

filtering.

TEXT BOOKS

1. da tive Signal Processing - ernard idro , Samuel D.Strearns, 2005, PE.

2. da tive ilter eory - Simon Hay in-, 4 ed., 2002,PE sia.

REFERENCES 

1. O timum signal rocessing: n introduction - So ocles. .Orfamadis, 2 ed., 1988, Mc ra -Hill, e yor .

2. da tive signal rocessing- eory and lications, S. omas le ander, 1986,

S ringer Verlag.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 30

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC202: SPEECH PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Fundamentals of Digital Speech Processing: natomy P ysiology of S eec Organs,

e rocess of S eec Production, e coustic eory of S eec Production, Digital

models fors eec signals.

UNIT - II 

Time Domain Models for Speech Processing

Introduction- indo considerations, S ort time energy and average magnitude S ort time

average ero crossing rate ,S eec vs. silence discrimination using energy and ero

crossing, Pitc eriod estimation using a arallel rocessing a roac , e s ort time

autocorrelation function, e s ort time average magnitude difference function, Pitc eriod

estimation using t e autocorrelation function.

UNIT – III 

Linear predictive coding (LPC) analysis

asic rinci les of inear Predictive nalysis: e utocorrelation Met od, e Covariance

Met od,Solution of c E uations: C oles y Decom osition Solution for Covariance

Met od, Durbin s Recursive Solution for t e utoCorrelation E uations, Com arision

bet een t e Met ods of Solution of t e PC nalysis E uations, lications of PC

Parameters: Pitc Detection using PC Parameters, ormant nalysis using PC Parameters.

Homomorphic Speech Processing

Introduction, Homomor ic Systems for Convolution: Pro erties of t e Com le Ce strum, Com utational Considerations, e Com le Ce strum of S eec , Pitc Detection, ormant Estimation, e Homomor icVocoder.

UNIT - IV

Speech enhancement: - ature of interfering sounds, S eec en ancment tec ni ues:

Single Micro one roac : s ectral substraction, En ancement by re-synt esis, Comb

filter, iener filter, Multimicro one roac .

Automatic speech recognition- asic attern recognition a roac es, Parametric

re resention of s eec , Evaluating t e similarity of s eec atterns, Isolated digit

Recognition System,.Contineous digit Recognition System

UNIT - V

Hidden Markov Model (HMM) for Speech

Hidden mar ov model HMM for s eec recognition, Viterbialgorit m, raining and testing

using HMMS, da ting to variability in s eec D , anguage models.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 31

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

Speaker recognition

Recognition tec ni ues, eatures t at distinguis s ea ers, S ea er Recognition Systems:

S ea er Verification System , S ea er Identification System.

TEXT BOOKS

1. Digital rocessing of s eec signals - .R Rabiner and S. .Sc afer. Pearson Education.

nd

2. S eec Communications: Human Mac ine - Douglas O S aug nessy, 2 ed., IEEE Press.

3. Digital rocessing of s eec signals. .R Rabinar and R Sc afer,1978, PHI.

REFERENCES

1. Discrete ime S eec Signal Processing : rinci les and Practice - omas . Quateri 1 ed., PE.

2. S eec udio Signal Processing- en old elson Morgan, 1 ed., iley.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 32

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC203: SOC ARCHITECTURE

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Introduction to Processor Design:

bstraction in Hard are Design, M O a sim le rocessor , Processor design trade off,

Design for lo o er consum tion.

ARM Processor as System-on-Chip: corn RISC Mac ine rc itecture in eritance

RM rogramming model RM develo ment tools 3 and 5 stage i eline RM

organi ation RM instruction e ecution and im lementation RM Co- rocessor interface

UNIT - II 

ARM Assembly Language Programming:

RM instruction ty es data transfer, data rocessing and control flo instructions RM

instruction set Co- rocessor instructions.

Architectural Support for High Level Language: Data ty es abstraction in Soft are

design E ressions oo s unctions and Procedures Conditional Statements se

of Memory

UNIT - III 

Memory Hierarchy: Memory si e and s eed On-c i memory Cac es Cac e design-an e am le memory management

UNIT - IV

Architectural Support for System Development: dvanced Microcontroller bus

arc itecture RM memory interface RM reference eri eral s ecification Hard are

system rototy ing tools rmulator Debug arc itecture

UNIT - V

Architectural Support for Operating System: n introduction to O erating Systems

RM system control co rocessor CP15 rotection unit registers RM rotection unit

CP15 MM registers RM MM rc itecture Sync roni ation Conte t S itc ing in ut

and out ut

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 33

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

TEXT BOOKS

nd

1. RM System on C i rc itecture Steve urber 2 ed., 2000, ddison esley

Professional.

st

2. Design of System on a C i : Devices and Com onents Ricardo Reis, 1 ed., 2004, S ringer

REFERENCES 

1. Co-Verification of Hard are and Soft are for RM System on C i Design

Embedded ec nology ason ndre s e nes, and CDROM System on

C i Verification Met odologies and ec ni ues Pra as Ras in ar, Peter

Paterson and eena Sing , 2001, lu er cademic Publis ers.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 34

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC204 : CODING THEORY AND TECHNIQUES

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Coding for Reliable Digital Transmission and storage: Mat ematical model of

Information, ogarit mic Measure of Information, verage and Mutual Information and

Entro y, y es of Errors, Error Control Strategies. inear loc Codes: Introduction to

inear loc Codes, Syndrome and Error Detection, Minimum Distance of a loc code, Error-Detecting and Error-correcting Ca abilities of a loc code, Standard array and Syndrome Decoding, Probability of an undetected error for inear Codes over a SC,

Hamming Codes. lications of loc codes for Error control in data storage system

UNIT- II

Cyclic codes: Descri tion, enerator and Parity-c ec Matrices, Encoding, Syndrome Com utation and Error Detection, Decoding ,Cyclic Hamming Codes, S ortened cyclic codes, Error-tra ing decoding for cyclic codes, Ma ority logic decoding for cyclic codes.

UNIT- III 

Convolutional codes: Encoding of Convolutional Codes, Structural and Distance Pro erties, ma imum li eli ood decoding, Se uential decoding, Ma ority- logic decoding of

Convolution codes. lication of Viterbi Decoding and Se uential Decoding, lications of

Convolutional codes in RQ system.

UNIT- IV

urst Error-Correcting codes: Decoding of Signle- urst error Correcting Cyclic codes, Single- urst-Error-Correcting Cyclic codes, urst-Error-Correcting Convoulutional Codes,

ounds on urst Error-Correcting Ca ability, Interleaved Cyclic and Convolutional Codes , P ased- urst Error-Correcting Cyclic and Convolutional codes.

UNIT – V

BCH – Codes: CH code- Definition, Minimum distance and CH ounds, Decoding Procedure for CH Codes- Syndrome Com utation and Iterative lgorit ms, Error ocation Polynomials and umbers for single and double error correction

TEXT BOOKS

1. Error Control Coding- undamentals and lications S u in, Daniel .Costello, r,

Prentice Hall, Inc.

2. Error Correcting Coding eory-Man oung R ee- 1989, Mc ra -Hill Publis ing.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 35

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 36

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC2051 : CPLD & FPGA ARCHITECTURES AND APPLICATIONS

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

-------------------------------------------------------------------------------------------------------------- 

UNIT - I

Programmable logic : ROM, P , P P D, P eatures, rogramming and

a lications using com le rogrammable logic devices ltera series Ma 5000/7000

series and ltera E logic-10000 series CP D, MD s- CP D Mac 1to 5 , Cy res

SH 370 Device tec nology, attice P S s arc itectures 3000 series S eed

erformance and in system rogrammability.

UNIT - II 

FPGAs: ield Programmable gate arrays- ogic bloc s, routing arc itecture, design flo

tec nology ma ing for P s, Case studies itir C4000 ER s E

8000/10000 P s: ORC s O timi ed Reconfigurable Cell rray : C E s C -

1,2,3 and t eir s eed erformance

UNIT - III 

lternative reali ation for state mac ine c at suing micro rogramming lin ed state mac ine

one ot state mac ine, etrinetes for state mac ines-basic conce ts, ro erties, e tended

etrinetes for arallel controllers.

UNIT - IV

Digital front end digital design tools for P s SICs: sing mentor gra ics ED tool

P dvantage Design flo using P s

UNIT - V

Case studies of araller adder cell araller adder se uential circuits, counters, multi le ers, arellel controllers.

TEXT BOOKS

1. ield Programmable ate rray ec nology - S. rimberger, Edr, 1994, lu er cademic Publications.

2. ield Programmable ate rrays, o n V.Oldfield, Ric ard C Dore, iley

Publications.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 37

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

REFERENCES

1. Digital Design sing ield Programmable ate rray, P. .C an S. Mourad, 1994, Prentice Hall.

2. Digital System Design using Programmable ogic Devices Parag. . ala, 2003, SP.

3. ield rogrammable gate array, S. ro n, R. . rancis, .Rose, . .Vranesic, 2007, SP.

4. Digital Systems Design it P s and CP Ds Ian rout, 2009, Elsevier

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 38

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC2052 : DESIGN FOR TESTABILITY

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

UNIT – I

Introduction to Test and Design for Testability (DFT) Fundamentals

Modeling: Modeling Digital Circuits at ogic evel, register evel, and Structural Models. evels of Modeling. ogic Simulation: y es of Simulation, Delay Models, Element Evaluation, Ha ard Detection, ate evel Event Driven Simulation.

UNIT – II

Fault Modeling: ogic ault Models, ault Detection and Redundancy, ault e uivalence and ault ocation. Single Stuc and Multi le Stuc - ault Models, ault Simulation lications, eneral ec ni ues for Combinational Circuits.

UNIT – III 

esting for Single Stuc aults SS utomated est Pattern eneration P / for

SS s in Combinational and Se uential Circuits, unctional esting it S ecific ault

Models, Vector Simulation P Vectors, ormats, Com action and Com ression,

Selecting P ool.

UNIT – IV

Design for estability testability rade-off s ec ni ues, Scan rc itectures and esting, Controllability and bsorbability, eneric oundary Scan, ull Integrated Scan, Storage Cells foe Scan Design, oard level and System level a roac es, oundary Scans Standards, Com ression ec ni ues Different ec ni ues, Syndrome test and Signature analysis.

UNIT – V

uilt-in Self test IS IS Conce ts and est attern eneration. S ecific IS

rc itectures OCS , S MPS, C IS , R D, I O. rief ideas on some advanced

IS conce ts and design for self-test at board level. Memory IS M IS : Memory est rc itectures and ec ni ues, Introduction to Memory est, y es of Memories and Integration, Embedded Memory esting Model, Memory est re uirements for M IS , esting eatures.

TEXT BOOKS

1. Digital Systems esting and estable Design Miron bramovici, Melvin . reur, rt uD. riedman, o n iley Sons.

2. Design for est for Digital ICs Embedded Core Systems lfred Crouc , 2008,

PE.

3. Introduction to V SI esting Robrt. . eugate , Steven M.Mclntyre, Engle ood

Cliffs, 1988, Prentice Hall.

REFERENCES 

1. Essentials of Electronic esting M. . us nell, Vis ani.D. gar al, S ringer.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 39

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC2061 : WIRELESS COMMUNICATION AND NETWORKS

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

UNIT - I 

Wireless Communications & System Fundamentals: Introduction to ireless

communications systems, e am les, com arisons trends, Cellular conce ts-fre uency

reuse, strategies, interference system ca acity, truc ing grade of service, im roving

coverage ca acity in cellular systems.

UNIT - II 

Multiple Access Techniques for Wireless Communication: DM , DM , SSM

HM /CDM /Hybrid tec ni ues , SDM tec ni ue S a licable to ireless

communications .Pac et radio access- rotocols, CSM rotocols, reservation rotocols,

ca ture effect in ac et radio, ca acity of cellular systems.

UNIT - III 

Wireless Networking: Introduction, differences in ireless fi ed tele one net or s,

traffic routing in ireless net or s circuit s itc ing, ac et s itc ing .25 rotocol.

Wireless data services cellular digital ac et data CDPD , advanced radio data

information systems, R M mobile data RMD . Common c annel signaling CCS , ISD -

road band ISD M, Signaling System no .7 SS7 - rotocols, net or services art,

user art, signaling traffic, services erformance

UNIT - IV

Mobile IP and Wireless Application Protocol: Mobile IP O eration of mobile IP, Co 

located address, Registration, unneling, P rc itecture, overvie , M scri ts, P

service, P session rotocol, ireless transaction, ireless datagram rotocol. ireless

ec nology, Infrared s, S read s ectrum s, arro ban micro ave s,

IEEE 802 rotocol rc itecture, IEEE802 arc itecture and services, 802.11 medium access

control, 802.11 ysical layer.

UNIT - V

Mobile Data Networks: Introduction, Data oriented CDPD et or , PRS and ig er data

rates, S ort messaging service in SM, Mobile a lication rotocol. d- oc ireless

et or s: Cellular and d oc ireless net or s, a lications, M C rotocols, Routing,

Multicasting, rans ort layer Protocols, uality of service bro sing, de loyment

considerations, d oc ireless Internet

TEXT BOOKS

1. ireless Communication and et or ing illiam Stallings, 2003, PHI.

nd

2. ireless Communications, Princi les, Practice- eodore, S.Ra a ort, 2 Ed. 2002, PHI.

3. Princi les of ireless et or s ave Pa aven and P. ris na Murt y, 2002, Pearson Education ublis ers

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 40

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

REFERENCES

1. ireless Digital Communications amilo e er, 1999, PHI.

2. elecommunication System Engineering Roger .reeman, 4/ed., iley-Interscience, o n iley Sons, 2004.

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 41

 

LAKIREDDY BALI REDDY COLLEGE OF ENGINEERING (AUTONOMOUS), MYLAVARAM - 521230

MEC2062 : VLSI SIGNAL PROCESSING

Lecture : 4 Periods/week Internal Marks : 40

External Marks : 60

Credits : 4 External Examination : 3 Hrs

UNIT - I

Introduction to DSP: y ical DSP algorit ms, DSP algorit ms benefits, Re resentation of DSP algorit ms Pipelining and Parallel Processing: Introduction, Pi elining of IR Digital filters, Parallel Processing, Pi elining and Parallel Processing for o Po er Retiming:

Introduction Definitions and Pro erties Solving System of Ine ualities Retiming

ec ni ues

UNIT - II 

Folding and Unfolding: olding : Introduction - olding ransform - Register minimi ation ec ni ues Register minimi ation in folded arc itectures folding of multirate systems

Unfolding: Introduction n lgorit m for nfolding Pro erties of nfolding critical

Pat , nfolding and Retiming lications of nfolding

UNIT - III 

Systolic Architecture Design: Introduction Systolic rray Design Met odology IR

Systolic rrays Selection of Sc eduling Vector Matri Multi lication and 2D Systolic

rray Design Systolic Design for S ace Re resentations contain Delays

UNIT – IV

Fast Convolution: Introduction Coo - oom lgorit m inogard algorit m Iterated

Convolution Cyclic Convolution Design of ast Convolution algorit m by Ins ection

UNIT – V

Low Power Design: Scaling Vs Po er Consum tion Po er nalysis, Po er Reduction

tec ni ues Po er Estimation roac es Programmable DSP : Evaluation of

Programmable Digital Signal Processors, DSP Processors for Mobile and ireless

Communications, Processors for Multimedia Signal Processing

TEXT BOOKS

1. V SI Digital Signal Processing- System Design and Im lementation es ab .

Part i, 1998, iley Inter Science.

2. V SI and Modern Signal rocessing ung S. , H. .ile House, . ailat ,

1985, Prentice Hall.

REFERENCES

1. Design of nalog Digital V SI Circuits for elecommunications and Signal

Processing ose E. rance, annis sividis, 1994, Prentice Hall.

2. V SI Digital Signal Processing Medisetti V. ,1995, IEEE Press , S

M.TECH (SYSTEMS AND SIGNAL PROCESSING), A.Y.2010-2011 Page 42

 

 

 

 

Studyof

OFDM modulation

Eldo Mabiala, Mathias Coinchon

Eurecom institute

Supervisor: Karim Maouche

 

Date: december 1999

Semester ro ect

 

Table des matières

1 Introduction 1

1.1 Aim of project 1

1.2 Abstract 1

1.3 Plan of report 2

2 OFDM 3

2.1 Channel 3

2.1.1 Frequency selectivity 3

2.1.2 Delay spread 3

2.2 Qualitative description 4

2.2.1 The problem of wideband transmission on a single carrier 4

2.2.2 Multicarrier transmission 4

2.2.3 Orthogonality 5

2.2.4 The guard interval 6

2.2.5 Comparison with other schemes 6

2.3 Mathematical description 6

2.3.1 Multicarrier approach: 6

2.3.2 Inverse Fourier Transform: 7

2.3.3 Adding a guard interval: 7

2.3.4 Limitation of bandwidth: 8

2.3.5 Channel and receiver parts: 9

2.4 Synoptic of an OFDM transmission system 10

2.5 Problems in OFDM 11

2.5.1 Orthogonality: 11

2.5.2 Synchronization: 12

2.5.3 Peak Average Power Ratio (PAPR): 12

2.5.4 Effect of clipping in OFDM signals: 13

2.5.5 Phase noise: 13

2.5.6 Frequency error: 13

2.5.7 Conclusion 14

2.6 Design of an OFDM system 14

2.6.1 Bandwidth 14

2.6.2 Number of carriers 14

2.6.3 Guard interval 15

2.6.4 Modulation 15

2.6.5 Coding 15

3 Applications 16

3.1 General 16

3.2 DAB 16

3.2.1 What bandwidth to use for OFDM? 16

3.2.2 How many carriers could we use? 17

3.3 ADSL 18

3.3.1 HDSL 18

3.4 Hiperlan/2 19

3.4.1 Definition 19

3.4.2 What is the mission statement of H2GF? 19

3.4.3 Who are the founding members? 19

3.4.4 When will the H2GF be launched? 19

3.4.5 How does HiperLAN2 work? 19

 

3.4.6 Examples of use and scenarios: 20

3.5 Other applications 20

4 Simulation 21

4.1 Results 21

4.1.1 OFDM QAM transmission with only additive noise 21

4.1.2 OFDM QAM transmission with Raleigh channel model 22

4.1.3 OFDM BPSK transmission 22

4.1.4 BPSK transmission on one carrier 23

4.1.5 OFDM 4-PAM transmission 23

4.1.6 4-PAM Modulation one carrier 24

5 Conclusion 25

5.1 Future 25

5.1.1 Multi Carrier CDMA (MC-CDMA): 25

5.2 Conclusion 25

6 Annexes 27

6.1 Bibliography : 27

6.2 Glossary 28

6.3 Matlab code 28

6.3.1 OFDM simulator Error! Bookmark not defined.

6.3.2 Function QAM.m Error! Bookmark not defined.

6.3.3 Function gene_fad Error! Bookmark not defined.

6.3.4 Function mod_fad Error! Bookmark not defined.

6.3.5 Function propag.c Error! Bookmark not defined.

 

3

 

1 Introduction

General presentation andplan ofreport.

1.1 Aim of project

The aim of the project is the study of the Orthogonal Frequency Division Multiplexing modulation. The questions are what are the main principles that technology use ? What are the advantages and disadvantages in regard with the other schemes ? In which case, environment is it used ? what is the actual state-of-the art of it ? Where do we meet OFDM technology nowadays, and in the future ? What are today’s standards ? Is it a mature technology

1.2 Abstract

Orthogonal Frequency Division Multiplexing (OFDM) is a new digital modulation technique who consists of transmitting a data stream on several carriers instead of using only one carrier. The general concept has been first introduced in 1971, but it’s only in the last decade, with the development of Digital Signal Processors (DSP) that applications become visible.

OFDM is mainly used on wideband transmissions. We’ll see that OFDM is well suited for transmissions in frequency selective channels. Such a situation is met for example in multipath environments.

Applications nowadays find place in most of high data rate, wideband transmissions.

• In audio and television broadcasting (DAB and DVB) where usually high data rate transmissions are required in multipath environment. We’ll see too that OFDM permits the use of a single frequency network (SFN) where one broadcast multiplex can be sent at the same frequency with several transmitters.

• In wireline transmission: ADSL to transmit data at a very high rate on copper wire lines. The problem here is that characteristics of the line are not known and may change among users.

• In wireless LAN networks with HiperLAN/2 standard where transmissions occurs at very high data rate (~10 Mb/s) in indoors environment (strong multipath environment).

1

 

1.3 Plan of report

In the next chapter, we’ll explain the principle of OFDM modulation by introducing things intuitively. We’ll then see some points more in details by doing a mathematical description. Then, we’ll focus more on the practical aspects of OFDM : the implementation and the problems that modulation suffer from.

The third chapter is dedicated to the main applications of OFDM nowadays. We’ll focus particulary on Digital Audio Broadcasting (DAB) application since it covers most aspects of an OFDM transmission.

Also, in order to touch the problem of implementation, we’ll make an OFDM transmission simulator on MATLAB.

 

2

 

2 OFDM

Description of Orthogonal Frequency Division

Multiplexing modulation.

2.1 Channel

When doing radio transmission on high frequencies (VHF and higher) we are often confronted to a multipath environment. Such environment are found mostly in urban areas where buildings reflects waves.

2.1.1 Frequency selectivity

One problem of multipath is that the resultant of waves from different paths can be constructive or destructive depending on the position, so signal change over time when moving. You can experiment that when listening on FM radios while driving in a city: signal is cut by noise when moving.

Generally we speak of a frequency selective channel. Characteristics may change fast when moving. That’s why the channel is also time varying.

 

 

 

f

Typicalfrequency response of a channel sufferingfrom mutlipath propagation

The aim of equalization is to compensate the problems introduced by frequency selectivity.

2.1.2 Delay spread

Another effect that affect digital transmission, is that the signal coming from

different paths has different time delays depending on the length of path. A

3

 

consequence of that is memory of channel which cause interference between symbol received (ISI).

 

 

t

2.2 Qualitative description

2.2.1 The problem of wideband transmission on a single carrier

When transmitting wideband on frequency selective channels, equalisation must be performed in order to avoid intersymbol interference. Equalisation try to make the channel flat. In order to do that channel state information is needed. Training sequences have then to be transmitted periodically to estimate channel. Channel estimation is performed by several calculations and is then CPU time consuming. So when data rate is high and when characteristics of channel change rapidly CPU power needed is high and system become expensive.

Also is the symbol period very small compared to the channel memory. That cause strong Intersymbol Interference (ISI) and equalization is needed to correct that problem.

 

0)o f

2.2.2 Multicarrier transmission

The idea of multicarrier transmission is to divide bandwidth in several narrow band transmissions so that the channel looks flat on each carrier. The data stream to transmit is then split among the carriers instead of being tranmsitted on one carrier with large signal bandwidth. That’s what is meant by Frequency Division Multiplex (FDM). The advantage is that no or less complex equalization is needed.

4

 

The symbol period on each carrier become large and the effect of channel memory (length of channel impulse response) become less destructive on the symbols, so intersymbol interference (ISI) is reduced and less equalisation is needed.

Each carrier is modulated using any known modulation scheme like QAM and PSK schemes.

Reexp(0)0t+ gyp)}

 

 

2.2.3 Orthogonality

In a transmission system, we want the occupied bandwidth on the channel to be as small as possible. For that, in a multicarrier system, we try to set a minimum frequency space between carriers without having intercarrier interference (ICI). The minimum space is reached when carriers are orthogonal to each other, signal from each can have a small overlap on the other without causing interference. That’s what is meant by the « O » (Orthogonal) of OFDM. We’ll see further that the inverse fourier transform has that orthogonality property.

5

 

 

 

  f

1/ T

2.2.4 The guard interval

One way to avoid Inter-symbol interference is to set a small gap equal to the duration of delay spread between the symbols. So, each symbol does not affect the next one. We’ll also see later that this interval plays an important role in the implementation.

2.2.5 Comparison with other schemes

Simple schemes (QAM, ... ) as we have just seen, suffer from strongISI when channel memory increase and then strong equalisation is needed. Also the risk of signal loss is high in frequency selective fading channels

CDMA (or spread spectrum) schemes. In order to have high data rate transmissions, receiver has to compute correlations at a high rate and it is CPU time consuming.

2.3 Mathematical description

Let us consider the general problem of transmitting a signal S(t) over the time-varying channel c(t,T). The sampling version of S(t) is a bit stream S[n] that is demultiplexed onto N carriers.

2.3.1 Multicarrier approach:

The bit stream S[n] is shared onto all carriers, each one to be produced some data to transmit. Then the bit stream is divided in sub-stream (Xn [k0<_k<_N_1 called OFDM symbol, we use two indexes n being time index and k the sub-stream index (index used for carrier recognition). The carriers do not receive an amount of bit since the bit stream is pre modulated, the required amplitude and phase of the carrier is then calculated using BPSK, QPSK or QAM before demultiplexing. A variant of OFDM is COFDM (coded orthogonal frequency division multiplexing), where forward error coding is applied to the signal before transmission, to overcome errors due to lost carriers from frequency selective fading, channel noise and other effects of propagation. But the main focus of this project is OFDM.

6

 

Xn [0]

Xn [1]

Xn [ N-1]

 

 

2.3.2 Inverse Fourier Transform:

OFDM uses the available spectrum efficiently by spacing the channels much closer together. This is achieved by making all the carriers orthogonal to one another, preventing interference between the closely spaced carriers. To generate OFDM successfully the relationship between all carriers must be carefully to maintain the orthogonality of the carriers. For that, after choosing the spectrum required, we have to convert it back to its time domain signal using an Inverse Fourier Transform. In most applications, an Inverse Fast Fourier Transform is used, it performs the transformation very efficiently, and provides a simple way of ensuring the carrier signals produced are orthogonal.

Xn[0] Xn[1]

Xn[ N-1] IFFT xn[0]

xn [1]

xn [N-1]

1 N k

Xkexp j 2 l

1r

N

k_0 N

-1

n _ n

For l = 0, 1, ..., N-1

2.3.3 Adding a guard interval:

One of the most important properties of OFDM transmissions is the robustness against multipath delay spread. This is achieved by having a long symbol period, which minimizes the inter-symbol interference. The level of robustness, can in fact be increased even more by addition of a guard interval between transmitted symbols. The guard period allows time for multipath signals from the previous

7

 

symbol to die away before the information from the current symbol gathered. The most effective guard interval to use is a cyclic extension of the symbol. Why?

The Fast Fourier Transform that we will use at the receiver, transforms a cyclic time domain signal into its equivalent frequency spectrum. The signal xn k is not

necessary cyclic... Let form a cyclic signal of N+L-1 samples by repeating the last L-1 samples xnk at the beginning of the signal. This technique is called guard

interval by cyclic prefix. The number L of samples to repeat is taken more than the memory of the channel. So we the resulting cyclic prefixed signal is:

x N  L  1 ,..., x N 1 , x 0 , x 1 ,...x N 1

n n n n n

2.3.4 Limitation of bandwidth:

The frequencies used in OFDM are regularly spaced in the spectrum required

so, the time domain signal xn k has to be limited in the frequency domain. For

this reason we will make a convolution between the signal and a window function that has a spectrum limited exactly on the required one. (All window functions have a limited spectrum). We have chosen to use the raised cosine function to limit our signal spectrum and we will note it g* t. The signal to send over the channel is:

NL 1

 

 

l 0

 

 

8

 

 

Example of an OFDM spectrum

2.3.5 Channel and receiver parts:

Let consider the problem of transmitting the signal xnm

~ over the time 

varying linear channel ct, without additional noise. If we call cm the sampling version of the channel, then the output of obtained by the channel is:

NL 1

~ yn m xnlcm l 

~ 

l 0

m = 0,1,..., N+L-1

The receiver basically does the reverse operation to the transmitter. The signal

received is ynm

~ that has N+L-1 samples, before the demodulation, we have to drop the L last samples of the received signal, and then remove the guard period, in order to use correctly the Fourier Transform properties. Indeed the demodulation operation is a simple FFT according to the IFFT used as modulation. We need to find out N samples (one per carrier) as at the modulator input, let apply an FFT to the signal yn  m

~ .

Y~k  yl

~ exp 2jk

n n N

~    1

Y k

       

 

N 1 N L ~    

n l 

 x m c l  m exp  2  jk

n 

l  0  m  0  N

9

 

~ N  L  1 N 

       

 1 

x m c l m

~      l

Y k   exp 2

n n   jk

 m0 l0 N

~ NL1  m

Y k x m

~ jk ck

n   exp 2

 

 

n  

 m 0 N 



The first term of the multiplication above looks like a Fourier Transform expression. In fact, if we restrict the summation index from m = 0 to m = N-1 it is equivalent to drop the L last samples of the signal y~n m. So the signal produces

by the demodulator is:

~N 1

   

 m 

Y k x m

~ jk c  k

n   exp 2

  

 

n  

 m 0 N 

 

Yn k  xn kck

~

Yn k  Xn k Ck

~

A simple division by the channel frequency response gets back the transmitted

signal. This modulation does not need any equalization and the data samples are

then combined back to the same size as the original data.

2.4 Synoptic of an OFDM transmission system

Implementation of OFDM modulation nowadays is almost everytime performed digitally.

Here’s a model of an OFDM transmission system

 

• Data coming from the input are arranged into vectors with number ~

of components equal to the number of carriers. Each component N

is composed by a number of bits depending on the alphabet of the modulation scheme used on the next stage. For example, if we use a 1536 carriers system with BPSK, we’ll have vectors of 1536 component each one composed by 1 bit (BPSK is 2-ary).

• Each component (group of bits) is mapped into a complex symbol depending on the alphabet of the modulation scheme used. For example, with BPSK the alphabet is { -1 ; +1 }.

• In order to obtain real samples after IFFT, a 2*Number of carrier points IFFT is done with:

~

= X k

*,=1,...,N

k k

X = Re ( X ), X ~ = Im( X

0 0 N 0

• The Inverse Fast Fourier Transform algorithm (IFFT) is applied to the vector giving a real samples vector.

•The guard interval is added at the beginning of the vector by repeating the components of the end. Vectors are concatenated to form a time signal (parallel/serial conversion)

• Windowing the signal is necessary to limit the bandwidth. Most used window is the raised cosine.

• The signal is then passed trough the channel. Channel is modeled by a linear system with frequency response c(t) together with a source of additive Gaussian noise.

• At the reception, signal is rearranged again into vectors (serial/parallel conversion) and guard interval is dropped.

• Fast Fourier Transform (FFT) is computed in order to get back the complex vector of symbols.

2.5 Problems in OFDM

2.5.1 Orthogonality:

As seen above the fact to have several carriers is actually advantageous whenever they are mathematically orthogonal. So carriers orthogonality is a constrain that can leads to a wrong operation of OFDM systems if not respected. The orthogonality is provided by IFFT that a numerical manipulation, an error of computation could change lightly spacing between to consecutive carriers and break the orthogonality of the whole system. In this case OFDM loses all its efficiency, because the notion of orthogonality is an absolute one.

 

11

 

2.5.2 Synchronization:

One of the crucial problems in the receiver is to sample the incoming signal correctly. If the wrong sequence of samples is processed, the Fast Fourier Transform shall not correctly recover the received data on the carriers. The problem is more embarrassing when the receiver is switched on. There is therefore a need for acquiring timing lock. If the signal transmitted is really time domain periodic, as required for the FFT to be correctly applied, then the effect of the time displacement is to modify the phase of all carriers by a known amount. This is due to the time shift theorem in convolutional transform theory.

Symbols

 

Time

 

Sampling

 

However, the signal is not really repetitive, we have cheated and performed the mathematical transform as if it were repetitive, but then chosen different symbols and transmitted them one after the other. The effect of the time shift would then be not only to add the phase shift referred to above, but also to add some intersymbol interference with adjacent symbols. This interference could hardly degrade reception.

To avoid these problems, we decide to transmit more than one complete sequence of time samples in order to increase the tolerance in timing. It’s an additional data guard interval. It is built by repeating a set as long as channel memory of last samples taken in the original sequence. The longer the guard interval, the more rugged the system, but guard interval does not carry any useful information and its transmission leads to a penalty of power.

One technique used to obtain good synchronization is to add between each OFDM symbol a null (zero samples) symbol. This technique is used in DAB for time synchronization.

2.5.3 Peak Average Power Ratio (PAPR):

When the phase of different subcarriers add up to form large peaks, an important complication comes in OFDM systems. This problem is called Peak Average Power Ratio (PAPR) and it is defined for each OFDM signal on a time interval [n, n+Ts] by the following formula:

For continuous signals

xn

For sampled signals

 

max

Xn = E x k

k { [ ] 1

n 2

In OFDM systems PAPR can have very high values for certain input sets of sample (Xn [k] and overload non-linear characteristics of systems, causing inter 

modulations among different carriers and undesired out-of-band radiation.

Another main drawback of PAPR can be seen as quantization noise domination towards the performance of system. This domination can be excited by avoiding the clipping effect of the maximum level of the Digital to Analog Converter (DAC) that is set too high.

Various techniques are proposed to reduce PAPR in OFDM signals, but that reduction is not obvious because PAPR and SNR are closely linked. We will not expound those techniques in this paper, they can be found in reference [14].

2.5.4 Effect of clipping in OFDM signals:

When transmitted signals have high PAPRs, amplifiers may produce “clipping”. In some way, clipping can be regarded as peaks of the input signal being simply cut-off by amplifiers. Consequences of clipping are out-of-band radiation and inter symbol interference between subcarriers. In order to avoid these undesired effects that reduce OFDM performances, one has either to use amplifiers with dynamic range, or try to reduce PAPR. The first alternative is expensive, the second one is more often used.

2.5.5 Phase noise:

At the receiver, a local oscillator can add phase noise to an OFDM signal, for example. The phase noise could so have two effects those are: Common Phase Error (CPE) due to a rotation of the signal constellation and, Inter Carrier Interference (ICI), similar to additive Gaussian noise. The BBC R&D have made analysis of the effects of phase noise on an OFDM signal, this analysis shows that CPE arises simultaneously on all carriers. Indeed, the signal constellation within a given symbol is subject to the same rotation for all carriers and this effect can be corrected by using reference information within the same symbol. Unfortunately, ICI is more difficult to overcome, due to the additive noise, which is different for all carriers. This difference can be interpreted as a loss of orthogonality.

2.5.6 Frequency error:

An OFDM system can be subject to two types of frequency error. They are Frequency offset (as might be caused by the tolerance of the local oscillator frequency) and, Error in the receiver master clock frequency (which will cause the spacing of the demodulating carriers to be different from those transmitted).

13

 

Before to find solutions to those problems, the system designer needs to determine how much residual frequency error is permissible, and understand exactly how errors affect the received signal.

Both of these error situations have been analyzed so, a frequency offset affects most carriers equally, with the very edge carrier less affected. ICI resulting from a fixed absolute frequency offset increases with the number of carriers, if the system bandwidth is kept constant. About error in the receiver clock frequency, in absence of frequency offset, it affects carriers unequally (the center carrier suffers a little while the worst affected carrier lies close to, but not at, the edge).

2.5.7 Conclusion

At the reception, it is very important to distinguish the starting point of FFT to avoid wrong demodulation. And so synchronization has to be precise. It explains the use of special symbols (pilot) for synchronization in transmission.

Hardware design of transmitter and receiver is important because of high peak to average ratio which cause distorsions if dynamic range of amplifiers and converters is not high enough.

OFDM is very sensitive to carrier frequency offsets. Such offsets are mainly the cause of receiver local oscillators instability and doppler effect when mobile is moving.

2.6 Design of an OFDM system

2.6.1 Bandwidth

Occupied bandwidth is of course directly related to the data rate to transmit. However, the question is , what is the minimum bandwidth to take in order to obtain enough diversity and avoid the loss off all the signal in frequency selective fading environments. On the other hand much bandwidth means also much transmitting power. There is a tradeoff between bandwidth and transmitted power.

That optimal bandwidth is found by channel simulations and field test trials. In DAB, for example, a bandwidth of 1,5 Mhz is a good compromise for the type of propagation conditions that apply.

2.6.2 Number of carriers

We have seen that the greater the number of carriers, the greater the symbol period on each carrier and so lees equalization is needed and the greater the diversity offered by the system. However, with differential modulation, it is important that the channel not vary too much during one symbol period. This is not the case when the receiver is moving because of dopler effect and short term fading. Then a great number of carrier will limit the moving speed. This is another tradeoff of OFDM.

Another problem is the complexity in the implementation increase when carrier number increase because large FFT are needed.

14

 

To continue with the DAB example, 1536 carriers has been found to be a good compromise. That lead to a carrier spacing of 1kHz and a symbol period of 1ms. Moving speed of mobiles shouldn’t get over 160 km per hour.

2.6.3 Guard interval

The tradeoff of guard interval is to set it large enough to avoid intersymbol interference depending on the memory of channel and transmitter position spacing in a single frequency network. On the other hand, we want it to be as small as possible as it carries no information and can be seen as a spoil of bandwidth.

In wireless systems, a guard interval of 25% of symbol period is often met and seems to be a good compromise. That is the value taken for DAB, it allows a maximum distance of about 80 kilometers between transmitters

2.6.4 Modulation

The modulation scheme used on each carrier depends on the BER needs.

In DAB, QPSK is used but for higher order systems 16, 64 or 256 QAM is used.

2.6.5 Coding

Channel coding is very important in OFDM systems. When we speak of diversity,

that is possible because information is redundant among the carriers.

Coding associated with frequency (among carriers) and time interleaving make the system very robust in frequency selective fading.

In litterature, coded OFDM systems are sometime called COFDM.

 

15

 

3 Applications

Applications ofOFDM in digital communications

nowadays.

3.1 General

We have seen that OFDM is digital transmission technique well suited for wideband, high data rate transmissions. The main advantage is that less equalisation is necessary.

A consequence of that is that OFDM is not a very good solution for one to one communications with several users on a shared channels, because of the problem of frequency allocation. However on super high frequency bands (SHF) and Extremely high bands (EHF) where occupied bandwidth is not a great problem, OFDM may be a good solution for one to one communications. But, nowadays, OFDM is mainly used for one to many (broadcast) communications like radio or television broadcasting. That’s why we find OFDM on several new digital broadcasting systems such as DAB and DVB.

3.2 DAB

Digital Audio Broadcasting (DAB) is an international, standardized digital broadcasting system developed by the European EUREKA-147 Project. The system should completely replace, in the future, the well-known analog FM (Frequency Modulation) radio system on the 88-108 MHz frequency band.

The DAB system is digital and provides CD-like audio quality. DAB is much more robust to interferences and is well suited for mobile reception like in a car. New possibilities are available on receivers like for example multimedia features (image and texts). The transmission scheme used for DAB is OFDM modulation

DAB is being deployed around Europe and some other countries nowadays, receivers are still expensive like CD players in the beginning of eighties but we might expect to see a wide use of it in the next 5 years.

3.2.1 What bandwidth to use for OFDM?

The wider the bandwidth, the more probably that the system overcome the

correlation bandwidth of the channel.

Problem to overcome: Short delay echoes are the main problems to overcome, and as these are always present there is no hard bound. The narrower the bandwidth,

 

16

 

the more likely it is that the whole signal will be affected. There is a trade off between bandwidth and transmitter power.

Bandwidth: 7MHz few problems

2 MHz degradation of 1dB in performance at each point

<1.5MHz degradation starts to increase

200KHz used for FM sound, then the margin required would be an additional 6dB or so.

Trade off: 1.5MHz for the type of propagation conditions that apply to mobile and portable radio reception.

Bit-rate: On each carrier the modulation system used is QPSK, the carriers are separated by a gap of around 1/Ts, where Ts is symbol period. The maximum bit rate available is so 2bit/s/Hz of the bandwidth. This figure is reduced by the inefficiency (signal redundancy) of the guard interval, the null symbol and the error coding. For DAB, this brings the useful bit-rate down about 1 bit/s/Hz of the bandwidth.

Therefore a DAB system will provide just less than 1.5Mbit/s of useful data. This is considerably more than 256kbit/s that needed for high-quality stereophonic program, so the implication is that several broadcast programs will share the same multiplex.

3.2.2 How many carriers could we use?

About the number of carriers, the more there are, the greater is the resolution of the diversity offered by the system. For the differential demodulation to work properly, the multipath environment must change slowly from symbol by symbol. Thus, there is limit to the symbol and hence the number of carriers. For static reception, this is not a major problem. But for mobile reception, the motion of the vehicle leads to changes in the multipath environment. Over a symbol period, a vehicle moving at a velocity v m/s will travel vTs*f/c wavelengths. This is fd*Ts, where fd is the maximum Doppler shift. If this is to introduce negligible phase distortion, then the function fd must be small. A figure of fd<0.02 has been proposed as suitable for general use. To achieve high vehicle velocities (>160km/h) it has been necessary to adopt three modes of operation, each mode being suited to a difference part of VHF and UHF frequency bands. The main difference between modes is the symbol period and, as a direct consequence, the number of carriers.

If it will be only a question of single transmitter, then the significant echoes would all be relatively short. Surveys indicate that guard interval of the order of 10µs would satisfactory for the majority of location, in the UK for example. The use of several transmitters puts a limit on the minimum guard interval that should be used. The transmissions from areas that are some distance away can reach quite high levels on occasions of anomalous propagation. This has to be a compromise between a small number of high power transmitters spaced by about 50km, or a much larger number of low-power transmitters. Because the first option is likely to

 

17

 

be cheapest, the guard interval is to set about 250µs, equivalent to a maximum in transmission distance at the receiver.

The symbol period need not be directly related to the guard interval. It is a just question of how much of the symbol period is repeated in the guard interval. This is purely a matter of efficiency, as the power transmitted in the guard interval does not form a useful part of the data information in the receiver unless there are substantial echoes. To minimize the power loss by the system, it is desirable to keep the guard interval to as low a percentage as possible of the symbol period. In practice, a guard interval of the order of 25% of the symbol period has been found to be good compromise.

If we consider a guard interval of 250µs, this leads to a symbol period of 1ms and a carrier spacing of about 1 kHz. It means approximately 1500 carriers (W=1.5MHz) in the minimum bandwidth that would be desirable for one OFDM transmission. In practice, one uses 1536 carriers with a guard interval of 512. Then the channel receives symbols that have 2048 length so, the guard interval is the quarter of the symbol total length.

3.3 ADSL

Asymetric Digital Subscriber Line (ADSL) is a technique to transmit high data rates (up to 6 Mb/s downlink, 640kb/s uplink) on Subscriber Lines (telephone lines). Such lines consist of twisted copper wires. The idea is to use the full capacity of the line instead of using only 4 khz needed to transmit voice. Occupied bandwdth goes to 1.1 MHz.

The main problem is that the characteristics of the line change among users. They change with distance, presence of bridged taps in the line, neighbourhood of other lines, ... . The results are reflections at certain frequencies which cause attenuation, velocity dependant of the frequency which causes ISI. The situation is very similar to wireless channels.

There are 2 possible modulation schemes usable for ADSL : CAP (Carrierless amplitude phase) that is similar to QAM and Discrete Multitone (DMT) that is another appellation for OFDM.

Nowadays, it seems that DMT is the retained candidate for ADSL. The downlink consist of 222 tones (carriers) and uplink is splitted in 24 tones. 2 to 15 bits are coded by tone. The transmission rate is optimized with respect to line conditions. If transmission on one of the tone is disrupted because of strong reflections and interferences at the frequency band, transmission is suspended on that tone by modem.

3.3.1 HDSL

HDSL : High bit rate Digital Subscriber Line is another implementation for symetric speeds (uplink rate = downlink rate). The rate is fixed to 1,6 Mb/s. The sampling rate is 640 KHz, the number of carriers is 256. 256 QAM subsymbols are applied to a treillis encoder and then modulated by a 512 points IFFT.

 

18

 

3.4 Hiperlan/2

3.4.1 Definition

HiperLAN2 is the all new high performance radio technology, specifically suited for operating in LAN environments. HiperLAN2 is a technology being developed within the European Telecommunications Standardisation Institute (ETSI) and a final specification is due to be finalised at the end of 1999 or beginning of 2000.

HiperLAN2 operates in the unlicensed 5 GHz frequency band, which has been specifically allocated to wireless LANs. In contrast to the IEEE 802.11 wireless Ethernet technology, HiperLAN2 is connection-oriented. Connections over the air are time-division multiplexed.

Connections can also be assigned different Quality of Service (QoS). This QoS support allows for the transmission of a mix of different types of technologies, e.g. voice, video, and data. There are also specific connections for unicast, multicast, and broadcast transmission.

HiperLAN2 allows for interconnection into virtually any type of fixed network technology. Thus, HiperLAN2 can carry, for example, Ethernet frames, ATM cells, IP packets, etc. A likely first scenario for HiperLAN2 is to use it between a mobile terminal such as a laptop, and an access point.

OFDM is the modulation used in the physical layer of HiperLAN2, with a 64 point Fast Fourier Transform. For the subcarrier modulation we have choice between BPSK, QPSK, and 16-64 QAM; the symbol period used is 3.6µs with a guard interval of 0.8µs (optionally 0.4µs). The demodulation is coherent. OFDM obviously provides intentionally wide frequency band and a potential bit-rate of 54Mbit/s.

3.4.2 What is the mission statement of H2GF?

Drive the adoption of HiperLAN2 as the globally accepted, broadband wireless technology in the 5GHz band, providing connectivity for mobile devices in corporate, public and home environments.

3.4.3 Who are the founding members?

Bosch, Dell, Ericsson, Nokia, Telia, Texas Instruments

3.4.4 When will the H2GF be launched?

The world-wide launch of HiperLAN2 will take place on the 14th September 1999

in Atlanta and simultaneously in London

3.4.5 How does HiperLAN2 work?

HiperLAN2 operates in the dedicated spectrum in the 5 GHz band. It provides connections, which can be assigned a specific QoS, over the air between the terminal and the base station. While on the move, HiperLAN2 automatically performs handover to the nearest base station (called access point in HiperLAN2). HiperLAN2 also has strong security support, including both authentication and

 

19

 

encryption and has a built-in facility for automatic frequency allocation, removing the need for frequency planning.

3.4.6 Examples of use and scenarios:

With HiperLAN2 installed, workers on a construction site can use laptops to collect blue prints, order materials and communicate with experts. By sending short video sequences via the integrated camera to an expert in real time, a problem can be looked at and discussed, using the high quality audio function, then promptly solved. The broadcast function also means that everyone working on site can be contacted with any information - and that creates a more efficient on-site operation. This application looks like DVB-T and uses OFDM by the same way.

HiperLAN2 benefits companies with a flexible workforce. Employees can transfer their laptops computers from one project to another and continue to exchange large amounts of information between project members and the company server. It is also possible to connect several desktop computers and video projectors via HiperLAN2.

Domestic electronics like televisions, cameras, stereo equipment and computers can all be interconnected by HiperLAN2 using small H2 modules which automatically establish connectivity. HiperLAN2 allows multimedia equipment to be intelligently controlled from any computing device in the home without the need for network cables.

HiperLAN2 enables travellers and employees to work whilst on the move, giving them access to the company network, the Internet as well as allowing them to make and receive multimedia calls. Aircraft Engineers can also benefit from the technology. With customised software, they can access information from databases and get in touch with experts on site.

HiperLAN2 benefits both students and lecturers, allowing wireless access to the university intranet. Covering the entire campus, students can access information, such as videotaped lectures and remote supervision transmitted by their lecturer and two-way communication can take place between students and lecturers through laptops.

3.5 Other applications

• Wireless ATM transmission system. For description, please refer to [17].

• Proposed scheme for UMTS air interface for bit rates higher than 384 kbps (by Telia). But the problem of synchronization of all mobiles in the uplink to the base station is still under study.

 

20

 

4 Simulation

Some results ofthe simulation ofan OFDM system.

4.1 Results

The OFDM model used for the implementation is described in 2.4.

The program has been implemented in matlab. The channel simulation program « propag.c » (has been given to us by a person who doesn’t know the authors of the sources) associated with gene_fad.m computes a vector composed y delays and amplitude of each path. The model used is Raleigh fading.

Here are some results with an OFDM transmission on 1536 carriers with differents modulation schemes on each carrier.

4.1.1 OFDM QAM transmission with only additive noise

 

 

21

 

4.1.2 OFDM QAM transmission with Raleigh channel model

Constellation at Reception

4.1.3 OFDM BPSK transmission

Constellation at Reoeptioi

 

4.1.4 BPSK transmission on one carrier

Constellat on at Reception

4.1.5 OFDM 4-PAM transmission

Constellation at leoepticn

0.3

0.2

0.1

o_ 0 0

- -0.1

a

-0.2

-0.3

-0.4  

-0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8

lnphase compoient

23

 

 

5 Conclusion

5.1 Future

5.1.1 Multi Carrier CDMA (MC-CDMA):

The project of MC-CDMA addresses the design of improved wireless radio networks. The mobile or indoor radio channel is characterized by multipath reception: the signal offered to the receiver does not contain only a direct line-of-sight radio wave, which causes significant degradation of the network performances. The effects of multipath radio propagation, modulation and coding added to signal processing techniques on the spectrum efficiency and performance of the wireless radio networks are studied, in particular OFDM and related transmission methods. MC-CDMA intends to further develop the insight in the performance of multi-user systems using OFDM. In particular, this concept is combined with CDMA.

MC-CDMA uses Direct Sequence CDMA (DS-CDMA) merely for multiplexing, but chooses the signal waveforms using the OFDM principle. Indeed, MC-CDMA is a form of DS-CDMA, but after spreading, a Fast Fourier Transform is performed; it can be seen also as a form of OFDM, but we first apply an orthogonal matrix operation to the user bit-stream. Therefore, MC-CDMA is sometimes called CDMA-OFDM. The main improvement of MC-CDMA between OFDM is avoidance of bit errors on subcarriers that are in a deep fade. For that, OFDM generally applies coding (It is the variant of OFDM, which we have called COFDM). Hence the number of subcarriers needed is larger than the number of bits or symbols transmitted simultaneously. MC-CDMA replaces this encoder by an N*N matrix operation. The initial results reveal an improved BER.

The applications of CM-CDMA are almost same than OFDM one and try to achieve best performances. For example about Digital Audio Broadcasting (DAB), MC-CDMA is robust against fading caused by natural multipath and, it can also work if signals are received from two different transmitter sites.

5.2 Conclusion

Because of the great work on third generation mobile networks (UMTS,...) nowadays, everybody speak and think CDMA for digital transmission. However, we’ve seen that OFDM is able to handle very high data rate transmission without too much complexity at the receiver. For that, it is a prefered scheme for digital wideband transmission. So, in the near future, as offering digital wideband transmissions is the new challenge for operators, we might expect to see an explosion in the use of communications systems using OFDM modulation:

25

 

• In the subscriber line with XDSL techniques. ADSL commercial tests are currently running in France.

•In the broadband multimedia transmissions with DAB and DVB. DAB networks are being deployed around big cities

• In wireless LAN with HiperLAN/2 standard. HiperLAN/2 allow wide area coverage and for that some people think that such a network may be deployed before UMTS.

However some areas in OFDM are always under researches. This the case for synchronization and coding methods to reduce PAPR.

Eldo Mabiala Mathias Coinchon

 

26

 

6 Annexes

6.1 Bibliography:

1. J.H. Stott :The effects of frequency errors in OFDM. BBC R&D Report BBC RD 1995/15.

2. P. Shelswell : The COFDM Modulation System. BBC Report BBC RD 1996/8.

3. J. Stott :The effect of phase noise in COFDM, BBC R&D

4. Explaining some magic of OFDM

5. J. Proakis : Digital Communications, McGrawHill

6. Sari, Karam, Jeanclaude : Transmission techniques for digital terrestrial TV broadcasting; IEEE communications magazine Febr 1995 vol.33 no.2

7. P. Duhamel : Le système de transmission du projet DAB: Porteuses orthogonales (OFDM)

8. Lazlo Hazy: Initial Channel Estimation and Frame Synchronization in OFDM Systems for Frequency Selective Channels. http://www.sce.carleton.ca/~Laszlo.Hazy/OFDM/ 

9. John M. Cioffi : A multicarrier primer.

http://www-isl.stanford.edu/~cioffi/papers.html 

10. E. Lawrey : The suitability of OFDM as a modulation technique for wireless telecommunications, with a CDMA comparison; http://www.eng.jcu.edu.au/eric/thesis/chapter1.htm#Introduction

11. Speth, Fechtel, Fock, Meyr : Optimum receiver design for wireless broad-band systems using OFDM ; IEEE trans. on communications vol 47 no 11 February 99

12. Chow, Tu : A discrete Multitone Transceiver System for HDSL Applications, IEEE Journal on selected areas in communications vol.9 (6) August 1991.

13. Maddocks, Pullen, Green: Digital Audio Broadcasting, Measuring techniques and coverage performance for a medium power VHF single frequency network. BBC R&D Report BBC RD 1995/2.

27

 

14. Bonaccorso, Mhirsi : Reducing peak to average power ratio in OFDM systems. Student Eurecom project 1998.

15. Dutta-Roy :A second wind for Wiring (ADSL) ; IEEE Spectrum spetember 1999

16. ADSL Forum on http://www.adsl.com 

17. Prasad : Universal Wireless Personal Communications ; Artech House

18. P.Humblet : Telecommunications; Eurecom course notes Spring 1998

19. G. Caire : Wireless Communications; Eurecom course notes 1999

All BBC R&D articles given here are available on their web site: http://www.bbc.co.uk/rd

A list of interesting links concerning OFDM can be found at : http://studwww.eurecom.fr/~coinchon/ofdm.html

6.2 Glossary

ADSL : Asymetric Digital Subscriber Line

BER : Bit Error Rate

DSP: Digital Signal Processor

ICI : InterCarrier Interference

ISI : InterSymbol Interference

LAN: Local Area Network

Multiplex: mix of several sources

OFDM : Orthogonal Frequency multiplexing

QAM : Quadrature Amplitude Modulation

SFN : Single Frequency Network

VHF : Very High Frequency

6.3 Matlab code

The code has been removed as parts of it comes from industrial partners who want to keep it confidential. Sorry for this inconvenience.

 

28

 

SSCH: Slotted Seeded Channel Hopping for Capacity

Improvement in IEEE 802.11 Ad-Hoc Wireless Networks

 

Paramvir Bahl

Microsoft Research

One Microsoft Way

Redmond, WA 98052

bahl@microsoft.com

 

Ranveer Chandra

Department of Computer Science

Cornell University

Ithaca, NY 14853

ranveer@cs.cornell.edu

 

John Dunagan

Microsoft Research

One Microsoft Way

Redmond, WA 98052

jdunagan@microsoft.com

 

 

ABSTRACT

Capacity improvement is one of the principal challenges in wireless networking. We present a link-layer protocol called Slotted Seeded Channel Hopping, or SSCH, that increases the capacity of an IEEE 802.11 network by utilizing fre¬quency diversity. SSCH can be implemented in software over an IEEE 802.11-compliant wireless card. Each node using SSCH switches across channels in such a manner that nodes desiring to communicate overlap, while disjoint communica¬tions mostly do not overlap, and hence do not interfere with each other. To achieve this, SSCH uses a novel scheme for distributed rendezvous and synchronization. Simulation re¬sults show that SSCH significantly increases network capac¬ity in several multi-hop and single-hop wireless networking scenarios.

Categories and Subject Descriptors

C.2.2 [Computer-Communication Networks]: Network Protocols

General Terms

Algorithms, Performance

Keywords

ad-hoc wireless networks, channel assignment, frequency di¬versity, pseudo-randomness, scheduling, medium access con¬trol

1. INTRODUCTION

The problem of supporting multiple senders and receivers in wireless networks has received significant attention in the

Ranveer Chandra was supported, in part, by DARPA un¬der AFRL grant RADC F30602-99-1-0532 and by AFOSR under MURI grant F49620-02-1-0233. Additional support was provided by Microsoft Corporation. 

 

past decade. One domain where this communication pattern naturally arises is fixed wireless multi-hop networks, such as community networks [1, 4, 6, 23]. Increasing the capacity of such wireless networks has been the focus of much recent re¬search (e.g., [14,20,27]). A natural approach to increase the network capacity is to use frequency diversity [9, 31]. Com¬modity wireless networking hardware commonly supports a number of orthogonal channels, and distributing the com¬munication across channels permits multiple simultaneous communication flows.

Channelization was added to the IEEE 802.11 standard to increase the capacity of infrastructure networks — neigh¬boring access points are tuned to different channels so traffic to and from these access points does not interfere [9]. Non-infrastructure (i.e., ad-hoc) networks have thus far been un¬able to capture the benefits of channelization. The current practice in ad-hoc networks is for all nodes to use the same channel, irrespective of whether the nodes are within com¬munication range of each other [4, 6].

In this paper, we propose a new protocol, Slotted Seeded Channel Hopping (SSCH), that extends the benefits of chan-nelization to ad-hoc networks. Logically, SSCH operates at the link layer, but it can be implemented in software over an IEEE 802.11-compliant wireless Network Interface Card (NIC). The SSCH layer in a node handles three aspects of channel hopping (i) implementing the node’s channel hop¬ping schedule and scheduling packets within each channel, (ii) transmitting the channel hopping schedule to neighbor¬ing nodes, and (iii) updating the node’s channel hopping schedule to adapt to changing traffic patterns. SSCH is a distributed protocol for coordinating channel switching de¬cisions, but one that only sends a single type of message, a broadcast packet containing that node’s current channel hopping schedule. Our simulation results show that SSCH yields a significant capacity improvement in ad-hoc wireless networks, including both single-hop and multi-hop scenar¬ios.

The primary research contributions of our paper can be summarized as follows:

 

 

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.

MobiCom’04, Sept. 26-Oct. 1, 2004, Philadelphia, Pennsylvania, USA. Copyright 2004 ACM 1-58113-868-7/04/0009 ...$5.00.

 

• We present a new protocol that increases the capac¬ity of IEEE 802.11 ad-hoc networks by exploiting fre-quency diversity. This extends the benefits of channel-ization to ad-hoc networks. The protocol is suitable for a multi-hop environment, does not require changes to the IEEE 802.11 standard, and does not require mul¬tiple radios.

 

• We introduce a novel technique, optimistic synchro-nization, for distributed rendezvous and synchroniza¬tion. This technique allows control traffic to be dis¬tributed across all channels, and thus avoids control channel saturation, a bottleneck identified in prior work on exploiting frequency diversity [31].

• We introduce a second novel technique to achieve good performance for multi-hop communication flows. The partial synchronization technique allows a forwarding node to partially synchronize with a source node and partially synchronize with a destination node. This synchronization pattern allows the load for a single multi-hop flow to be distributed across multiple chan¬nels.

The rest of this paper is organized as follows: we pro¬vide background and motivate the problem in Section 2. In Section 3 we describe SSCH in detail, and in Section 4 we analyze its performance. We discuss design alternatives in Section 5, and we consider related work in Section 6. Fi¬nally, we discuss future work in Section 7, and conclude in Section 8.

2. BACKGROUND AND MOTIVATION

In this paper, we will limit our discussion to the widely-deployed IEEE 802.11 Distributed Coordination Function (DCF) protocol [8]. We begin by reviewing some relevant details of this protocol. IEEE 802.11 recommends the use of a Request To Send (RTS) and Clear To Send (CTS) mech¬anism to control access to the medium. A sender desiring to transmit a packet must first sense the medium free for a DCF interframe space (DIFS). The sender then broadcasts an RTS packet seeking to reserve the medium. If the in¬tended receiver hears the RTS packet, the receiver sends a CTS packet. The CTS reserves the medium in the neighbor¬hood of the receiver, and neighbors do not attempt to send a packet for the duration of the reservation. In the event of a collision or failed RTS, the node performs an exponential backoff. For additional details, we refer the reader to [8].

The IEEE 802.11 standard divides the available frequency into orthogonal (non-overlapping) channels. IEEE 802.11b has 11 channels in the 2.4 GHz spectrum, 3 of which are or¬thogonal, and IEEE 802.11a has 13 orthogonal channels in the 5 GHz spectrum. Packet transmissions on these orthog¬onal channels do not interfere if the communicating nodes on them are reasonably separated (at least 12 inches apart for common hardware [9]).

Using only a single channel limits the capacity of a wire¬less network. For example, consider the scenario in Figure 1 where there are 6 nodes within communication range of each other, all the nodes are on the same channel, and 3 of the nodes have packets to send to distinct receivers. Due to in¬terference on the single channel, only one of them, in this case node 3, can be active. In contrast, if all 3 orthogo¬nal channels are used, all the transmissions can take place simultaneously on distinct channels. SSCH captures the ad¬ditional capacity provided by these orthogonal channels.

We imposed three constraints on ourselves in the design of SSCH:

• SSCH should require only a single radio per node. Some of the previous work on exploiting frequency diversity has proposed that each node be equipped 

 

Figure 1: Only one of the three packets can be trans¬mitted when all the nodes are on the same channel.

with multiple radios [9,33]. Multiple radios draw more power, and energy consumption continues to be a sig¬nificant constraint in mobile networking scenarios. By requiring only a single standards-compliant NIC per node, SSCH faces fewer deployability hurdles than schemes with additional hardware requirements.

• SSCH should use an unmodified IEEE 802.11 proto¬col (including RTS/CTS) when not switching chan¬nels. Requiring standards-compliant hardware allows for easier deployment of this technology.

• SSCH should not cause a logical partition, which we de¬fine to occur when two nodes in communication range are unable to communicate. Because SSCH switches each NIC across frequency channels, different NICs may be on different channels most of the time. De¬spite this, any two nodes in communication range will overlap on a channel with moderate frequency (e.g., at least 10 ms out of every half second) and discovery is accomplished during this time. The mathematical properties of SSCH guarantee that this overlap always occurs.

SSCH exploits frequency diversity using an approach we call optimistic synchronization. SSCH is designed to make the common case be that nodes are aware of each other’s channel hopping schedules, yet SSCH allows any node to change its channel hopping schedule at any time. If node

A has traffic to send to another node B, and A knows B’s hopping schedule, A will probably be able to quickly send to

B by changing its own schedule. In the uncommon case that A does not know B’s schedule, or A has out-of-date informa¬tion about B, then the traffic incurs a latency penalty while A discovers B’s new schedule. The SSCH design achieves this good common case behavior when SSCH is used with a workload where traffic patterns change (i.e., new flows are started) less often than hopping schedule updates are propagated. Because hopping schedule update propagation requires only tens of milliseconds, this is a good workload assumption for many wireless networking scenarios. Our performance evaluation in Section 4 gives absolute numbers for these qualitative claims.

SSCH is designed to work in a single-hop or multi-hop environment, and therefore SSCH must support multi-hop flows. We introduce the partial synchronization technique to allow one node B to follow a channel hopping schedule

 

that overlaps half the time with another node A, and half the time with a third node C; this is necessary for node B to efficiently forward traffic from node A to node C. Although it is trivially possible for node B to have a channel hopping schedule that is an interleaving of A and C’s schedules, this leaves open how B will represent its schedule when a fourth node desires to synchronize with B. The design for channel hopping that we describe in Section 3.3 resolves this issue.

3. SSCH

SSCH switches each radio across multiple channels so that multiple flows within interfering range of each other can si-multaneously occur on orthogonal channels. This results in significantly increased network capacity when the network traffic pattern consists of such flows.

SSCH is a distributed protocol, suitable for deployment in a multi-hop wireless network. It does not require synchro¬nization or leader election. Nodes do attempt to synchro¬nize, but lack of synchronization results in at most a mild reduction in throughput.

SSCH defines a slot to be the time spent on a single chan¬nel. We choose a slot duration of 10 ms to amortize the over¬head of channel switching. At 54 Mbps (the maximum data rate in IEEE 802.11a), 10 ms is equivalent to 35 maximum-length packet transmissions. A longer slot duration would have further decreased the overhead of channel switching, but would have increased the delay that packets encounter during some forwarding operations. The channel schedule is the list of channels that the node plans to switch to in sub¬sequent slots and the time at which it plans to make each switch. Each node maintains a list of the channel schedules for all other nodes it is aware of – this information is al¬lowed to be out-of-date, but the common case will be that it is accurate. The good performance exhibited by SSCH (Section 4) validates this claim.

We develop the SSCH protocol by first describing its as¬sumptions about the underlying hardware and Medium Ac¬cess Control (MAC) protocol (Section 3.1). We then de¬scribe the packet transmission attempts that are made by each node within a slot, and we refer to this as the packet schedule (Section 3.2). We then define the policy for updat¬ing the channel schedule and for propagating the channel schedule to other nodes (Section 3.3). We then describe the mathematical properties that guided SSCH’s design (Sec¬tion 3.4). Finally, we discuss implementation considerations for SSCH (Section 3.5).

3.1 Hardware and MAC Assumptions

We assume that all nodes are using IEEE 802.11a – SSCH could also be used with other MACs in the IEEE 802.11 family, but this evaluation is beyond the scope of our pa¬per. IEEE 802.11a supports 13 orthogonal channels, and we assume no co-channel interference, a reasonable assumption for physically separated nodes [9]. We expect wireless cards to be capable of switching across channels. As we discuss in more detail at the beginning of Section 4, recent work has re¬duced this switching delay to approximately 80 µs ( [3,17]). We assume that each wireless card contains only a single half-duplex single-channel transceiver.

We require that NICs with a buffered packet wait after switching for the maximum length of a packet transmission before attempting to reserve the medium. This prevents hid¬den terminal problems from occurring just after switching. 

 

This requirement about the hardware is not necessary if the NIC packet buffer can be cleared whenever the channel is switched.

3.2 Packet Scheduling

SSCH maintains packets in per-neighbor FIFO queues. Using FIFO queues maintains standard higher-layer assump¬tions about in-order delivery. The per-neighbor FIFO queues are maintained in a priority queue ordered by perceived neighbor reachability. At the beginning of a slot, packet transmissions are attempted in a round-robin manner among all flows. If a packet transmission to a particular neighbor fails, the corresponding flow is reduced in priority until a pe¬riod of time equal to one half of a slot duration has elapsed – this limits the bandwidth wasted on flows targeted at nodes that are currently on a different channel to at most two packets per slot whenever a flow to a reachable node also exists. Packets are only drawn from the flows that have not been reduced in priority unless only reduced priority flows are available.

Because nodes using SSCH will often be on different chan¬nels, broadcast packets transmitted in any one slot are likely to reach only some of the nodes within physical communi¬cation range. The SSCH layer handles this issue through repeated link-layer retransmission of broadcast packets en-queued by higher layers. Although broadcast packets sent this way may reach a different set of nodes than if all nodes had been on the same channel, we have not found this to present a difficulty to protocols employing broadcast packets — in Section 4 we show that as few as 6 transmissions allows DSR (a protocol that relies heavily on broadcasts) to func¬tion well. This behavior is not surprising because broadcast packets are known to be less reliable than unicast packets, and so protocols employing them are already robust to their occasional loss. However, the SSCH retransmission strategy may not be compatible with all uses of broadcast, such as its use to do synchronization [15]. Also, deploying SSCH in an environment with a different number of channels might require the choice of 6 transmissions to be revisited. Finally, although retransmission increases the bandwidth consumed by broadcast packets, SSCH still delivers significant capacity improvement in the traffic scenarios we studied (Section 4).

An SSCH node with a packet to send may discover that a neighbor is not present on a given channel when no CTS is received in response to a transmitted RTS. However, the node may very well be present on another channel, in which case SSCH should still deliver the packet. To handle this, we initially retain the packet in the packet queue. Packets are dropped only when SSCH gives up on all packets to a given destination, and this dropping of an entire flow occurs only when we have failed to transmit a packet to the destina¬tion node for an entire cycle through the channel schedule. We will explain the meaning of a cycle through the channel schedule in Section 3.3, but with our chosen parameter set¬tings the timeout is 530 ms. After a flow has been garbage collected, new packets with the same destination inserted in the queue are assigned to a new flow, and attempted in the normal manner.

This packet scheduling policy is simple to implement, and yields good performance in the common case that node sched¬ules are known, and information about node availability is accurate. A potential drawback is that a node crash (or other failure event) can lead to a number of wasted RTSs

 

to the failed node. When added across channels, the num¬ber may exceed the IEEE 802.11 recommended limit of 7 retransmission attempts. In Section 4, we quantify the cost of such failures and show that it is small.

3.3 Channel Scheduling

We begin our description of channel scheduling by de¬scribing the data structure used to represent the channel schedule. We then describe the policy nodes use to act on their own channel schedule, the mechanism to communi¬cate channel schedules to other nodes, and finally the policy nodes implement for updating or changing their own channel schedule.

The channel schedule must capture a given node’s plans for channel hopping in the future, and there is obvious over¬head to representing this as a very long list. Instead, we compactly represent the channel schedule as a current chan¬nel and a rule for updating the channel – in particular, as a set of 4 (channel, seed) pairs. Our experimental results show that 4 pairs suffice to give good performance (Section 4). We represent the (channel, seed) pair as (xi, ai). The channel xi is represented as an integer in the range [0, 12] (13 possi¬bilities), and the seed ai is represented as an integer in the range [1, 12]. Each node iterates through all of the channels in the current schedule, switching to the channel designated in the schedule in each new slot. The node then increments each of the channels in its schedule using the seed,

xi  (xi + ai) mod 13

and repeats the process.

We introduce one additional slot to prevent logical parti¬tions. After the node has iterated through every channel on each of its 4 slots, it switches to a parity slot whose chan¬nel assignment is given by xparity = a1. The term parity slot is derived from the analogy to the parity bits appended at the end of a string in some error correcting codes. The mathematical justification for this design is given in Sec¬tion 3.4. We use the term cycle to refer to the 530 ms iter¬ation through all the slots, including the parity slot.

In Figure 2, we illustrate possible channel schedules for two nodes in the case of 2 slots and 3 channels. In the Fig¬ure, node A and node B are synchronized in one of their two slots (they have identical (channel, seed) pairs), and they also overlap during the parity slot. The field of the channel schedule that determines the channel during each slot is shown in bold. Each time a slot reappears, the chan¬nel is updated using the seed. For example, node A’s slot 1 initially has (channel, seed) = (1,2). The next time slot 1 is entered, the channel is updated by adding the seed to it mod 3 (mod 3 because in this example, there are 3 channels). The resulting channel is given by (1 + 2) mod 3 = 0.

Nodes switch from one slot to the next according to a fixed schedule (every 10 ms in our current parameter settings). However, the decision to switch channels may occur while a node is transmitting or receiving a packet. In this case we delay the switch until after the transmission and ACK (or lack thereof) have occurred.

Nodes learn each other’s schedules by periodically broad¬casting their channel schedule and offset within this cycle. We use the IEEE 802.11 Long Control Frame Header format (see Section 3.5) to embed both the schedule and the node’s current offset. The SSCH layer at each node schedules one of these packets for broadcast once per slot. 

 

Figure 2: Channel hopping schedules for two nodes with 3 channels and 2 slots. Node A always overlaps with Node B in slot 1 and the parity slot. The field of the channel schedule that determines the channel during each slot is shown in bold.

Nodes also update their knowledge of other nodes’ sched¬ules by trying to communicate and failing. Whenever a node sends an RTS to another node, and that node fails to re¬spond even though it was believed to be in this slot, the node sending the RTS updates the channel schedule for the other node to reflect that it does not currently know the node’s schedule in this slot.

We now turn to the question of how a given node changes its own schedule. Schedules are updated in two ways: each node attempts to maintain that its slots start and stop at roughly the same time as other nodes, and that its chan¬nel schedule overlaps with nodes for which it has packets to send. We embed the information needed for this syn¬chronization within the Long Control Frame Header as well. Using this information, a simple averaging scheme such as described by Elson et al [15] can be applied to achieve the loose synchronization required for good performance (Sec¬tion 4 shows that a 100 µs skew in clock times leads to less than a 2% decrease in capacity).

At a high level, each node achieves overlap with nodes for which it has traffic straightforwardly, by changing part of its own schedule to match that of the other nodes. However, a number of minor decisions must be made correctly in order to achieve this high level goal.

Nodes recompute their channel schedule right before they enqueue the packet announcing this schedule in the NIC (and so at least once per slot). In a naive approach, this node could examine its packet queue, and select the (chan¬nel, seed) pairs that lead to the best opportunity to send the largest number of packets. However, this ignores the interest this node has in receiving packets, and in avoiding congested channels. An example of the kind of problem that might arise if one ignores the interest in receiving packets is given in Figure 3. Here, A synchronized with B, and then B synchronized with C in such a way that A was no longer synchronized with B. This could have been avoided if B had used its other slot to synchronize with C, as it would have if it considered its interest in receiving packets.

 

 

Figure 3: The problem with a naive synchronization scheme. Node A has two slots, with (channel, seed) pairs represented by A1 and A2; nodes B and C are similarly depicted. At time t1, node A synchronizes with node B. Node B synchronizes with node C at time t2, after which A and B are no longer synchro¬nized.

To account for this node’s interest in receiving packets, we maintain per-slot counters for the number of packets re¬ceived during the previous time the slot was active (ignor¬ing broadcast packets). Any slot that received more than 10 packets during the previous iteration through that slot is labeled a receiving slot; if all slots are receiving slots, any one is allowed to be changed. If some slots are receiving slots and some are not, only the (channel, seed) pair on a non-receiving slot is allowed to be changed for the purpose of synchronizing with nodes we want to send to.

To account for channel congestion, we compare the (chan¬nel, seed) pairs of all the nodes that sent us packets in a given slot with the (channel, seed) pairs of all the other nodes in our list of channel schedules. If the number of other nodes synchronized to the same (channel, seed) pair is more than twice as many as this node communicated with in the pre¬vious occurrence of the slot, we attempt to de-synchronize from these other nodes. De-synchronization just involves choosing a new (channel, seed) pair for this slot. In our experiments, this de-synchronization mechanism was both necessary and sufficient to prevent the nodes from all con¬verging to the same set of (channel, seed) pairs.

The final constraints we add moderate the pace of change in schedule information. Each node only considers updating the (channel, seed) pair for the next slot, never for slots further in the future. If the previous set of criteria suggest updating some slot other than the next slot, we delay that decision. Given these constraints, picking the best possible (channel, seed) pair simply requires considering the choice that synchronizes with the set of nodes for which we have the largest number of queued packets. Additionally, the (channel, seed) pair for the first slot is only allowed to be updated during the parity slot – this helps to prevent logical partition, as will be explained in more detail in Section 3.4.

This strategy naturally supports nodes acting as sources, sinks, or forwarders. A source node will find that it can as¬sign all of its slots to support sends. A sink node will find that it rarely changes its slot assignment, and hence nodes sending to it can easily stay synchronized. A forwarding node will find that some of its slots are used primarily for receiving; after re-assigning the channel and seed in a slot to support sending, the slots that did not change are more likely to receive packets, and hence to stabilize on their cur-rent channel and seed as receiving slots for the duration of 

 

the current traffic patterns. Our simulation results (Sec¬tion 4) support this conclusion. We refer to the technique of enabling this synchronization pattern as partial synchro¬nization.

3.4 Mathematical Properties of SSCH

Our discussion of the mathematical properties of SSCH will initially focus on the static case. The behavior of SSCH when channel schedules are not changing assures us that in a steady-state flow setting, nodes will rendezvous appropri¬ately, in a sense that we make precise below. We will then expand our discussion to include the dynamics of channel scheduling in an environment where flows are starting and stopping.

The channel scheduling mechanism has three simultane¬ous design goals: allowing nodes to be synchronized in a slot, infrequent overlap between nodes that do not have data to send to each other, and ensuring that all nodes come into contact occasionally (to avoid a logical partition). To achieve these goals, we rely on a very simple mathematical technique, addition modulo a prime number.

Consider two nodes that want to be synchronized in a given slot. If they have identical (channel, seed) pairs for this slot, then clearly they will remain synchronized in future iterations (using the static assumption). Now consider two nodes that are not synchronized because they have differ¬ent seeds. A simple calculation shows that these two nodes will overlap exactly one out of every 13 iterations in this slot (recall that 13 is the number of channels). This is the behavior we want from these nodes: they overlap regularly enough that they can exchange their channel schedules, but they are mostly on different channels, and so do not interfere with each other’s transmissions.

Now consider the rare case that two nodes share identical seeds in every slot, but different channels accompany each seed – this has at most a 1 in 134  28, 000 chance of oc¬curring for randomly chosen (channel, seed) pairs. In this case, the nodes will march in lock-step through the same set of channels in each slot, never overlapping. This would be problematic, and it is this situation that the parity slot prevents. To justify this claim, we consider two distinct sit¬uations. If both nodes enter their parity slot at the same time, then they overlap there because the parity channel is equal to the seed for the first slot for both nodes. With our chosen parameter settings of 10 ms per slot, 4 slots, and 13 channels, this overlap occurs once every 530 ms and lasts for 10ms. If their parity slots do not occur at the same time, then the first node’s parity slot offers a fixed target for the slot in which the second node is changing channels, and again, the two nodes will overlap. This overlap occurs once every 7 seconds. Although both these cases will be rare, the SSCH time synchronization mechanism allows us to ignore the second case entirely – a relative clock skew of 5 ms or less is sufficient to guarantee that two parity slots overlap in time.

Now considering the dynamic case (and assuming clock synchronization to within 5 ms), we note that nodes are not permitted to change the seed for the first of their four slots except during a parity slot. Therefore they will always overlap in either the first slot or the parity slot, and hence will always be able to exchange channel schedules within a moderate time interval.

The use of addition modulo a prime to construct channel

 

hopping schedules does not restrict SSCH to scenarios where the number of channels is a prime number. If one desired to use SSCH with a wireless technology where the number of channels is not a prime, one could straightforwardly use a larger prime as the range of xi, and then map down to the actual number of channels using a modulus reduction. Though the mapping would have some bias to certain chan¬nels, the bias could be made arbitrarily small by choosing a sufficiently large prime.

A final point about the use of addition modulo a prime is that SSCH can be modified to require fewer bits to rep¬resent a node’s schedule by reducing the number of choices for a seed. The only penalty to this reduction is increasing the protocol’s reliance on the parity slot for avoiding logical partitions.

3.5 Implementation Considerations

When simulating SSCH in QualNet [5], we made two tech-nical choices that seem to be relatively uncommon based on our reading of the literature. The first technical choice re¬lates to how we added SSCH to an existing system, and the second relates to a little-utilized part of the IEEE 802.11 specification.

In order to implement SSCH, we had to implement new packet queuing and retransmission strategies. To avoid re¬quiring modifications to the hardware (in QualNet, the hard¬ware model) or the network stack, SSCH buffers packets below the network layer, but above the NIC device driver. To maintain control over transmission attempts, we config¬ure the NIC to buffer at most one packet at a time, and to attempt exactly one RTS for each packet before returning to the SSCH layer. By observing NIC-level counters before and after every attempted packet transmission, we are able to determine whether a CTS was heard for the packet, and if so, whether the packet was successfully transmitted and acknowledged. All the necessary parameters to do this are exposed by the hardware model we used in QualNet. This also prevents head-of-line blocking from interfering with our desire to implement the SSCH transmission strategy.

For efficiency reasons, we choose to use the IEEE 802.11 Long Control Frame Header format to broadcast channel schedules and current offsets, rather than using a full broad¬cast data packet. The most common control frames in IEEE 802.11 (RTS, CTS, and ACK) use the alternative short for¬mat. The long format was included in the IEEE 802.11 standard to support inter-operability with legacy 1-Mbps and 2-Mbps DSSS systems [8]. The format contains 6 un¬used bytes; we use 4 to embed the 4 (channel, seed) pairs, and another 2 to embed the offset within the cycle (i.e., how far the node has progressed through the 530 ms cycle).

Lastly, we comment that the beaconing mechanism used in IEEE 802.11 ad-hoc mode for associating with a Basic Service Set (BSS) works unchanged in the presence of SSCH. A newly-arrived node can associate to a BSS as soon as it overlaps in the same channel with any already-arrived node.

4. SYSTEM EVALUATION

We simulate SSCH in QualNet and compare its perfor¬mance with the commonly used single-channel IEEE 802.11a protocol. In Section 4.1, we present microbenchmarks quan¬tifying the different SSCH overheads. In Section 4.2, we present macrobenchmarks on the performance of SSCH with a large number of nodes in a single hop environment. In Sec¬ 

 

tion 4.3, we extend the macrobenchmark evaluation to en¬compass mobility and multi-hop routing. Our results show that SSCH incurs very low overhead, and significantly out¬performs IEEE 802.11a in a multiple flow environment.

Our simulation environment comprises a varying number of nodes in a 200m × 200m area. All nodes in a single simula¬tion run use the same MAC, either SSCH or IEEE 802.11a. We set all nodes to operate at the same raw data rate, 54 Mbps. We assume 13 usable channels in the 5 GHz band. SSCH is configured to use 4 seeds, and each slot duration is 10 ms. All seeds are randomly chosen at the beginning of each simulation run. The macrobenchmarks in Sections 4.2 and 4.3 are averages from 5 independent simulation runs, while the microbenchmarks in Section 4.1 are drawn from a single simulation run.

We primarily measure throughput under a traffic load of maximum rate UDP flows. In particular, we use Constant Bit Rate (CBR) flows of 512 byte packets sent every 50 µs. This data rate is more than the sustainable throughput of IEEE 802.11a operating at 54 Mbps.

For all our simulations, we modified QualNet to use a channel switch delay of 80 µs. This choice was based on re¬cent work in solid state electronics on reducing the settling time of the Voltage Control Oscillator (VCO) [7]. Switching the channel of a wireless card requires changing the input voltage of the VCO, which operates in a Phase Locked Loop (PLL) to achieve the desired output frequency. The delay in channel switching is due to this settling time. The speci¬fication of Maxim IEEE 802.11b Transceivers [3] shows this delay to be 150 µs. A more recent work [17] shows that this delay can be reduced to 40-80 µs for IEEE 802.11a cards.

4.1 Microbenchmarks

We present microbenchmarks measuring the overhead of SSCH in several different scenarios. In Section 4.1.1, we measure the overhead during the successful initiation of a CBR flow. In Section 4.1.2, we measure the overhead on an existing session of failing to initiate a parallel CBR flow. In Section 4.1.3, we measure the overhead of supporting two flows simultaneously. In Section 4.1.4, we measure the over-head of continuing to attempt transmissions to a mobile node that has moved out of range. These scenarios cover many of the different dynamic events that a MAC must ap¬propriately handle: a flow starting while a node is present, a flow starting while a node is absent, simultaneous flows where both nodes are present, simultaneous flows where one node moves out of range, etc. Finally, the scenario in Sec¬tion 4.1.5 measures the overhead of SSCH with respect to a different kind of event, clock skew.

4.1.1 Overhead of Switching and Synchronizing

In this experiment, we measured the overhead of success¬fully initiating a CBR flow between two nodes within com¬munication range of each other. The first node initiates the flow just after the parity slot. This incurs a worst-case delay in synchronization, because the first of the four slots will not be synchronized until 530 ms later.

Figure 4 shows the moving average over 20 ms of the throughput at the receiver node. The sender quickly syn¬chronizes with the receiver on three of the four slots, as it should, and on the fourth slot after 530 ms. The fig¬ure shows the throughput while synchronizing (oscillating around 3/4 of the raw bandwidth), and the time required

 

10 12 14 16 18 20 22 24

Time (in seconds)

SSCH 802.11a

Figure 4: Switching and Synchronizing Overhead: Node 1 starts a maximum rate UDP flow to Node 2. We show the throughput for both SSCH and IEEE 802.11a.

to synchronize. After synchronizing, the channel switch¬ing and other protocol overheads of SSCH lead to only a 400 Kbps penalty in the steady-state throughput relative to IEEE 802.11a. This penalty conforms to our intuition about the overheads in SSCH: a node spends 80 µs every 10 ms switching channels (80 µs/10 ms = .008), and then must wait for the duration of a single packet to avoid colliding with pre-existing packet transmissions in the new channel (1 packet/35 packets = .028). Adding these two overheads together leads to an expected cumulative overhead of 3.6%, which is in close agreement with the measured overhead of (400 Kbps/12 Mbps) = 3.3%.

Note that the throughput reaches a maximum of only 13 Mbps, although the raw data rate is 54 Mbps. This low uti¬lization can be explained by the IEEE 802.11a requirement that the RTS/CTS packets be sent at the lowest supported data rate, 6 Mbps, along with other overheads [18].

4.1.2 Overhead of an Absent Node

SSCH requires more re-transmissions than IEEE 802.11 to prevent logical partitions. These retransmissions waste bandwidth that could have been dedicated to a node that was present on the channel. To quantify this overhead, we initiated a CBR flow between two nodes, allowed the system to quiesce, and then initiated a send from the first node to a non-existent node. Figure 5 shows the moving average over 80 ms of the throughput. The Figure shows that the sender takes 530 ms to timeout on the non-existent node. During this time the throughput drops by 550 Kbps, which is a small fraction (4.6%) of the total throughput.

4.1.3 Overhead of a Parallel Session

Our next experiment quantifies the ability of SSCH to fairly share bandwidth between two flows, and to quickly achieve this fair sharing. To measure this we start with Node 1 sending a maximum rate UDP flow to Node 2. At 21.5 seconds, Node 1 starts a second maximum rate UDP flow to Node 3. Figure 6 presents the moving average over

 

 

23 23.5 24 24.5 25 25.5 26 26.5 27

Time (in seconds)

Figure 5: Overhead of an Absent Node: Node 1 is sending a maximum rate UDP flow to Node 2. Node 1 then attempts to send a packet to a non-existent node.

14

12

10

8

6

4

2

0

15 20 25 30 35

Time (in seconds)

Node 2 Node 3

Figure 6: Overhead of a Parallel Session: Node 1 is sending a maximum rate UDP flow to Node 2. Node 1 then starts a second flow to Node 3.

140 ms of the throughput achieved by both receivers. The bandwidth is split between the receivers nearly perfectly, and with no decrease in net throughput.

4.1.4 Overhead of Mobility

We now analyze the effect of mobility at a micro-level on the performance of SSCH. Ideally, SSCH should be able to detect a link breakage due to movement of a node, and sub¬sequently re-synchronize to other neighbors. We show that SSCH can indeed handle this scenario with an experiment comprising 3 nodes and 2 flows, and in Figure 7 we present the moving average over 280 ms of each flow’s throughput.

Node 1 is initially sending a maximum rate UDP flow to Node 2. Node 1 initiates a second UDP flow to Node 3 at around 20.5 seconds. This bandwidth is then shared be¬tween both the flows (as in the experiment of Section 4.1.3) until 30 seconds, when Node 3 moves out of the communi¬cation range of Node 1. Our experiment configures Node 1 to continue to attempt to send to Node 3 until 43 seconds,

 

Node 2 Node 3

Figure 7: Overhead of Mobility: Node 1 is sending a maximum rate UDP flow to Node 2. Node 1 starts another maximum rate UDP flow to Node 3. Node 3 moves out of range at 30 seconds, while Node 1 continues to attempt to send until 43 seconds.

14

12

10

8

6

4

2

0

1 ns 10 ns 100 ns 1 µs 10 µs 100 µs 1 ms

Clock Drift

Figure 8: Overhead of Clock Skew: Throughput be¬tween two nodes using SSCH as a function of clock skew.

and during this time it continues to consume a small amount of bandwidth. In contrast, the experiment in Section 4.1.2 measured the overhead of enqueueing a single packet to an absent node. When the flow to Node 3 finally stops, Node 2’s received throughput increases back to its initial rate.

4.1.5 Overhead of Clock Drift

As we described in Section 3.3, SSCH tries to synchronize slot begin and end times, though it is also designed to be robust to clock skew. In this experiment, we quantify the robustness of SSCH to moderate clock skew. We measure the throughput between two nodes after artificially intro¬ducing a clock skew between them, and disabling the SSCH synchronization scheme for slot begin and end times. We vary the clock skew from 1 ns (106 ms) to 1 ms such that the sender is always ahead of the receiver by this value, and present the results in Figure 8. Note the log scale on the x-axis.

The throughput achieved between the two nodes is not significantly affected by a clock skew of less than 10 µs. 

 

These practical values of clock skew are extremely small to impact the performance of SSCH. The drop in throughput is more for larger clock skews, although the throughput is still acceptable at 10.5 Mbps when the skew value is an extremely high 1 ms.

These results provide justification for the design choice we made of not requiring nodes to switch synchronously across slots, as described in Section 3.3. For example, a node will delay switching to receive an ACK, or to send a data packet if its channel reservation is successful. In the 100 node ex-periment described in Section 4.3.2, we measured the skew in channel switching times for a traffic pattern of 50 flows to be approximately 20 µs. Figure 8 shows that this is a negligible amount.

4.2 Macrobenchmarks: Single-hop Case

We now present simulation results showing SSCH’s ability to achieve and sustain a consistently high throughput for a traffic pattern consisting of multiple flows. We first evaluate this using steady state UDP flows. We then extend our evaluation to consider a dynamic traffic scenario where UDP flows both start and stop. Finally, we study the performance of TCP over SSCH.

4.2.1 Disjoint Flows

We first look at the number of disjoint flows that can be supported by SSCH. All nodes in this experiment are in communication range of each other, and therefore two flows are considered disjoint if they do not share either endpoint. Ideally, SSCH should utilize the available bandwidth on all the channels on increasing the number of disjoint flows in the system. We evaluate this by varying the number of nodes in the network from 2 to 30 and introducing a flow between disjoint pairs of nodes — the number of flows varies from 1 to 15.

Figure 9 shows the per-flow throughput, and Figure 10 shows the total system throughput. IEEE 802.11a performs marginally better when there is just one flow in the network. When there is more than one flow, SSCH significantly out¬performs IEEE 802.11a.

An increase in the number of flows decreases the per-flow throughput for both SSCH and IEEE 802.11a. However, the drop for IEEE 802.11a is much more significant. The drop for IEEE 802.11a is easily explained by Figure 10, which shows that the overall system throughput for IEEE 802.11a is approximately constant.

It may seem surprising that the SSCH system throughput has not stabilized at 13 times the throughput of a single flow by the time there are 13 flows. However, this can be attributed to SSCH’s use of randomness to distribute flows across channels. These random choices do not lead to a perfectly balanced allocation, and therefore there is still un¬used spectrum even when there are 13 flows in the system, as shown by the continuing positive slope of the curve in Figure 9.

4.2.2 Non-disjoint Flows

We now consider the case when the flows in the network are not disjoint – nodes participate as both sources and sinks, and in multiple flows. This scenario stresses SSCH’s ability to efficiently support sharing among simultaneous flows that have a common endpoint. Each node in the net¬work starts a maximum rate UDP flow with one other ran 

 

 

2 4 6 8 10 12 14 16

# Nodes

Figure 13: Effect of Flow Duration: Ratio of SSCH throughput to IEEE 802.11a throughput for flows having different durations.

16

14

12

10

8

6

4

2

0

2 3 4 5 6 7 8

# Flows

Figure 14: TCP over SSCH: Steady-state TCP throughput when varying the number of non-disjoint flows.

evaluate this concern quantitatively, we run an experiment where we vary the number of nodes in the network from 2 to 9, such that all nodes are in communication range of one another. We then start an infinite-size file transfer over FTP from each node to a randomly selected other node. This choice to use non-disjoint flows is designed to stress the SSCH implementation by requiring nodes to be synchronized as either senders or receivers with multiple other nodes. In Figure 14 we present the resulting cumulative steady-state TCP throughput over all the flows in the network.

Figure 14 shows that the TCP throughput for a small number of flows is lower for SSCH than the throughput over IEEE 802.11a. However, as the number of flows increases, SSCH does achieve a higher system throughput. Although TCP over SSCH does provide higher aggregate throughput than over IEEE 802.11a, the performance improvement is not nearly as good as for UDP flows. This shows that jitter due to SSCH does have an impact on the performance of

 

 

0 6 12 18

# Nodes

SSCH 802.11a

Figure 15: Multi-hop Chain Network: Variation in throughput as chain length increases.

TCP. A more detailed analysis of the interaction between TCP and SSCH, and modifications to support better inter¬actions between TCP and SSCH, is a subject we plan to address in our future work.

4.3 Macrobenchmarks: Multi-hop Case and Mobility

We now evaluate SSCH’s performance when combined with multi-hop flows and mobile nodes. We first analyze the behavior of SSCH in a multi-hop chain network. We then consider large scale multi-hop networks, both with and without mobility. As part of this analysis, we study the interaction between SSCH and MANET routing protocols.

4.3.1 Performance in a Multi-hop Chain Network

IEEE 802.11 is known to encounter significant perfor¬mance problems in a multi-hop network [34]. For example, if all nodes are on the same channel, the RTS/CTS mechanism allows at most one hop in an A  B  C  D chain to be ac¬tive at any given time. SSCH reduces the throughput drop due to this behavior by allowing nodes to communicate on different channels. To examine this, we evaluate both SSCH and IEEE 802.11a in a multi-hop chain network.

We vary the number of nodes, which are all in communi-cation range, from 2 to 18. We initiate a single flow that encounters every node in the network. Although more than 4 nodes transmitting within interference range of each other would be unlikely to arise from multi-hop routing of a single flow, it could easily arise in a general distributed application. Figure 15 shows the maximum throughput as the number of nodes in the chain is varied. We see that there is not much difference between SSCH and IEEE 802.11a for flows with few hops. As the number of hops increases, SSCH performs much better than IEEE 802.11a since it distributes the com-munication on each hop across all the available channels.

4.3.2 Performance in a Multi-hop Mesh Network

We now analyze the performance of SSCH in a large scale multi-hop network without mobility. We place 100 nodes uniformly in a 200 × 200 m area, and set each node to trans¬mit with a power of 21 dBm. The Dynamic Source Routing

 

Figure 16: Mulithop Mesh Network of 100 Nodes: The per-flow throughput on varying the number of flows in the network.

(DSR) [22] protocol is used to discover the source route be¬tween different source-destination pairs. These source routes are then input to a static variant of DSR that does not per¬form discovery or maintain routes. We vary the number of maximum rate UDP flows from 10 to 50. We generate source-destination pairs by choosing randomly, and reject¬ing pairs that are within a single hop of each other.

We present the per-flow throughput in Figure 16. Increas¬ing the number of flows leads to greater contention, and the average throughput of both SSCH and IEEE 802.11a drops. For every considered number of flows, SSCH provides signif¬icantly higher throughput than IEEE 802.11a. For 50 flows, the inefficiencies of sharing a single channel are sufficiently pronounced that SSCH yields more than a factor of 15 ca¬pacity improvement.

4.3.3 Impact of Channel Switching on

MANET Routing Protocols

Previous work on multi-channel MACs has often over¬looked the effect of channel switching on routing protocols. Most of the proposed protocols for MANETs rely heavily on broadcasts (e.g., DSR [22] and AODV [28]). However, neighbors using a multi-channel MAC could be on different channels, which could cause broadcasts to reach significantly fewer neighbors than in a single-channel MAC. SSCH ad¬dresses this concern using a broadcast retransmission strat¬egy discussed in Section 3.2.

We study the behavior of DSR [22] over SSCH in the same experimental setup used in Section 4.3.2, with 100 nodes in a 200 m×200 m area. However, we reduce the transmission power of each node to 16 dBm to force routes to increase in length (and hence to stress DSR over SSCH). We select 10 source-destination pairs at random, and we use DSR to discover routes between them. In Figure 17 we compare the performance of DSR over SSCH, when varying the SSCH broadcast transmission count parameter (the number of con¬secutive slots in which each broadcast packet is sent once).

Figure 17 shows that the performance of DSR over SSCH improves with an increase in the broadcast transmission count. The DSR Route Request packets see more neighbors when SSCH broadcasts them over a greater number of slots. 

 

Figure 17: Impact of SSCH on Unmodified MANET Routing Protocols: The average time to discover a route and the average route length for 10 randomly chosen routes in a 100 node network using DSR over SSCH.

This increases the likelihood of discovering shorter routes, and the speed with which routes are discovered. However, there seems to be little additional benefit to increasing the broadcast parameter to a value greater than 6. We attribute the slight bumpiness in the curves to the stochastic nature of DSR, and its reliance on broadcasts.

Comparing SSCH to IEEE 802.11a, we see that the SSCH discovers routes that are comparable in length. However, the average route discovery time for SSCH is much higher than for IEEE 802.11a. Because each slot is 10 ms in length, broadcasts are only retransmitted once every 10 ms, and this leads to a significantly longer time to discover a route to a given destination node. We believe that this latency is a fundamental difficulty in using a reactive protocol such as DSR with SSCH. We plan to explore the interaction of other proactive and hybrid routing protocols with SSCH in the future.

4.3.4 Performance in Multi-hop Mobile Networks

We now study the impact of mobility in a network using DSR over IEEE 802.11a and SSCH. In this experiment, we place 100 nodes randomly in a square and select 10 flows. Each node transmits packets at 21 dBm. Node movement is determined using the Random Waypoint model. In this model, each node has a predefined minimum and maximum speed. Nodes select a random point in the simulation area, and move towards it with a speed chosen randomly from the interval. After reaching its destination, a node rests for a period chosen from a uniform distribution between 0 and 10 seconds. It then chooses a new destination and repeats the procedure. In our experiments, we fix the minimum speed at 0.01 m/s and vary the maximum speed from 0.2 to 1.0 m/s. Although we have studied SSCH at higher speeds, the results are not significantly different. We performed this experiment using two different areas for the nodes, a 200m × 200m area and a 300m × 300m area. We refer to the smaller area as the dense network, and the larger area as the sparse network – the average path is 0.5 hops longer

 

Figure 18: Dense Multi-hop Mobile Network: The per-flow throughput and the average route length for 10 flows in a 100 node network in a 200m × 200m area, using DSR over both SSCH and IEEE 802.11a.

 

Figure 19: Sparse Multi-hop Mobile Network: The per-flow throughput and the average route length for 10 flows in a 100 node network in a 300m × 300m area, using DSR over both SSCH and IEEE 802.11a.

 

 

in the sparse network. For all these experiments, we set the SSCH broadcast transmission count parameter to 6.

Figure 18 shows that in a dense network, SSCH yields much greater throughput than IEEE 802.11a even when there is mobility. Although DSR discovers shorter routes over IEEE 802.11a, the ability of SSCH to distribute traffic on a greater number of channels leads to much higher overall throughput. Figure 19 evaluates the same benchmarks in a sparse network. The results show that the per-flow through¬put decreases in a sparse network for both SSCH and IEEE 802.11a. This is because the route lengths are greater, and it takes more time to repair routes. However, the same qual¬itative comparison continues to hold: SSCH causes DSR to discover longer routes, but still leads to an overall capacity improvement.

DSR discovers longer routes over SSCH than over IEEE 802.11a because broadcast packets sent over SSCH may not reach a node’s entire neighbor set. Furthermore, some opti¬mizations of DSR, such as promiscuous mode operation of nodes, are not as effective in a multi-channel MAC such as SSCH. Thus, although the throughput of mobile nodes us¬ing DSR over SSCH is much better than their throughput over IEEE 802.11a, we conclude that a routing protocol that takes the channel switching behavior of SSCH into account will likely lead to even better performance.

5. DISCUSSION

In this Section we discuss alternative designs for SSCH within the constraints that we enumerated in Section 2. We will discuss prior work related to SSCH in detail in Section 6.

SSCH distributes the rendezvous and control traffic across all the channels. One straightforward alternative scheme, which still only requires one radio, is to use one of the chan¬nels as a control channel, and all the other channels as data channels (e.g., [21]). Each node must then somehow split its time between the control channel and the data channels.

Such a scheme will have difficulty in preventing the con¬trol channel from becoming a bottleneck. Suppose that two nodes exchange RTS/CTS on the control channel, and then 

 

switch to a data channel to do transmission. Unless all other nodes were also on the control channel during the RTS/CTS exchange, these two nodes will still need to do an RTS/CTS on this channel in order to avoid the hidden terminal prob¬lem. The two nodes should wait to even do the RTS/CTS until after an entire packet transmission interval has elapsed, because another pair of nodes might have also switched to this channel, orchestrating that decision on the control chan¬nel during a time that the first pair of nodes were not on the control channel. In order to amortize this startup cost, the nodes should have several packets to send to each other. However, while any one node remains on a data channel, any other node that desires to send it a packet must remain idle on the control channel waiting for the node it desires to reach to re-appear. If the idle node on the control channel chooses not to wait, and instead switches to a data channel with another node for which it has traffic, it may repeatedly fail to rendezvous with the first node, leading to a significant imbalance in throughput and possibly a logical partition.

The problems with a dedicated control channel may be solvable, but it is clear that a straightforward approach with un-synchronized rendezvous presents several difficulties. If one instead tried to synchronize rendezvous on the control channel, the control channel could again become a bottle¬neck simply because many nodes simultaneously desire to schedule packets on that channel.

6. RELATED WORK

We divide the prior work relevant to SSCH into three categories: prior uses of pseudo-random number generators in wireless networking, channel switching for reasons other than capacity improvement, and alternative approaches to exploiting frequency diversity. In the first category, we find that pseudo-random number generators have been used for a variety of tasks in wireless networking. For example, the SEEDEX protocol [29] uses pseudo-random generators to avoid RTS/CTS exchanges in a wireless network. Nodes build a schedule for sending and listening on a network, and publish their seeds to all the neighbors. A node will attempt

 

a transmission only when all its neighbors (including the receiver) are in a listening state. Assuming relatively con¬stant wireless transmission ranges, this protocol also helps in overcoming the hidden and exposed terminal problem. The TSMA protocol [11,12] is a channel access scheme pro¬posed as an alternative to ALOHA and TDMA, for time-slotted multihop wireless networks. TSMA aims to achieve the guarantees of TDMA without incurring the overhead of transmitting large schedules in a mobile environment. Each node is bootstrapped with a fixed seed that determines its transmission schedule. The schedules are constructed using polynomials over Galois fields (which have pseudo-random properties), and the construction guarantees that each node will overlap with only a single other node within a certain time frame. The length of the schedule depends on the num-ber of nodes and the degree of the network. Porting these schedules to a multichannel scenario, where the number of channels is fixed, remains an open problem, and even such a porting would not meet the SSCH goal of supporting traffic-driven overlap. Redi et al. [13] use a pseudo-random gen¬erator to derive listening schedules for battery-constrained devices. Each device’s seed is known to a base station, which can then schedule transmissions for the infrequent moments when the battery-constrained device is awake. Although pseudo-random generators have been used for a number of tasks (as this survey of the literature makes clear), to the best of our knowledge, SSCH is the first protocol to use a pseudo-random generator to construct a channel hopping schedule.

The second category of prior work is channel switching for reasons other than capacity improvement. MultiNet [10] is the main piece of work that we are aware of in this cate¬gory. MultiNet allows a NIC to periodically hop between two channels, enabling a single wireless NIC to connect to two logically distinct networks, such as an AP network and an ad-hoc network. MultiNet is designed to provide new functionality: simultaneous connectivity to distinct net¬works using a single NIC. In contrast, SSCH is designed to yield capacity improvement within a single ad-hoc network.

The third category of prior work we define encompasses all prior approaches to increasing network capacity by ex¬ploiting frequency diversity. This is a significant body of work. The first division we make in this body of work is between research that assumes a single NIC capable of com¬municating on a single channel at any given instance in time, and research that assumes more powerful radio technology, such as multiple NICs [9,30] or NICs capable of listening on many channels simultaneously [21, 26], even if they can only communicate on one. Our work falls in to the former cate¬gory; the SSCH architecture can be deployed over a single standards-compliant NIC supporting fast channel switching.

Dynamic Channel Assignment (DCA) [33] and Multi-radio Unification Protocol (MUP) [9] are both technologies that use multiple radios (in both cases, two radios) to take advan¬tage of multiple orthogonal channels. DCA uses one radio on a control channel, and the other radio switches across all the other channels sending data. Arbitration for channels is embedded in the RTS and CTS messages, and is exe-cuted on the control channel. Although this scheme may fully utilize the data channel, it does so at the cost of us¬ing an entire radio just for control. MUP uses both radios for data and control transmissions. Radios are assigned to orthogonal channels, and a packet is sent on the radio with 

 

better channel characteristics. This scheme gives good per-formance in many scenarios. However, it still only allows the use of as many channels as there are radios on each physi¬cal node. From our perspective, the key drawback to both DCA and MUP is simply that they require the use of mul¬tiple radios. Recently, commercial products have appeared that claim the ability to place multiple radios on a single NIC [2]. It is still not known whether these products will ever achieve as many radios on a NIC as there are available channels, nor what their power consumption will be.

A straightforward way to view the different potential gains of SSCH compared to a true multiple radio design is to con¬sider two distinct sources of bottleneck in a single-radio, single-channel system: the saturation of the channel, and the saturation of any particular radio. Conceptually, SSCH significantly increases the channel bandwidth, without in¬creasing the bandwidth of any individual radio. In contrast, a true multiple radio design increases both. A specific ex¬ample of this difference is that a node using MUP (a true multiple radio design) can simultaneously send and receive packets on separate channels, while a node using SSCH can only perform one of these operations at a time.

We next turn our attention to work assuming more pow¬erful radio technology than is currently technologically fea¬sible. HRMA [35] is designed for frequency hopping spread spectrum (FHSS) wireless cards. Time is divided into slots, each one of which corresponds to a small fraction of the time required to send a packet, and the wireless NIC is on a dif¬ferent frequency during each slot. All nodes are required to maintain synchronized clocks, where the synchronization is at the granularity of slot times that are much shorter than the duration of a packet. Each slot is subdivided in to four segments of time for four different possible communica¬tions: HOP-RESERVED/RTS/CTS/DATA. The first three segments of time are assumed to be small in comparison with the amount of time spent sending a segment of the packet during the DATA time interval. To the best of our knowl¬edge, a FHSS wireless card that supports this type of MAC protocol at high data rates is not commercially available.

Another line of related work assumes technology by which nodes can concurrently listen on all channels. For example, Nasipuri et al [26] and Jain et al [21] assume wireless NICs that can receive packets on all channels simultaneously, and where the channel for transmission can be chosen arbitrarily. In these schemes, nodes maintain a list of free channels, and either the sending or receiving node chooses a channel with the least interference for its data transfer. Wireless NICs do not currently support listening on arbitrarily many channels, and we do not assume the availability of such technology in the design of SSCH.

We finally consider prior work that only assumes the pres¬ence of a single NIC with a single half-duplex transceiver. The only other approach that we are aware of to exploit¬ing frequency diversity under this assumption is Multichan¬nel MAC (MMAC) [31]. Like SSCH, MMAC attempts to improve capacity by arranging for nodes to simultaneously communicate on orthogonal channels. Briefly, MMAC oper¬ates as follows: nodes using MMAC periodically switch to a common control channel, negotiate their channel selections, and then switch to the negotiated channel, where they con¬tend for the channel as in IEEE 802.11. This scheme raises several concerns that SSCH attempts to overcome. First, MMAC extends IEEE 802.11 Power Save Mode (PSM) for

 

ad-hoc networks. This implies a relatively stringent reliance on clock synchronization, which is particularly hard to pro¬vide in multi-hop wireless networks [19]. In contrast, SSCH does not require tight clock synchronization because SSCH does not have a common control channel or a dedicated neighbor discovery interval. Secondly, synchronization traf¬fic in MMAC can be a significant fraction of the system traf¬fic, and the common synchronization channel can become a bottleneck on system throughput. SSCH addresses this concern by distributing synchronization and control traf¬fic across all the available channels. A third concern with MMAC is that it can lead to packet delays of hundreds of milliseconds for even a single hop. MMAC switches channels every 100 ms, so a node with packets for two different desti¬nations will have to wait at least 100 ms to send traffic to one of them whenever the two destinations decide to use different channels. In contrast, SSCH performs well while switching channels every 10 ms. A fourth concern with MMAC is that it does not specify how to support broadcasts, which are required by most MANET routing protocols (e.g., DSR). SSCH addresses this using a broadcast retransmission strat¬egy that we experimentally validated to be compatible with DSR.

Although this survey does not cover all related work, it does characterize the current state of the field. At the level of detail in this section, prior work such as CHMA [32] is similar to HRMA [35], and MAC-SCC [25] and the MAC protocols implicit in the work of Li et al [24] and Fitzek et al [16] are similar to DCA [33]. However, a final related channel hopping technology that is worth mentioning is the definition of FHSS channels in the IEEE 802.11 [8] specifica¬tion. At first glance, it may seem redundant that SSCH does channel hopping across logical channels, each one of which (per the IEEE 802.11 specification) may be employing fre¬quency hopping across distinct frequencies at the physical later. The IEEE 802.11 specification justifies this physi¬cal layer frequency hopping with the scenario of providing support for multiple Basic Service Sets (BSS’s) that can co¬incide geographically without coinciding on the same logical channel. In contrast, SSCH does channel hopping so that any two nodes can coincide as much or as little of the time as they desire. This is also at the heart of the difference between SSCH and past work on channel-hopping proto¬cols where nodes overlap a fixed fraction of the time [12] – the degree of overlap between any two nodes using SSCH is traffic-dependent.

7. FUTURE RESEARCH

SSCH is a promising technology. In our future work, we plan to investigate how SSCH will perform when imple¬mented over actual hardware, and is subject to the normal environmental vagaries of wireless networks, such as unpre¬dictable variations in signal strength. As part of this im¬plementation effort, we also plan to evaluate how metrics reflecting environmental conditions, such as ETX [14], can be integrated into SSCH.

Our results in Section 4.3.3 show that running existing routing protocols over SSCH can incur a large route discov¬ery time and an increase in the route length. In the future, we plan to more thoroughly evaluate routing over SSCH (as opposed to classical single channel routing), and to explore SSCH’s interaction with a wider variety of proactive and hybrid routing protocols.

 

There are at least four additional topics that would also need to be addressed before SSCH can be deployed. One is interoperability with nodes that are not running SSCH. Another is the evaluation of power consumption under this scheme. We have not attempted to evaluate the energy cost of switching channels, nor have we attempted to enable a power-saving strategy such as in the IEEE 802.11 specifica-tion for access-point mode. A third topic of investigation is the evaluation of SSCH in conjunction with auto-rate adap¬tation mechanisms. A fourth topic is a more detailed eval¬uation of the interplay between SSCH and TCP.

8. CONCLUSION

We have presented SSCH, a new protocol that extends the benefits of channelization to ad-hoc networks. This protocol is compatible with the IEEE 802.11 standard, and is suitable for a multi-hop environment. SSCH achieves these gains using a novel approach called optimistic synchronization. We expect this approach to be useful in additional settings beyond channel hopping.

We have shown through extensive simulation that SSCH yields significant capacity improvement in a variety of single-hop and multi-hop wireless scenarios. In the future, we look forward to exploring SSCH in more detail using an imple-mentation over actual hardware.

Acknowledgment

The authors would like to thank Ken Birman for his insight¬ful comments on early drafts of this paper.

9. REFERENCES

[1] Bay Area Wireless Users Group, http://www.bawug.org.

[2] Engim, http://www.engim.com/.

[3] Maxim 2.4GHz 802.11b Zero-IF Transceivers. http://pdfserv.maxim-ic.com/en/ds/MAX2820-MAX2821.pdf.

[4] MIT RoofNet, http://www.pdos.lcs.mit.edu/roofnet/.

[5] QualNet, http://www.qualnet.com/.

[6] Seattle Wireless, http://www.seattlewireless.net/.

[7] Tracking Advances in VCO Technology, http://pdfserv.maxim-ic.com/en/an/AN1768.pdf.

[8] IEEE 802.11b/D3.0, Wireless LAN Medium Access Control(MAC) and Physical (PHY) Layer Specification: High Speed Physical Layer Extensions in the 2.4 GHz Band, 1999.

[9] A. Adya, P. Bahl, J. Padhye, A. Wolman, and L. Zhou. A Multi-Radio Unification Protocol for IEEE 802.11 Wireless Networks. In IEEE International Conference on Broadband Networks (Broadnets) 2004.

[10] R. Chandra, P. Bahl, and P. Bahl. MultiNet: Connecting to Multiple IEEE 802.11 Networks Using a Single Wireless Card. In IEEE Infocom 2004.

[11] I. Chlamtac and A. Farago. Making Transmission Schedules Immune to Topology Changes in Multi-Hop Packet Radio Networks. IEEE/ACM Transactions on Networking, 2(1):23–29, February 1994.

[12] I. Chlamtac and A. Farago. Time-Spread Multiple-Access (TSMA) Protocols for Multihop Mobile Radio Networks. IEEE/ACM Transactions on Networking, 5(6):804–812, December 1997.

 

[13] I. Chlamtac, C. Petrioli, and J. Redi.

Energy-Conserving Access Protocols for Identification Networks. IEEE/ACM Transactions on Networking, 7(1):51–61, February 1999.

[14] D. D. Couto, D. Aguayo, J. Bicket, and R. Morris. A High-Throughput Path Metric for Multi-Hop Wireless Routing. In ACM MobiCom 2003.

[15] J. Elson, L. Girod, and D. Estrin. Fine-Grained Network Time Synchronization using Reference Broadcast. In USENIX OSDI 2002.

[16] F. Fitzek, D. Angelini, G. Mazzini, and M. Zorzi. Design and performance of an enhanced IEEE 802.11 MAC protocol for multihop coverage extension. IEEE Wireless Communications, 10(6):30–39, December 2003.

[17] F. Herzel, G. Fischer, and H. Gustat. An Integrated CMOS RF Synthesizer for 802.11a Wireless LAN. IEEE Journal of Solid-state Circuits, 18(10), October 2003.

[18] M. Heusse, F. Rousseau, G. Berger-Sabbatel, and

A. Duda. Performance Anomaly of 802.11b. In IEEE

Infocom 2003.

[19] L. Huang and T.-H. Lai. On the scalability of IEEE 802.11 ad hoc networks. In ACM MobiHoc 2002.

[20] K. Jain, J. Padhye, V. Padmanabhan, and L. Qiu. Impact of Interference on Multi-hop Wireless Network Performance. In ACM MobiCom 2003.

[21] N. Jain and S. R. Das. A Multichannel CSMA MAC Protocol with Receiver-Based Channel Selection for Multihop Wireless Networks. In IEEE International Conference on Computer Communications and Networks (IC3N) 2001.

[22] D. Johnson, D. Maltz, and J. Broch. DSR: The Dynamic Source Routing Protocol for Multihop Wireless Ad Hoc Networks. In C. Perkins, editor, Ad Hoc Networking, chapter 5, pages 139–172. Addison-Wesley, 2001.

[23] R. Karrer, A. Sabharwal, and E. Knightly. Enabling Large-scale Wireless Broadband: The Case for TAPs. In ACM HotNets 2003.

[24] J. Li, Z. J. Haas, M. Sheng, and Y. Chen. Performance Evaluation of Modified IEEE 802.11 MAC for Multi-Channel Multi-Hop Ad Hoc Network. In IEEE International Conference on Advanced Information Networking and Applications (AINA) 2003. 

 

[25] Y. Li, H. Wu, D. Perkins, N.-F. Tzeng, and

M. Bayoumi. MAC-SCC: Medium Access Control with a Separate Control Channel for Multihop Wireless Networks. In IEEE International Conference on Distributed Computing Systems (ICDCS) Workshop 2003.

[26] A. Nasipuri and S. R. Das. Multichannel CSMA with Signal Power-Based Channel Selection for Multihop Wireless Networks. In IEEE Vehicular Technology Conference (VTC) 2000.

[27] J. Padhye, R. Draves, and B. Zill. Routing in Multi-radio, Multi-hop Wireless Mesh Networks. In ACM MobiCom 2004.

[28] C. Perkins, E. Belding-Royer, and S. Das. Ad hoc On-Demand Distance Vector (AODV) Routing. In IETF RFC 3561, July 2003.

[29] R. Rozovsky and P. Kumar. SEEDEX: A MAC Protocol for Ad Hoc Networks. In ACM MobiHoc 2001.

[30] E. Shih, P. Bahl, and M. Sinclair. Wake on Wireless: An event driven power saving strategy for battery operated devices. In ACM MobiCom 2002.

[31] J. So and N. H. Vaidya. Multi-Channel MAC for Ad Hoc Networks: Handling Multi-Channel Hidden Terminals Using a Single Transceiver. In ACM MobiHoc 2004.

[32] A. Tyamaloukas and J. J. Garcia-Luna-Aceves. Channel-Hopping Multiple Access. In IEEE International Communications Conference (ICC) 2000.

[33] S.-L. Wu, C.-Y. Lin, Y.-C. Tseng, and J.-P. Sheu. A New Multi-Channel MAC Protocol with On-Demand Channel Assignment for Mobile Ad Hoc Networks. In International Symposium on Parallel Architectures, Algorithms and Networks (I-SPAN) 2000.

[34] S. Xu and T. Saadawi. Does the IEEE 802.11 MAC Protocol Work Well in Multihop Wireless Ad Hoc Networks? IEEE Communications Magazine, pages 130–137, June 2001.

[35] Z. Tang and J. J. Garcia-Luna-Aceves.

Hop-Reservation Multiple Access (HRMA) for Ad-Hoc Networks. In IEEE Infocom 1999.

 

CURRICULUM VITAE

Bruce J. Perlman, Ph.D.

bperlman@unm.edu

HOME: 8409Vintage Dr. N.E. OFFICE: School of Public Administration

Albuquerque, New Mexico, 87122 Social Sciences Bldg, Room 3004

(505) 856-5026 MSC05 3100

(505) 321-3642 (cellular) 1University of New Mexico

Albuquerque, New Mexico 87131-0001

(505) 277-3313

CURRENT POSITION:

Professor, School of Public Administration, University of New Mexico, Albuquerque, New Mexico

EDUCATION:

Ph.D. in Government, Claremont Graduate School, Claremont, California,1980

B.A. in Social Science/History, California State Polytechnic University, 1972

FELLOWSHIPS, HONORS, and AWARDS:

Hatch Chair in Law and Public Administration, 2001 to 2004, University of New Mexico, School of Public

Administration, Albuquerque, New Mexico

Presidential Lecturer in Public Administration, 1988 to 1990, University of New Mexico, School of Public Administration, Albuquerque, New Mexico

Fulbright Senior Lecturer, 1986 to 1987, Law Faculty, Public Administration Career, National Autonomous University of Honduras, Tegucigalpa, Honduras

Dissertation Fellow, Haynes Foundation, 1978 to 1980, Claremont Graduate School, Claremont, California

Interdisciplinary Crime and Delinquency Fellow, Law Enforcement Assistance Administration, 1973 to 1976, Claremont Graduate School, Claremont, California

GRANTS:

Research Grant, “The “North-South Project: the Politics of Infrastructure in Eurasia and the Middle East”,

July 2005 to June 2006, International Research and Exchange (IREX), U.S. Department of State

Research Grant, “Government Reform in Central Asia”, July, 2004, International Research and Exchange (IREX), U.S. Department of State

Research Grant, “Collaboration on Government Reform and Curriculum Development with Urals Academy of Government Service”, September 2003 to December 2003, Network of Institutes and Schools of Public Administration in Central and Eastern Europe (with Gleason and Sazanova)

Research Grant, “Why So Few Women in Information Technology: A Comparative Study”, October 2001

to September 2002, National Science Foundation, Information Technology Workforce Program, Co-Principal Investigator (with R. Varma and D. Kapur)

Academic Specialist Grant in Public Administration: Municipal Development, 1994, Colombia Externado University, United States Information Agency, Bogotá, Colombia

Academic Specialist Grant in Political Science: Public Administration, 1989, Law Faculty, Public Administration Career, National Autonomous University of Honduras, United States Information Agency, Tegucigalpa, Honduras

 

. Perlman Curriculum Vita Page 2 of 16

Inter-American Field Research Grant, 1986, National Personnel Directorate of Ecuador, Mellon Foundation, Quito, Ecuador

REPORTED RESEARCH:

Books and Monographs

The New Institutionalism in Mexico: Essays on Mexican Public Administration, B. Perlman and J.

Pineda, (eds.), Mexico City, Mexico: University of Aguascalientes, 2004

“Large Scale Job Dislocation and Reduced Economic Growth: Market Failure or Government Failure?” (with S. Nystrom), U.S. Congress, Joint Economic Committee, September 24, 1996

Journal and Other Publications

“Introduction: Energizing Facilities,” State and Local Government Review, 44:2 May 2012, pp. 37- 39

“Social Media Sites at the State and Local Level: Operational Success and Governance Failure,” State and Local Government Review, 44:1 April, 2012, pp. 67-75

“The Great Recession and Compensation Practices: New Rules for Compensation?" State and Local Government Review, 43:3 December, 2011, pp. 221-223

“Capital Ideas: National Capitals as a Fourth Jurisdiction,” State and Local Government Review, 43:2 August, 2011, pp. 140-143

“Rethinking a Megalopolis: A Metropolitan Government Proposal for the Mexico City Metro Area,” State and Local Government Review, 43:2 August, 2011, pp. 144-150 (with J. Pineda)

“Introduction: Risks and Rewards in State and Local Collaboration,” State and Local Government Review, 43:1 2011, pp 46-48

“Governance Challenges and Options for State and Local Government,” State and Local Government Review, 42:3 2010, pp 246-257

“Creative Regionalism: Governance for Stressful Times,” State and Local Government Review, 42:2, 2010 (with J. Jimenez), pp 151-156

“Introduction: New Rules and Approaches for Succession Planning,” State and Local Government Review, 42:1, 2010, pp 48-50

“Fiscal Distress and Governance Challenges: The Perfect Storm of the Fiscal Crisis,” State and Local Government Review, 41:3, 2009, pp. 201-207

“The ARRA of Our Ways,” State and Local Government Review, 41:2, 2009, pp. 120-122 “Governance and Sustainability,” State and Local Government Review, 41:1, 2009, pp. 37-38

“Cultural Determinism versus Administrative Logic: Asian Values and Administrative Reform in Kazakhstan and Uzbekistan,” International Journal of Public Administration (with G. Gleason), 30:12, 2007, pp. 1-16 (1327-1342)

“Planeación e Integración Estratégica: Un Modelo con Estudio de Caso del Sector Judicial”, Estado, Gobierno, Gestión: Revista Chilena de Administración Pública, Diciembre, 2006 (with M. Rivera), pp. 112-124

 

. Perlman Curriculum Vita Page 3 of 16

“Barely Managing: Attitudes of Information Technology Professionals on Management Technique”, The Social Science Journal, (with R. Varma), 42:4, 2005, pp. 583-594

“El Estado Actual de la Ciencia Política ”, Enlace: La Revista del Colegio Nacional de Ciencias Políticas y Administración Pública de México, June, 2005

"Issues in Comparative and International Administration," Handbook of Public Administration, Jack Rabin, (ed), 3rd Edition, New York: Marcel Dekker, Public Administration and Public Policy Series (with F. Heady and M. Rivera), 2005, pp. 605-626

“Presentación,” Conflicto Político y Negociación, A. Acosta Leon and J. Pineda (eds), Mexico City, Mexico: National College of Political Science and Public Administration, 2005

“Una Revisión Institucional Sobre Cuatro Experiencias de Servicio Civil y/o Profesional de Carrera”, Servicio Profesional de Carrera, (with J. Pineda), Vol. II, núm. 4, segundo semestre de 2005

“Comparative Perspectives on Third Generation Reform: Realignment and Misalignment in Central Asian Reform Programs”, International Public Management Review, (with G. Gleason), 6:1, 2005, pp. 100¬117

“Report on the International Public Management Network Research Conference in Rio de Janeiro, Brazil,” International Public Management Review, (with E. Caperchione and T. Virtanen), 6:1, 2005, pp. 169¬178

“The Rubric of Reform in Muslim Central Asia”, PA Times, (with G. Gleason), Vol. 27, No. 8, August, 2004, pp. 8-10

“Estudio Introductorio: La Gestión Pública Local en los Estados de América Latina”, La Gestión Pública en Gobiernos Locales: Experiencias Latinoamericanas, (J. Pineda and R. Grandinetti, eds.), pp. 11-15, México: Innovación Editorial Lagares, 2004

“Algunas Lecciones en Gestión Pública Comparativa”, Revista del Instituto de Administración Pública del Estado de México, No. 110, Enero-Abril, 2004, pp. 229-243

“Seguridad Jurídica y Moralidad Gubernamental”, Revista del Instituto de Administración Pública del Estado de México, No. 57, Enero-Abril, 2004, pp. 79-102

“Managing the War Against Terror (Part 2): Emergency Response to Terrorist Incidents”, Book Review Essay, Public Performance and Management Review, Vol. 26, No. 4, June, 2003, pp. 441-448

“Managing the War Against Terror: A Policy Framework for the United States”, Book Review Essay, Public Performance and Management Review, Vol. 26, No. 2, December, 2002, pp. 214-220

“Understanding and Improving Engineering Practice”, IEEE Technology and Society, (with R. Varma), 21:1, 2002, pp. 40-47

"Un Nuevo Modelo para la Formulación de Estrategias en la Planificación", Carta Económica Regional, Año 14, No. 78, (with M. Rivera), Octubre-Diciembre 2001, pp. 25-34

“What’s Wrong and what should be done: Comments on the Case Study, Who’s on First,” Public Performance and Management Review, Vol. 25, No. 1, September, 2001

 

. Perlman Curriculum Vita Page 4 of 16

"Bridging the Cultural Chasm: Improving Collaboration and Cooperation between the Computer and Social Sciences", Proceedings of the International Symposium on Technology and Society (with R. Varma), IEEE Catalog No. PR01209, 2001, pp.19–27

"Teaching Engineering Ethics", Proceedings of the American Society of Engineering Education Annual Conference & Exposition, Session 2793 (with R. Varma), June, 2001

"Institutional Constraints on Market Liberalization Policies in Cuba", Policy and Management Review, (with M. Rivera), Volume 1, No. 1, 2000, pp. 64-91

“Development and Reform of Post Sandinista Nicaraguan Public Administration”, Research in Public Administration, Volume 5, J. Perry (ed), JAI Press, Inc., 1999, pp. 189-231

“The Emergence of Collective Bargaining in the Guatemalan Public Sector: A Case of ‘Union Bulwarism’?” Public Personnel Management, (with T.Z. Reeves), Vol. 27, No.2, Summer, 1998

“Nicaragua: From Patrimonialism to Positivism: Assessing Past and Planned Reforms and a Theory of Configuration”, Proceedings, Conference on Civil Service Systems in Comparative Perspective, School of Public and Environmental Affairs, University of Indiana, Bloomington (April 5-8, 1997)

"La Evolución de las Gestiones Realizadas por Estados Unidos Para Controlar el Narcotráfico a Nivel Internacional Y Sus Repercusiones en Latinoamérica: Una Investigación Preliminar", (with G. LaFree), Chapter 4, La Criminología en America Latina, Birkbeck & Martínez, (eds.), Merida, Venezuela: Universidad de los Andes, Consejo de Desarrollo Científico, Humanístico y Tecnológico, 1992

"Teaching Methods in Public Administration Revisited,"(with T. De Young), Policy Studies Review, Vol. 8, No. 4, summer 1989, pp 852-858

"Bandidos: The Varieties of Latin American Banditry," Book Review, Criminal Justice Review, Vol. 13, No. 1, spring, 1988, pp. 73-75

"Modernizing the Public Service in Latin America: Paradoxes of Latin American Public Administration," International Journal of Public Administration, Vol. 12 No. 4, 1988, pp.671-704

"Who Was That Masked Man: Local Law Enforcement in New Mexico and the Southwest," Urban Resources, Vol. 4, No. 3, spring, 1987

"Ethical Dimensions of Public Policy: A Conceptual Framework," (with D. Rosenthal), Polity Vol. XIX, No. 1, autumn, 1986, pp. 56-73

"Tools for Program Evaluation", Proceedings, 9th National Conference on Teaching Public Administration, (with T. De Young), May 1, 1986, New York, New York

"Microcomputers in Public Administration: Prerequisite to Productivity", Public Productivity Review (with T. De Young), Vol. IX, Number 2-3, Summer/Fall, 1985, pp. 247-259

"Teaching Research Methods in Public Administration" Teaching Political Science: Politics in Perspective (with T. De Young), Vol. 11, No. 2, winter 1983-1984: 63

"A Political Economy Approach to Criminal Justice Education: Concerns for Study and Teaching," Proceedings, 7th National Conference on Teaching Public Administration, March 8, 1984, Albuquerque, New Mexico

 

. Perlman Curriculum Vita Page 5 of 16

"Teaching Methods for Teaching Methods in Public Administration," Proceedings, 6th National Conference on Teaching Public Administration (with T. De Young), March 1, 1983, Miami, Fl.

Technical Reports

“Actualización de la estructura orgánica del MEC: Avance en el fortalecimiento organizacional del MEC,” Programa para el desarrollo la educación de Guinea Ecuatorial (PRODEGE), Ministerio de Educación y Ciencias de Guinea Ecuatorial, April 2011

“Annual Evaluation Report,” WESST Corp JOLI Grant, U.S. Department of Health and Human Services, October, 2004

“Monitoring Report,” WESST Corp JOLI Grant, U.S. Department of Health and Human Services, April, 2003

“Hardmark Telecommunications Grant: Project Review and Evaluation,” New Mexico Department of Labor, Albuquerque New Mexico, March, 2003

“Statewide Vulnerability Assessment,” State of New Mexico Department of Public Safety Office of Emergency Services and Security, January, 2003

“Brief Comparison: Functions of Elected At Large County Officials and Appointed County

Administration,” Urban County Charter Commission, County of Bernalillo, April, 2001 to May, 2001

“Audit Response for NMDOL United States Department of Labor,” Office of Inspector General, New Mexico Department of Labor, July, 2001 to August, 2001

“Large Scale Job Dislocation and Reduced Economic Growth: Market Failure or Government Failure?” U.S. Congress, Joint Economic Committee, September 24, 1996 (with S. Nystrom)

“Guatemala: Modernization of the Public Sector and Human Resource Management,” International Bank for Reconstruction and Development (World Bank), Washington, D.C., January, 1996

“Re-engineering the Costa Rican Central Government: Action Brief,” Sigma One Corp. and USAID, Costa Rica, San Jose, Costa Rica, March, 1995

“Technical Assistance Action Plan, Nicaragua Civil Service Reform Technical Assistance Credit,” International Bank for Reconstruction and Development (World Bank), Washington, D.C., October, 1994

“Nicaraguan Civil Service Review, Modernization of the State, Civil Service Component,” International Bank for Reconstruction and Development (World Bank), Washington, D.C., August, 1994

"Evaluation of Southeast Regional Biomass Program, Chapter IV: Lessons Learned," Sandia National Laboratories and DOE/EE/OPA, Albuquerque, New Mexico, May, 1994

"Analysis of EE Sector Program Evaluation Indicators, Key Sector, and Key Written Sources," Sandia National Laboratories and DOE/EE/OPA, Albuquerque, New Mexico, March, 1994

"Estudio de Instituciones Claves," Development Training Project, Creative Associates International Inc. and USAID/GDO, Nicaragua, October, 1994

"Manual de Funciones del Ministerio de Educación," Education and Human Resources Training Support, Juarez and Associates and USAID/GDO, Nicaragua, March, 1993

 

. Perlman Curriculum Vita Page 6 of 16

"The Future of Municipal Service Delivery in Phoenix and Maricopa County", Arizona State University, School of Public Affairs, April, 1992

"Sub Sector Assessment for Basic Education", Academy for Educational Development and USAID/GDO, Nicaragua, August, 1991

"A Paper on Issues in the Study of Problems Concerning and the Design of Programs for the Reintegration of Salvadoran Ex Combatants into Civilian Life", Creative Associates International Inc. and USAID/IRD, El Salvador, December, 1990

"A Report and Recommendations Including Training Plan on Guatemala Public Administration, Financial Management and Anti-corruption," USAID/HRD Guatemala, June, 1990

"Citizen Satisfaction with Police Service Delivery," University of New Mexico, Institute for Criminal Justice Studies, May, 1990

"Drug Enforcement on the Argentine Border: Report and Recommendations," Procuración de la Nación y Policía Federal de Argentina, August, 1989

"Training Plan for APRE Leadership," USAID/HRD, El Salvador, July, 1987

"Estudio Administrativo del Poder Judicial de Honduras, USAID/ODI, Honduras," December, 1986

"Program Summary: Programa de Capacitación en Administración Pública, Grupo Ecuatoriano", USAID/HRD, Ecuador" August, 1985

"Management Information Systems Evaluation Study," New Mexico State Highway Department, May, 1984.

"Honor Farm Feasibility Study," Bernalillo County Detention Center, December, 1983

SELECTED PAPERS AND PRESENTATIONS:

Academic Papers and Presentations

Paper, “La Agenda Pendiente en el Sistema Institucional del Estado Mexicano ”, Panel Convener, “Análisis Institucional, Profesionalización, y Modernización del Sector Público en México”, X International Conference on Reform of the State and Public Administration, Latin American Center for Development Administration, Santiago, Chile, Pública, November 18-21, 2005

Paper, “Comparative Perspectives on Third Generation Reform: Realignment and Misalignment in Central Asian Reform Programs”, Conference of the International Public Management Network at the Getulio Vargas School of Public and Business Administration, Rio De Janeiro, Brazil, November 17-19, 2004

Paper, “Formalismo y Pragmatismo en las Administraciones Públicas en el Continente Americano: Filosofía y Práctica en el Norte y en el Sur”, IX International Conference on Reform of the State and Public Administration, Latin American Center for Development Administration, Instituto Nacional de Administración Pública, Madrid, Spain, November 2-5, 2004

Paper, Una revisión inicial a la nueva Ley del Servicio Profesional de Carrera en la Administración Pública Federal de México: Oportunidades y Desafíos en la Medición de Resultados, VIII Internacional Conference on Reform of the State and Public Administration, Latin American Center for Development Administration, Panama City, Panama, November 28-31, 2003

 

. Perlman Curriculum Vita Page 7 of 16

Paper, “Algunas Lecciones en Gestión Pública Comparativa”, VIII Internacional Conference on Reform of

the State and Public Administration, Latin American Center for Development Administration, Panama City, Panama, November 28-31, 2003

Paper, Hazards vs. Targets in the State of New Mexico: Fitting Local Problems to National Priorities in the Homeland Security Planning Process, 64th Annual Conference of the American Society of Public Administration, Washington, D.C. March 14-19, 2003

Paper, “Barely Managing: Perceptions of Computer Scientists on Management and the Social Sciences in Public Institutions”, 44th Annual Conference of the Western Social Science Association, Albuquerque, New Mexico, April 10-13, 2002

Paper, “Managing Computer Scientists in Government Service”, 63rd Annual Conference of the American Society of Public Administration, Phoenix, Arizona, March 23-26, 2002

Paper, Seguridad Jurídica y Capital Social, 6th Ibero-American Juridical Conference, National Judicature Council, San Salvador, El Salvador, September 3-6, 2000

Paper, "Managing on the Border", Panel Convener, "Managing on the Border," 53rd National Conference, American Society for Public Administration, Chicago, Illinois, April, 1992

Panel Convener, "Comparative Public Administration" Western Social Science Association, 33rd Annual Conference, Denver, Colorado, April, 1992

Paper, "Issues in Multi-cultural Training," Western Social Science Association, 31st Annual Conference, Portland, Oregon, April, 1990

Paper, "Police Professionalism: Guild Ethics or Professional Ethics?" 50th National Conference, American Society for Public Administration, Miami, Florida, April, 1989

Paper, "Technical Assistance in Civil Conflict Areas," 50th National Conference, American Society for Public Administration, Miami, Florida, April, 1989

Presentation, "What to Do With Excellence after You Find It," New Mexico Municipal League, City Managers' Section, Taos, New Mexico, April, 1989

Presentation, "Performance and Performing," International Municipal Clerks National Conference, Albuquerque, New Mexico, January, 1989

Paper "Justice under the Gun: Judicial Administration in El Salvador, "Panel Convener, "Administration in

Conflict," 49th National Conference, American Society for Public Administration, Portland, Oregon, April, 1988

Paper, "Law Enforcement Firearms Training, "Panel Convener, "Democracy Under the Gun: A Round Table on Firearms Policy, Administration and Training." 48th National Conference, American Society for Public Administration, Boston, Massachusetts, April, 1987

Paper and Computer Demonstration, "Using Microcomputers to Make Decisions," 9th Annual National Conference on Teaching Public Administration, White Plains, New York, May, 1986

Paper, "The Role of Local Police in Combating Terrorism", 47th National Conference, American Society for Public Administration, Anaheim, California, April, 1986

 

. Perlman Curriculum Vita Page 8 of 16

Paper, "Varieties of Violence", Panel Convener, "Varieties of Violence", 47th National Conference, American Society for Public Administration, Anaheim, California, March, 1986

Discussant, "Air Quality in Albuquerque", New Mexico Chapter, American Society for Public Administration, Albuquerque, New Mexico, February, 1986

Paper, "La Evolución de las Gestiones Realizadas por Estados Unidos Para Controlar el Narcotráfico a Nivel Internacional Y Sus Repercusiones en Latinoamérica: Una Investigación Preliminar," Seminario Internacional, La Criminología en America Latina: Balance y Perspectivas, Universidad de los Andes, Mérida, Venezuela, November, 1985 (with G. LaFree)

Paper, "New Techniques in Public Administration," Panel Convener, "New Techniques in Public Administration", 8th Annual Conference on Teaching Public Administration, St. Louis, Missouri, April, 1985

Paper, "Computers in Law Enforcement: A Review," Western Political Science Association, Annual Conference, Las Vegas, Nevada, March, 1985

Paper, "Law Enforcement Intelligence Units: The Myth of Strategic Intelligence, Western Political Science Association, Annual Conference, Las Vegas, Nevada, March, 1985

Paper, "Police Assessment in One Southwestern City," American Society for Public Administration, National Conference, Denver, Colorado, April, 1984

Paper, "Reasons for Joining the Police Reserves: Experience in One Southwestern State," Western Social Science Association, 26th Annual Conference, San Diego, California, April, 1984

Paper, "Education and Performance in Assessment Centers: Experience in One Southwestern Police Department," Western Social Science Association, 26th Annual Conference, San Diego, California, April, 1984

Paper, "Ethical Dimensions of Public Policy and Public Administration," Southwestern Political Science Association Annual Conference, Fort Worth, Texas, March, 1984 (with D. Rosenthal)

Paper, "Public Impact on Corrections," New Mexico Association of Criminal Justice Professionals Annual Conference, Albuquerque, New Mexico, October, 1983

Paper, "Care and Feeding of the Violent Juvenile Offender," American Society for Public Administration, Region VIII Conference, Phoenix, Arizona, April, 1983

Paper, "Collective Bargaining in New Mexico Corrections: The Strange Case of AFSCME vs. NMCWA," Western Social Science Association, 25th Annual Conference, Albuquerque, New Mexico, April, 1983

Paper, "The Moral Fiction of Expertise: The Cases of the Policy and Management Sciences," Western Social Science Association, 25th Annual Conference, Albuquerque, New Mexico, April, 1983 (with D. Rosenthal)

Paper, "Reaganomics and Corrections: Correctional Policy Making In An Era of New Federalism" American Society for Public Administration, Region VIII Conference, Denver, Colorado, October, 1982. Paper, "Productivity Enhancing Programs: Salary Structure for Rewarding Exceptional Performance," International Conference, San Diego, California October, 1982

 

. Perlman Curriculum Vita Page 9 of 16

Paper, "The Future of Corrections," Panel Convener, "The Future of Corrections", American Society for Public Administration Region X and Western Governmental Research Association Joint Annual Conference, Anaheim, California, October, 1982

Invited Addresses

Invited Address, “Políticas Públicas y Educación Internacional ,” IV Annual Latin American Network

Conference, Puebla, México, September 2, 2011

Invited Address, “Transformación organizacional en el estado contemporáneo,” El congreso

internacional en gerencia de proyectos y mejoramiento organizacional, EAN University, Bogotá, Colombia, September 24, 2010

Invited Address, El Estado y Las Políticas Públicas, Bases Para un Proyecto de Nación, Conferencia

Inaugural de la Cuarta Promoción, Maestría en Políticas Públicas, Universidad Rafael Landivar, Guatemala City, Guatemala, July 7, 2003

Invited Address, El Estado Actual de la Ciencia Política, Semana de Administración Pública, Escuela Nacional de Estudios Profesionales UNAM-Acatlán, México, June 5, 2003

Invited Address, Conferencia Magistral, Gerencia Pública y Rediseño Institucional en las Administraciones Públicas”, Segundo Congreso Internacional en Ciencia Política y Administración Pública, Colegio Nacional de Ciencia Política y Administración Pública de México, Mexico City, Mexico, November, 13-14, 2002

Invited Address, Algunas Lecciones sobre Administración Pública Comparada, Seminar on Comparative Methodology in Public Management, Universidad de Chile, Santiago, Chile, August26-28, 2002

Invited Address, La Dicotomía entre la Política y la Administración, Universidad de Guadalajara, Guadalajara, México, July, 1994

Invited Address, "The Philosophy of Social Science Methods" USIA Guatemala, 1st USAC-UNM Social Science Seminar, Guatemala, February, 1989

Invited Address, "U.S. Drug Enforcement Policy, 3rd U.N. Conference on Trans-national Crime, Albuquerque, New Mexico, November, 1988

Invited Address, "La Lucha Contra Las Drogas," Procuración de la Nación and Policía Federal de la Republica, Seminario Sobre Narcotráfico, Buenos Aires, Argentina, May, 1988

Invited Address, "Power, Politics, and Professional Groups, New Mexico Dietician Association Annual Meeting, Albuquerque, New Mexico, April, 1986

Invited Address, "Market Research", Work Unlimited Marketing Workshop, Albuquerque, New Mexico, February, 1986

Invited Address, "How to Fail Miserably with Your Staff", New Mexico Association of Counties, Treasurers Affiliate Legislative Conference, Santa Fe, New Mexico, January, 1986

Invited Address, "Burnout: Don't Make an Ash of Yourself", New Mexico Municipal League Annual Conference, Las Cruces, New Mexico, September, 1985

ACADEMIC AND PROFESSIONAL EXPERIENCE:

Chief Financial Officer, Office of the State Treasurer, State of New Mexico, Santa Fe, New Mexico, July,

2009 to January, 2010 (TDY from Governor’s Office; on leave from UNM)

 

. Perlman Curriculum Vita Page 10 of 16

Deputy Chief of Staff, Office of the Governor, State of New Mexico, Santa Fe, New Mexico, July, 2008 to July, 2009 (TDY to State Treasurer in July 2009; on leave from UNM)

Chief Administrative Officer, City of Albuquerque, Albuquerque, New Mexico, December, 2005 to June 2008 (on leave from UNM)

Director, School of Public Administration, University of New Mexico, May, 2004 to December, 2005

Chief of Party and Resident Advisor, Project for Reform of the Justice Sector, Inter American Development Bank and Associates for Rural Development, May, 1999 to July, 2000 (on leave from UNM)

Interim Director, School of Public Administration, University of New Mexico, May, 1994 to January, 1996

Visiting Associate Professor, School of Public Affairs, Arizona State University, January, 1991 to June, 1992

Associate Professor, School of Public Administration, University of New Mexico, August, 1989 to the January, 2006 (took leave to fill other posts)

Associate Director for Programs in Public Administration, Office of International Technical Cooperation, University of New Mexico, July, 1988 to July, 1990

Associate Director, Division of Public Administration, University of New Mexico, August, 1988 to August, 1989

Director, Master of Arts in Public Administration in Spanish, Division of Public Administration, University of New Mexico, January, 1987 to August, 1988

Assistant Professor, Division of Public Administration, University of New Mexico, August, 1982 to May, 1988

Academic Coordinator, Programa de Capacitación en Efectividad Organizacional, Office of International Technical Assistance, University of New Mexico, May, 1986 to February, 1987

Project Manager, A.C.O.P.S. Integration Project, Albuquerque Police Department, January, 1984 to July, 1985

Academic Coordinator, Programa de Capacitación en Administración Pública, Latin American Programs

in Education and Division of Public Administration, University of New Mexico, May, 1985 to August, 1985

Director, Policy and Systems Development, Albuquerque Police Department, January, 1983 to January, 1984 (on leave from UNM)

Lecturer, California State University, Dominguez Hills, September, 1981 to June, 1982

Program Manager, University of Southern California, June, 1980 to August, 1981

Adjunct Faculty, California State University, Dominguez Hills, September, 1979 to June, 1981

Research Associate, Executive Management Program: Humanities Component, Claremont Graduate School. September, 1977 to June, 1979

 

. Perlman Curriculum Vita Page 11 of 16

Evaluation Consultant, Noble and Associates, August, 1976 to September, 1977

Teaching Assistant, Sociology Department, Pitzer College, Claremont Colleges, September, 1975 to June, 1976

Director, Project Upstream, Los Angeles Unified and Pomona Unified School Districts, September, 1974 to June, 1975

CONSULTING AND CONTRACT RESEARCH:

International:

Academy for Educational Development, PRODEGE, Organizational Strengthening Consultant for the

Ministry of Education of Equatorial Guinea, October, 2010 to September, 2011

Inter American Development Bank and Associates for Rural Development, Project for Reform of the

Justice Sector, Chief of Party and Resident Advisor, May, 1999 to July, 2000 (on leave from UNM)

The International Bank for Reconstruction and Development (World Bank) and the Government of Guatemala, Ministry of Finance, Department of External Financing, Guatemala: Modernization of the Public Sector and Human Resource Management, Public Administration Specialist. Guatemala, Guatemala, July, 1995 to December, 1995

USAID Costa Rica, Office of General Development and Sigma One Corporation, Public Administration Specialist for State Reform for The Technical Secretariat of State Transformation and the Ministry of Planning, June, 1995 to September, 1995

The International Bank for Reconstruction and Development (World Bank), Nicaragua Reform of the State Project, Civil Service Modernization Component, Public Administration Specialist, Managua, Nicaragua, May, 1994 to August, 1994

USAID Nicaragua, Office of General Development, and Juárez and Assoc., Inc., Public Administration Specialist, BASE Project, Managua, Nicaragua, January, 1994 to April, 1994

USAID Nicaragua, Office of General Development, and Creative Assoc. Int'l., Inc., Public Administration Specialist, Development Training Project, August, 1993 to October, 1993

USAID Guatemala, Office of Education and Human Resources and Juárez and Assoc., Inc., Chief of Party and Education Administration Specialist, Rural Primary Education Improvement Project Final Evaluation, Guatemala City, Guatemala, January, 1993 to April, 1993

USAID Nicaragua, General Development Office, and Juárez and Associates Inc., Administration

Specialist, Functions Manual for Ministry of Education. Managua, Nicaragua, September, 1992 to March, 1993

USAID Costa Rica, Projects Development Office, and Sigma One Corporation, Public Law and Administration Consultant, Costa Rica State Public Sector Reform Project. San Jose, Costa Rica, November, 1992

USAID Guatemala, Office of Education and Human Resources, and Creative Associates International, Inc., Education Administration Specialist, Mid-term Evaluation of the Basic Education Strengthening Project, Guatemala City, Guatemala, July, 1992 to September, 1992

USAID Nicaragua, General Development Office, and Juárez and Associates Inc., Institutional Strengthening of the Nicaraguan Ministry of Education, Managua, Nicaragua, June, 1991 to March, 1992

 

. Perlman Curriculum Vita Page 12 of 16

USAID Nicaragua, General Development Office, and Juárez and Associates Inc., Project Paper Team Leader/Public Administration Specialist, Project Paper. Managua, Nicaragua, December, 1991 to January, 1992

The International Bank for Reconstruction and Development (World Bank), El Salvador Social Sector Project, Specialist in Public Administration and Organizational Development. San Salvador, El Salvador, October, 1991

USAID Nicaragua, General Development Office, and Juárez and Associates Inc., Sub Sector Assessment in Basic Education. Managua, Nicaragua, February, 1991 to May, 1991

USAID El Salvador, Office of Infrastructure and Rural Development, and Creative Associates International Inc., Study on the Reintegration of Ex Combatants into Salvadoran Civil Society, San Salvador, El Salvador, November, 1990 to December, 1990

USAID Guatemala, Office of Human Resource Development, Development Training and Support

Program, Materials Inventory and Operations Review of Controllers General Offices of Costa Rica and Spain. San Jose, Costa Rica; Guatemala, Guatemala, June, 1990 to July, 1990

USAID Guatemala, Office of Human Resource Development, Development Training and Support Program, Diagnosis and Needs Assessment for Administrative Reform for the Controller General's Office of Guatemala, Guatemala, Guatemala, November, 1989 to June, 1990

USAID El Salvador, Infrastructure and Rural Development Program, Administrative Reform and

Management Training and Technical Assistance Program for Engineers, Albuquerque, New Mexico; Antigua, Guatemala; San Salvador, El Salvador, January, 1989 to June, 1989

USIA Costa Rica and USAID Regional Administration of Justice Office, Court Administration Course, Albuquerque, New Mexico, May, 1989 to August, 1989

USIA Guatemala, University of San Carlos, Guatemala, UNM, Partnership Program in Political Science, Guatemala, Guatemala; Albuquerque, New Mexico, January, 1989 to December, 1989

Procuración de la Nación y Policía Federal de Argentina, Study on Organizational Cooperation in Drug

Enforcement on the Argentine-Bolivian Border. Buenos Aires, Argentina, Salta, Argentina; Positos, Bolivia, August, 1989

USAID Guatemala, Office of Human Resources, National Institute of Public Administration of Guatemala, Training Program for Executives. Guatemala, Guatemala, July, 1988

USAID El Salvador, Office of Education and Training, and UNM Office of International Technical Assistance, El Mejoramiento de la Productividad en el Sector Público, una Evaluación y Taller de Trabajo. San Salvador, El Salvador, June, 1987 to July, 1987

USAID El Salvador, Office of Democratic Initiatives, and Proyecto Ilanud-FIU-El Salvador, Estudio Sobre El Órgano Judicial en El Salvador. San Salvador, El Salvador, May, 1987 to June, 1987

USAID El Salvador, Office of Education and Training and UNM Office of International Technical Assistance, Proyecto para Revitalización de Educación. San Salvador, El Salvador, March, 1987 to April 1987

USAID El Salvador, Office of Education and Training, and, UNM Office of International Technical

Assistance, Program in Organizational Effectiveness, San Salvador, El Salvador; Albuquerque, New Mexico, May, 1986 to February, 1987

 

. Perlman Curriculum Vita Page 13 of 16

USAID Honduras, Office of Democratic Initiatives, and Proyecto Ilanud-FIU-Honduras, Estudio de La Administración de Justicia, Estudio Administrativo. Tegucigalpa, Honduras, November, 1986 to December, 1986

USAID Ecuador, Office of Education and Training, and UNM Latin American Programs in Education and Division of Public Administration, Program in Public Administration, Quito, Ecuador; Albuquerque, New Mexico, May, 1985 to August, 1985

Domestic:

City of Albuquerque, Albuquerque Fire Department, Update and Review of Standard Operating Guidelines,

October, 2004 to December, 2005

City of Albuquerque, Environmental Health Department, Metropolitan Medical Response System Consultant, May, 2004 to December, 2005

Techonex, Program Evaluator for the Hardmark Telecommunications Implementation Grant, United States Department of Labor, October, 2003 to December, 2005

City of Albuquerque, Office of Emergency Preparedness, Emergency Planning and Management Consultant, July, 2002 to January, 2005

State of New Mexico Department of Public Safety Office of Emergency Services and Security, Consultant for Statewide Vulnerability Assessment, July, 2002 to January, 2003

Women’s Economic Self Sufficiency Team Corp. (WESST Corp), Program Evaluator for the JOLI Micro 

enterprise Development Grant, United States Department of Health and Human Services, Albuquerque New Mexico, December, 2002 to present

Central Area Workforce Investment Board, Workforce Investment Act, Grants Administration Consultant, Albuquerque New Mexico, October, 2002 to December 2003

New Mexico Department of Labor, Program Evaluator for the Hardmark Telecommunications Demonstration Grant, United States Department of Labor Albuquerque New Mexico, September, 2002 to March, 2003

State of New Mexico Department of Public Safety Office of Emergency Services and Security, Consultant for Statewide Vulnerability Assessment, July, 2002 to January, 2003

Urban County Charter Commission, County of Bernalillo, Consultant for Functional Comparison of Elected and Appointed Officials, April-May, 2001

New Mexico Department of Labor, Consultant for Preparation of Audit Response to United States Department of Labor, Office of Inspector General, July-Aug, 2001

Sandia National Laboratories and the Federal Aviation Administration, Flight Services Division. Consultant for Organizational Systems, Aviation Technical Oversight System, September, 1997 to February, 1999

Sandia National Laboratories and the Federal Highway Administration, Office of Motor Carriers, Consultant for Field Research and International Administration, Advanced Technologies for International, Intermodal, Ports of Entry, July, 1996 to February 1997

 

. Perlman Curriculum Vita Page 14 of 16

Sandia National Laboratories and Department of Energy, Conservation and Renewable Energy, Office of Planning and Evaluation, Consultant in Evaluation, July, 1993 to June, 1994

New Mexico Municipal League and New Mexico Association of Counties, Mediator and Chair, Joint Executive Board Meeting for Legislative Agenda Setting, December, 1984 to 1990

City of Albuquerque, New Mexico, Albuquerque Police Department, Citizen Satisfaction with Service Delivery Study. January to 1990, May, 1990

Council for International Development, Washington, D.C., Proposal for Afghanistan Opium Reduction Project, April, 1990

New Mexico State Legislative Finance Committee, Planning Retreat, February, 1990

City of Santa Fe, New Mexico, City Council Planning Retreat, February, 1990

City of Silver City, New Mexico, City Manager and Department Directors' Planning Retreat, July, 1989

New Mexico Municipal League, Statewide Training Needs Assessment, February, 1989 to August, 1989

Silva & Silva, El Paso, Texas, Expert Witness, Pérez v. F.B.I., December, 1989 to May, 1989

City of Albuquerque, Office of Senior Affairs, Strategic Planning Review, May, 1986 to August, 1986

City of Albuquerque, New Mexico, Environmental Health Division, Project Assistance for Inspections and Maintenance Program. July, 1985

New Mexico Attorney General's Office, Expert Testimony, Childers v. Hobbs. April, 1985, to June, 1985

County of Bernalillo, New Mexico, Bernalillo County Detention Center, Honor Farm Feasibility Study. May, 1985

Communication Companies International, Malibu, California, Corporate Reorganization Study and Executive Development Program, June, 1984

Ergodotics Inc., La Verne, California, Executive Development Program, June, 1983

New Mexico State Highway Department, Management Information Systems Evaluation Study, October, 1983 to January, 1984

UNIVERSITY, PROFESSIONAL, AND COMMUNITY SERVICE:

Member, Impact Fee Review Committee, City of Albuquerque, Office of the Mayor, October 2010 to

present

Editor, Governance Matters Section, State and Local Government Review, January, 2008 to present Member, Faculty Senate Government Relations Committee August, 2010, to present

Member, Editorial Board, Public Performance and Management Review, May, 2003 to present Editor in Chief, Journal of Public Administration Education, August 2004 to July 2007

Member, Executive Board, Section on International and Comparative Administration, American Society for Public Administration, November, August 2004 to September, 2007

 

. Perlman Curriculum Vita Page 15 of 16

Associate Editor for Latin America, International Public Management Review, October, 2004 to November, 2007

Associate Editor for Latin America, Encyclopedia of Public Administration, September, 2004 to November, 2007

Member, International Awards Committee, American Society for Public Administration, June, 2004 to July 2007

Faculty Senator, School of Public Administration, University of New Mexico, August, 2003 to September, 2004

Member, Faculty Senate Government Relations Committee August, 2002, May to 2004

Member, Editorial Board, Public Performance and Management Review, May, 2003 to present

Faculty Intern, Office of the Vice President for Student Affairs, University of New Mexico, January, 2003 to December, 2003

Member, Executive Board, Section on Public Administration Research, American Society for Public Administration, May, 2002 to present

Member, National Conference Program Committee, American Society for Public Administration, June, 2002 to March, 2003

Representative, Faculty Senate Curriculum Committee, School of Public Administration, University of New Mexico, August, 2001 to May, 2003

Chair, Mayor’s Transition Team for the Albuquerque Police Department, November, 2001 to December, 2002

Member, City of Albuquerque Mayor's Law Enforcement Transition Team, December, 1996 to February, 1997

Member, Community Oriented Policing Steering Committee, City of Albuquerque, January, 1995 to December, 1997

Member, Faculty Senate Graduate Committee, September, 1993 to May, 1997

Member, President's Task Force on International Issues, American Society for Public Administration, September, 1992 to present

Member, Office of U.S. Senator Jeff Bingaman, U.S. - Mexico Border Business Task Force, September, 1990 to December, 1990

Member, City of Albuquerque Mayor's Law Enforcement Transition Team, December, 1989 to February, 1990

Member, City of Albuquerque Mayor's Task Force on the Police, January, 1988 to May 1988

Chairman of the Board and Faculty Associate, UNM Institute for Criminal Justice Studies, January, 1987 to December, 1989

 

. Perlman Curriculum Vita Page 16 of 16

Member, National Conference Program Committee, American Society for Public Administration, March, 1986 to March, 1988

National Conference Program Chair, American Society for Public Administration, Section on Criminal Justice Administration, March, 1986 to March, 1987

Member, Publications Subcommittee, Latin American Institute, University of New Mexico, August, 1986 to May, 1988

Member, University Budget Review Committee, August, 1985 to May, 1987

Regional Representative, Western Governmental Research Association; Judge, Samuel C. May Research Paper Award, January, 1985 to December, 1985

Steering Committee, City of Albuquerque, Mayor's Task Force on Alcohol and Drug Abuse, January, 1985 to June, 1985

Executive Council Member, New Mexico Chapter, American Society for Public Administration. January, 1984 to May, 1986

National Program Committee, American Correctional Association, January, 1984 to December, 1984 Vice President, New Mexico Correctional Association. January, 1983 to June, 1984

Member, Bernalillo County Adult Local Selection Panel New Mexico State Department of Corrections, January, 1983 to May, 1984

Executive Board Member, Albuquerque Police Department Reserve Corps, March, 1983 to June, 1985 Member, Educational Leave Committee, Albuquerque Police Department, January, 1983 to June, 1987

University Speakers Bureau Talks: for Kiwanis, Optimists, Exchange Clubs, N.M. Computer Society and others

LANGUAGES:

Speaking Fluency: English and Spanish

 

Abstrak . Mach Zhender Inferometer menggunakan Semiconductor Amplifier optik ( SOA ) dan coupler optik . Perangkat ini digunakan untuk menghasilkan fungsi logika ( AND, XOR ) & multiplexer dan Encoder yang diperoleh menggunakan perangkat ini di Optical Pohon Arsitektur . Simulasi Encoder dan Multiplexer dilakukan pada tingkat 10 Gbit / s dan keduanya disimulasikan untuk input kombinasi logis yang berbeda . Simulasi menunjukkan bahwa Perangkat ini cocok untuk beroperasi pada bit rate yang lebih tinggi dan juga untuk entitas logis yang berbeda .

Banyak aliran data kecepatan rendah dapat di-multiplexing menjadi satu kecepatan tinggi aliran melalui waktu Optical division multiplexing ( OTDM ) , sehingga setiap saluran masukan mentransmisikan data dalam ditugaskan slot waktu . Tugas ini dilakukan dengan cepat beralih multiplexer ( mux ) . The routing data yang berbeda aliran di akhir link TDM dilakukan oleh switch demultiplexer ( demux ) dan demultiplexer ini digunakan menggunakan MZI saklar terdiri semikonduktor amplifier optik ( SOA ) dan optik coupler . Dalam bab empat channel OTDM disimulasikan pada 40 Gbit / s dan selanjutnya diselidiki dampak dari kekuatan sinyal , lebar pulsa dan kekuatan sinyal kontrol pada BER . Indeks Situs Syarat . Semua saklar optik , Mach – Zehnder interferometer ( MZI ) , Semiconductor amplifier optik ( SOA ) , Skema Switching , analisis Spectrum .

I PENDAHULUAN

Dalam era informasi , teknologi melihat tanpa henti permintaan untuk jaringan kapasitas yang lebih tinggi dengan biaya lebih rendah . Teknologi komunikasi optik telah berkembang pesat ke mencapai kapasitas transmisi yang lebih besar dan transmisi lebih lama jarak . Untuk itu kecepatan data tersebut dapat dicapai jika data tetap dalam domain optik menghilangkan kebutuhan untuk mengkonversi sinyal optik . Oleh karena itu , untuk berhasil dapat mencapai kecepatan data yang lebih tinggi , jaringan optik canggih akan mewajibkan semua pemrosesan sinyal ultra cepat optik seperti konversi panjang gelombang , logika optik dan aritmatika pengolahan, fungsi add -drop , dll Berbagai arsitektur , algoritma , operasi logis dan aritmatika telah diusulkan di bidang optik / komputasi optoelektronik dan pemrosesan paralel dalam tiga dekade terakhir . nonlinier optik lingkaran cermin ( NOLM ) memberikan dukungan besar untuk optik switching yang berbasis semua logika optik dan pengolahan aljabar di mana mekanisme switching didasarkan pada serat Kerr nonlinier .

Solusi yang lebih efisien dan kompak dapat diwujudkan oleh semua switching optik dalam semikonduktor amplifier optik ( SOA ) di mana koefisien non linear jauh lebih tinggi . Konfigurasi switching berbasis Berbagai SOA telah menunjukkan sebelumnya seperti Tetrahertz asimetris optik demultiplexers ( kodok ) , inferometers nonlinier ultra-cepat ( Unis ) dan Mach - Zehnder inferometers ( MZIs ) . Diantara topologi yang berbeda , switch MZI monolithically terintegrasi merupakan solusi yang paling menjanjikan karena mereka kompak ukuran, stabilitas termal dan daya rendah . Dalam komputasi optik , sistem interkoneksi optik adalah primitif yang merupakan berbagai algoritma optik dan arsitektur . Arsitektur pohon optik ( OTA ) juga mengambil peran penting dalam hal ini . Jadi dalam era teknologi berubah dengan cepat kami mewakili skema alternatif baru yang mengeksploitasi keunggulan dari kedua SOA - MZI dan OTA , untuk implementasi dari semua logika dan aritmatika operasi paralel optik biner data.

1,1 Mach Zehnder Inferometer

The Mach - Zehnder Inferometer adalah alat yang digunakan untuk menentukan pergeseran fasa yang disebabkan oleh contoh kecil yang ditempatkan di jalan salah satu dari dua balok collimated dari dua cahaya koheren sumber . A Mach - Zehnder Interferometer dibuat dari dua skrup dihubungkan dengan lengan panjang optik yang tidak sama . itu Mach - Zehnder Interferometer

 

memiliki dua port input dan dua output port . Lampu dibagi dalam dua lengan input coupler interferometer , dan mereka kemudian digabungkan dalam coupler output interferometer . Panjang optik dua lengan yang tidak sama , sehingga fase sesuai dengan menunda Fig.1.1 menjadi fungsi dari panjang gelombang . relatif fase cahaya di dua port input output coupler adalah Oleh karena itu, fungsi dari panjang gelombang . Sebagai tahap keterlambatan ( d ) meningkat , siklus MZI antara lintas negara , di mana sebagian besar cahaya muncul dalam Waveguide pada sisi yang sama dengan input , dan negara bar , di mana sebagian besar bergerak cahaya untuk Waveguide di sisi lain .

Fig.1.1 Mach - Zehnder Inferometer

Sistem Komunikasi Serat Optik

Kinerja Menggunakan MZI Switching

Sachin Kumar , Indu Bala Pauria , Anoop Singhal

Kinerja Sistem Komunikasi Serat Optik Menggunakan MZI Switching

99

1.2 Semiconductor Optical Amplifier

Semiconductor amplifier optik adalah amplifier yang menggunakan semikonduktor untuk menyediakan media gain . desain terbaru termasuk pelapis anti - reflektif dan Waveguide miring dan jendela daerah yang dapat mengurangi refleksi akhir tatap kurang dari 0,001 % . Karena ini menciptakan hilangnya daya dari rongga yang lebih besar dari keuntungan mencegah penguat dari bertindak sebagai laser . Amplifier tersebut sering digunakan dalam sistem telekomunikasi dalam bentuk serat – pigtailed komponen , yang beroperasi pada panjang gelombang sinyal antara 0,85 ƒÊm dan 1,6 ƒÊm dan menghasilkan keuntungan hingga 30 dB . itu semikonduktor amplifier optik adalah ukuran kecil dan elektrik dipompa . Hal ini dapat berpotensi lebih murah dibandingkan EDFA dan dapat diintegrasikan dengan laser semikonduktor , modulator , dll Namun , kinerja yang masih belum 5 sebanding dengan EDFA . SOA memiliki noise yang lebih tinggi , lebih rendah gain , dan moderat ketergantungan polarisasi dan tinggi nonlinier dengan cepat sementara waktu . Ini berasal dari nanodetik pendek atau kurang seumur hidup atas negara , sehingga gain bereaksi dengan cepat terhadap perubahan pompa atau kekuatan sinyal dan Perubahan keuntungan juga karena perubahan fase yang dapat mendistorsi sinyal . Nonlinier ini menyajikan paling parah masalah bagi aplikasi komunikasi optik . Namun itu memberikan kemungkinan untuk keuntungan dalam panjang gelombang yang berbeda daerah membentuk EDFA .

1.3 Kategori switch

1.3.1 MZI Beralih

The Mach - Zehnder interferometer ( MZI ) switch berbasis

terdiri dari splitter 3 dB dan Combiner 3 dB , dihubungkan dengan dua lengan interferometer . Dengan mengubah bias efektif Indeks dari salah satu lengan , perbedaan fase di awal dari combiner dapat diubah , sehingga lampu dari port satu output yang lain . Switch ini memiliki keuntungan bahwa fase pergeseran bagian dan modus kopling bagian dipisahkan , sehingga keduanya dapat dioptimalkan secara terpisah .

Beralih berbasis Fig.1.2 MZI

 

Perubahan indeks bias kecil efektif dalam interferometer cukup untuk switching . Kerugiannya adalah panjangnya dan akurat perubahan indeks bias yang diperlukan untuk switching. Ketika multimode gangguan skrup yang dipekerjakan sebagai 3 dB splitter dan combiner , toleran fabrikasi dan polarisasi gelombang sensitif membimbing struktur adalah diperoleh . Sebuah sinyal data daya rendah difokuskan ke pusat masukan waveguide sedemikian rupa sehingga terbagi menjadi dua bagian yang sama di Y - junction daya splitter . Kedua balok kemudian merambat melalui dua lengan Mach - Zehnder dan bergabung kembali konstruktif pada 6 keluaran Y - junction power combiner dan merambat sepanjang waveguide output. Sebuah kontrol daya tinggi sinyal juga difokuskan ke salah satu pemandu gelombang luar untuk menghasilkan perubahan indeks bias nonlinier di Waveguide melalui nonlinear efek Kerr optik . Ini menghasilkan fase perbedaan antara dua sinyal data pada output Y junction menyebabkan mereka untuk mengganggu destruktif ketika perbedaan fasa antara mereka adalah radian TC . Di bawah ini kondisi , sinyal data yang digabungkan ke dalam mode radiasi dan output turun ke nol . Selanjutnya perangkat dapat digunakan sebagai modulator .

1.3.2 DC Beralih

Dalam switch directional coupler dua pandu gelombang yang berdekatan adalah dirancang sedemikian rupa , bahwa cahaya dapat ditransfer dari satu Waveguide ke yang lain dengan kopling . Switching diperoleh dengan benar menyesuaikan indeks bias efektif salah satu pandu gelombang . Untuk beralih hanya indeks bias kecil perubahan yang dibutuhkan .

Gambar 1.3 Directional Coupler Beralih

Untuk transfer yang baik dari cahaya , panjang kopling akurat adalah diperlukan . Karena panjang ini biasanya polarisasi dan tergantung panjang gelombang dan sangat dipengaruhi oleh fabrikasi penyimpangan ( kedalaman etch , Waveguide spasi ) , switch yang baik kinerja sulit diperoleh .

1.3.3 berbasis SOA MZI Beralih

Sebuah semikonduktor amplifier optik dapat digunakan baik untuk amplifikasi dan redaman sinyal optik , dengan memutar keuntungan dan mematikan . Properti ini dapat digunakan untuk cara sederhana namun efektif untuk beralih dengan memisahkan optik sinyal dengan 3 dB splitter , setelah sinyal ini dilemahkan di satu tangan dan diperkuat di lengan lainnya . Sejak splitter kerugian dan kerugian tambahan (misalnya serat -chip kerugian kopling ) dapat dikompensasi oleh SOA , jenis switch dapat memiliki rendah kerugian atau bahkan keuntungan dan , di samping itu, rasio yang sangat baik on-off yang mengarah ke tingkat crosstalk rendah .

Berdasarkan Gambar 1.4 SOA MZI Beralih

Kerugian yang paling penting dari sebuah saklar SOA adalah tinggi tingkat tambahan kebisingan di ? ? \ on ? \ negara akibat spontan emisi yang dihasilkan dalam SOA .

1.4 OPTSIM

Optsim adalah sistem komunikasi optik canggih paket simulasi yang dirancang untuk teknik profesional dan penelitian mutakhir dari WDM , DWDM , TDM , CATV , optik LAN , bus paralel optik , dan lainnya muncul optik sistem dalam telekomunikasi , datacom , dan aplikasi lainnya . Hal ini dapat digunakan untuk merancang sistem komunikasi optik dan mensimulasikan mereka untuk menentukan kinerja mereka mempertimbangkan berbagai parameter komponen .

100

Ara 1,5 Optsim editor grafis

 

Optsim dirancang untuk menggabungkan akurasi terbesar dan pemodelan kekuatan dengan kemudahan penggunaan pada kedua Windows dan UNIX platform . Optsim merupakan komunikasi optik sistem sebagai set saling berhubungan blok , dengan setiap blok mewakili komponen atau subsistem dalam komunikasi sistem . Sebagai sinyal fisik lewat di antara kompone dalam sistem komunikasi dunia nyata , ? ? \ sinyal . Data dilewatkan antara model komponen dalam simulasi Optsim .

II PUSTAKA SURVEY

2.1 All- Optical Logic oleh MZI saklar

Koji Igarashi et al . dijelaskan pemrosesan sinyal optik berdasarkan modulasi fase optik dan selanjutnya optik filtering , yang berlaku untuk 160 - Gb / s optik time-division multiplexing ( OTDM ) subsistem . ultrafast modulasi fase sinyal optik dilakukan dengan self- phase modulasi ( SPM ) dan modulasi silang - fase ( XPM ) ketika pulsa optik melewati serat optik nonlinier . demikian modulasi fase menginduksi pergeseran spektral optik sinyal . Jian Wang ET . Al . disajikan logika ultrafast gerbang AND untuk kembali - ke - nol ( CSRZ ) sinyal carrier- ditekan oleh memanfaatkan dua jenis mengalir orde kedua nonlinier dalam neonatus lithium poled berkala ( PPLN ) Waveguide .

Solusi analitis berasal di bawah tidak ada deplesi pendekatan jelas menggambarkan prinsip operasi .

Pertama , berdasarkan mengalir generasi kedua - harmonik dan generasi - perbedaan frekuensi ( CSHG / DFG ) di PPLN , sebuah semua - optik 40 Gb / s CSRZ logika gerbang AND adalah berhasil diimplementasikan dalam percobaan dan diverifikasi oleh numerik simulasi . Hal ini ditemukan bahwa pemalas dikonversi , mengambil DAN hasilnya , terus format modulasi CSRZ unchanged.Second , dengan menggunakan mengalir sum – dan generasi - perbedaan frekuensi ( CSFG / DFG ) di PPLN . oleh memodifikasi desain yang sudah ada dua masukan nano fotonik Gerbang , yang operasi didasarkan pada optik dekat lapangan ( ONF ) interaksi antara tiga titik kuantum tetangga ( Qds ) , mereka meningkatkan rasio gerbang ON / OFF hingga sekitar 9 dB . Untuk melakukan hal ini , Arash Karimkhani et al . telah menghilangkan kemungkinan langsung ONF interaksi antara input dan titik output. Kemudian , oleh menambahkan QD lain , sebagai kontrol dot kedua baik yang ada dan dimodifikasi arsitektur dua - masukan , mereka mengusulkan dua baru tiga - masukan nanophotonic gerbang schemes.one dengan interaksi ONF langsung antara input dan output titik , dan yang lain tanpa interaksi tersebut . Flip-flop adalah terdiri dari dua lintas digabungkan VCSOA elektrik dipompa inverter dan menggunakan prinsip-prinsip modulasi cross-gain , gain polarisasi anisotropi , dan keuntungan yang sangat nonlinear karakteristik untuk mencapai fungsi flip-flop . mereka menyoroti bahwa , ketika terintegrasi pada chip , jenis ini semua - optik flip-flop membuka prospek baru untuk mengimplementasikan semua - optik kenangan cepat dan sirkuit waktu regenerasi .

Jingsheng Yang et al . disajikan strategi fungsi -lock untuk semua - optik gerbang logika ( AOLG ) memanfaatkan cross- polarisasi modulasi ( CPM ) efek dalam semikonduktor amplifier optik ( SOA ) . Dengan memonitor kekuatan logika cahaya , strategi menyadari metode terkendali untuk menangkap OR dan NOR fungsi dan beralih di antara mereka . Strategi ini telah berhasil diterapkan dalam percobaan dengan 10 - Gb / s tidak - kembali-ke - nol ( NRZ ) sinyal , yang memiliki tinggi Keberhasilan -rate di atas 95 % dan memastikan rasio kepunahan yang tinggi hasil cahaya di atas 11,4 dB . Setiap langkah dalam strategi memiliki evaluasi numerik yang pasti , yang memberikan potensi pelaksanaan otomatis .

2 . 2 OTDM BY MZI SWITCHING

 

D. Petrantonakis , P. Zakynthinos et . al mendemonstrasikan semua - optik empat panjang gelombang modus 3R meledak regenerator , bebas dari kesalahan operasi dengan 10 - Gb / s paket data panjang variabel bahwa pameran 6 - dB variasi daya packet - to- paket . itu sirkuit dilaksanakan dengan menggunakan urutan tiga terintegrasi berbasis amplifier semikonduktor quadruple optik Array interferomentric Mach.Zehnder . T. Ohara , H. Takara et . al memberikan laporan pertama dari 160 - Gb / s optik time-division multiplexing transmisi dengan semua – channel independen modulasi dan semua -channel simultan demultiplexing . Dengan menggunakan multiplexer dan demultiplexer berdasarkan poled berkala neonatus lithium dan semikonduktor amplifier optik hibrida cahaya planar terintegrasi gelombang sirkuit , 160 - km transmisi ini berhasil demonstrated.Colja Schubert et al . menyelidiki tiga interferometric switch semua - optik yang berbasis di kayu salib fase modulasi ( XPM ) dalam semikonduktor amplifier optik ( SOA ) , laser semikonduktor amplifier di cermin lingkaran ( Slalom ) switch, interferometer Mach.Zehnder ( MZI ) switch, dan interferometer ultrafast nonlinier ( UNI ) beralih . Switching jendela dengan lebar yang berbeda diukur dalam kondisi yang sama untuk ketiga beralih konfigurasi ? e . J. M. Verdurmen disorot waktu semua – optik domain add -drop multiplexing untuk fase termodulasi OTDM sinyal untuk pertama kalinya , untuk pengetahuan kita . XIN Ming , et . al menyatakan alternatif untuk label swapping , sebuah Skema label stripping semua - optik berdasarkan SOA - MZI . itu Proses pengupasan adalah mengendalikan diri tanpa proses sinkronisasi . Hasil simulasi menunjukkan bahwa tinggi kualitas pengupasan dapat dicapai , dengan tidak lebih dari 0.09dB fluktuasi daya dan 0.05dB fluktuasi fase di kedua ditelanjangi dan tetap label . Sebuah rasio kontras kekuatan 28dB antara label dilucuti tetap dan residu, dan 30dB ratio ( SNR ) signal-to -noise dapat dicapai masing-masing. Spalter et . al . sifat transmisi dinyatakan dan kecepatan tinggi beralih teknologi disajikan untuk 160 - Gb / s OTDM sistem , yang perlu membuktikan biaya - efektif dalam point- to-point menghubungkan transmisi dan harus menawarkan rute waktu – domain kemampuan untuk menjadi kenyataan komersial . Analisis toleransi parameter menunjukkan bahwa pengupasan kinerja memburuk sedikit ketika mempertimbangkan perangkat ? ? e ketidaksempurnaan dalam praktek . Hasil simulasi multi- hop juga menunjukkan bahwa skema kami berlaku untuk OPS skala besarjaringan . Hans- Georg Weber et al . disajikan ultrahigh – speed Kinerja Sistem Komunikasi Serat Optik Menggunakan MZI Switching 101

transmisi data pada serat optik berdasarkan waktu optikdivision multiplexing ( OTDM ) teknologi transmisi . Pemrosesan sinyal optik dalam pemancar dan penerima sebagai serta persyaratan pada transmisi data kecepatan ultrahigh melalui link fiber dibahas . Akhirnya , hasil dari beberapa Percobaan OTDM - transmisi , termasuk 160 - Gb / s pengiriman melalui 4320 km , 1.28 - Tb / s transmisi lebih dari 240 km , dan 2,56 - Tb / s transmisi melalui serat penghubung 160 - km , yang dijelaskan .

2.3 Tujuan

Dalam tesis ini , penelitian dilakukan dengan tetap melihat dari tujuan sebagai berikut .

1 . Untuk mengetahui tingkat kesalahan bit dan power control dari 4 X40 Gb / s waktu optik domain sistem multiplexing menggunakanMach - Zehnder beralih .

2 . Untuk menyelidiki operasi logis optik multiplexer dan encoder menggunakan Mach - Zehnder Inferometer .

3 . Untuk mengetahui tingkat kesalahan bit dari FTTH pada 40 Gbit / s denganMach - Zehnder Switching .

 

2.4 Penelitian Garis Setelah mempelajari pengenalan , survei literatur dasar, kita menentukan tujuan dalam bab II .

Dalam bab III , kita menyelidiki operasi logis optik multiplexer dan encoder oleh Mach - Zhender Inferometer pada 10 Gbit / s .

Dalam bab IV , kami praktis menyelidiki dan memvalidasi bit Tingkat dan kontrol kekuatan menormalkan kekuatan Mach - Zehnder beralih pada empat saluran yang berbeda pada waktu yang berbeda bergeser pada bit rate yang sama dari 40 Gb / s .Kami akhirnya mendiskusikan kesimpulan dalam bab V dan juga pekerjaan di masa depan .

III implentation OF ENCODER OPTIK DAN

MULTIPLEXER MENGGUNAKAN MACH - ZEHNDER

INFEROMETER

Dalam bab ini perangkat logika sederhana semua - optik , yang disebutMach Zhender Inferometer terdiri dengan menggunakan Semiconductor Optical Amplifier ( SOA ) dan optik coupler . Perangkat ini digunakan untuk menghasilkan logis fungsi ( AND, XOR ) dan multiplexer dan Encoder adalah diperoleh dengan menggunakan perangkat ini di Optik Pohon Arsitektur . itu simulasi Encoder dan Multiplexer dilakukan pada tingkat 10 Gbit / s dan keduanya simulasi untuk input yang berbeda logis kombinasi . Simulasi menunjukkan bahwa perangkat ini cocok untuk beroperasi pada bit rate yang lebih tinggi dan juga untuk berbeda logis entitas .

3.1 PENDAHULUAN

Seperti kita ketahui dalam beberapa hari terakhir penelitian dalam komputasi optik meningkat dari hari ke hari dan banyak ilmuwan yang bekerja pada mereka, tetapi dalam elektronik komputasi operasi logis memainkan peran yang sangat penting karena mereka membutuhkan daya yang lebih kecil , seperti mereka adalah sirkuit digital dan dibandingkan dengan sirkuit analog , mereka sangat fleksibel . Tapi mereka memiliki kelemahan tertentu juga bahwa mereka bekerja sampai frekuensi yang terbatas , tetapi jika kita menggunakan itu logika menggunakan instrumen optik kemudian memberikan stabilitas yang lebih baik , kecepatan yang lebih baik dan switching. Dalam komputasi optik digital , sistem interkoneksi optik adalah primitif yang merupakan berbagai algoritma optik dan arsitektur . tinggi mempercepat gerbang logika semua - optik merupakan elemen kunci dalam berikutnya jaringan optik generasi dan sistem komputasi untuk melakukan fungsi pemrosesan sinyal optik , seperti semua - optik label swapping , pengakuan header, paritas memeriksa , penambahan biner dan enkripsi data. Dalam beberapa terakhir tahun , beberapa pendekatan telah diusulkan untuk mewujudkan berbagai gerbang logika baik menggunakan serat nonlinier tinggi atau semikonduktor amplifier optik ( SOA ) . The berbasis SOA perangkat memiliki potensi integrasi monolit , yang menawarkan keuntungan dari kekompakan , meningkat kehandalan dan pengurangan biaya . Hingga saat ini , berdasarkan sebagian SOA gerbang logika telah dilakukan dengan menggunakan lintas gain modulasi ( XGM ) dan modulasi silang - fase ( XPM ) , yang pasti membatasi kecepatan operasi perangkat tersebut karena dengan waktu pemulihan yang lambat pembawa intrinsik SOA.Although kecepatan operasi dapat ditingkatkan menjadi 40GB / s atau lebih tinggi dengan penggunaan daya tinggi terus menerus gelombang memegang balok [ 48 ] atau struktur interferometer yang berbeda , kompleksitas dan biaya perangkat meningkat . Permintaan kecepatan tinggi pemrosesan sinyal semua - optik telah ditimbulkan oleh arus dan jaringan optik dekat - masa depan dalam upaya untuk melepaskan node jaringan dari latency yang tidak diinginkan dan kecepatan keterbatasan yang ditetapkan oleh O / tahap konversi E / O dan untuk mencocokkan

 

pengolahan dan kecepatan transmisi . Dalam hal ini, peningkatan yang signifikan dalam upaya penelitian terhadap penyebaran kecepatan tinggi semua pemrosesan sinyal optik teknologi , konsep aplikasi dan demonstrasi memiliki telah menyaksikan selama beberapa tahun terakhir . semikonduktor penguat optik ( SOA ) berbasis , gerbang optik interferometric telah muncul sebagai pemrosesan sinyal fotonik main –stream unit , mengeksploitasi respon cepat mereka untuk operasi kecepatan tinggi dan mengambil keuntungan dari kemajuan yang luar biasa hibrida dan teknik integrasi monolitik untuk menawarkan kompak elemen switching . Untuk tujuan ini, elemen tunggal , kecepatan tinggi gerbang -optik telah dibuktikan sebagai perangkat terintegrasi di sejumlah laboratorium di seluruh dunia dan telah dikembangkan sebagai produk komersial terutama untuk panjang gelombang konversi dan regenerasi tujuan .

3.2 MULTIPLEXER

Multiplexer atau mux adalah perangkat yang melakukan multiplexing ; akan memilih salah satu dari banyak sinyal input analog atau digital dan output yang menjadi satu baris . Multiplexer input 2n memiliki n pilih bit , yang digunakan untuk memilih baris masukan untuk mengirim ke output . Input A Input B output

Tabel 3.1 Tabel Kebenaran 2:1 Multiplexer

3.3 ENCODER

Sebuah encoder adalah perangkat , sirkuit , transduser , program perangkat lunak dan algoritma yang mengubah informasi dari satu format , atau kode yang lain , untuk tujuan standardisasi , kecepatan , kerahasiaan , keamanan , atau menghemat ruang dengan mengecilkan ukuran . sebuah encoder dapat menjadi sebuah perangkat yang digunakan untuk mengubah sinyal ( seperti sedikit stream) atau data ke dalam kode . Kode ini menyajikan salah satu dari nomor tujuan seperti mengompresi informasi untuk transmisi atau penyimpanan, mengenkripsi atau menambahkan redudansi untuk input kode , atau menerjemahkan dari satu kode yang lain . Hal ini biasanya dilakukan dengan cara algoritma diprogram , terutama jika ada bagian digital , sementara sebagian besar encoding analog dilakukan dengan analog sirkuit .

3.4 KERJA MULTIPLEXER

Seperti yang sudah kita bahas saklar MZI untuk semua – optik logika jadi di sini kerja pohon optik menggunakan MZI berbasis switch optik . Ada sumber konstan CW berkas yang mungkin menjadi sumber laser. Sinyal cahaya yang berasal dari CWLS dapat diambil sebagai sinyal yang masuk . incoming sinyal cahaya adalah insiden pada saklar s1 pertama . Sekarang kita bisa memperoleh cahaya di cabang yang diinginkan berbeda atau sub cabang oleh menempatkan tepat dari sinyal kontrol . Sinyal kontrol juga sinyal cahaya .

Ara 3,1 logika Optical menggunakan MZI saklar

Kasus 1 : ? ? ? ? Ketika A = e0 f dan B = e0 f ?

The CW sinar yang berasal dari CWLS konstan insiden pada saklar s1 pertama . Seperti di sini A = 0,0 ? ? E , sinyal kontrol A tidak ada, itu berarti sinyal cahaya yang masuk hanya hadir pada

s1 . Sesuai prinsip beralih dibahas di atas , cahaya

muncul melalui saluran yang lebih rendah dan jatuh pada saklar s3 di C.

Berikut kontrol sinyal B tidak ada. Sebagai sinyal B yang absen sehingga

 

light akhirnya keluar melalui saluran yang lebih rendah dan s3

mencapai output 1 . Dalam hal ini , tidak ada cahaya hadir di lain

output port , port1 sehingga output adalah salah satu negara dan yang lain berada dalam

state nol .

Kasus 2 : ? ? ? ? Ketika A = e0 f dan B = e1 f ?

Cahaya dari sumber cahaya CW adalah insiden pada s1 . Sebagai A = .0 ? E ? ,

sinar muncul melalui saluran yang lebih rendah dan jatuh pada

s3 . Pada s3 kontrol sinyal B hadir . Dalam kehadiran

sinyal kontrol muncul melalui saluran atas dan s3

akhirnya mencapai ke port output 2 . Dalam hal ini cahaya hanya

hadir dalam output port 2 . pelabuhan Oleh karena itu keluaran menunjukkan satu negara

sementara yang lain menunjukkan nol negara.

Kasus 3 : ? ? ? ? Ketika A = e1 f dan B = e0 f ?

Cahaya dari CWLS adalah insiden pada saklar s1 pertama . Seperti di sini A

= .1 ? ? E , kontrol sinyal A hadir . Karena itu , cahaya

muncul melalui saluran atas s1and jatuh pada s2 di

O.As B = 0,0 ? ? E , tidak ada sinyal kontrol hadir di B , yang berarti

cahaya yang keluar dari saluran yang lebih rendah s2 untuk mencapai keluaran

port 3 . Jadi keluaran port 3 adalah dalam satu negara dan lain-lain berada di nol

negara.

Kasus 4 : ? ? ? ? Ketika A = e1 f dan B = e1 f ?

Cahaya dari CWLS adalah insiden pada saklar s1 pertama . Seperti di sini A

= .1 ? ? E , kontrol sinyal input A hadir . Karena itu,

cahaya muncul melalui saluran atas s1 dan jatuh pada s2

di O. Sebagai B = 0,1 ? ? e , sinyal kontrol hadir di B. Oleh karena itu

cahaya mengikuti saluran atas s2 untuk mencapai output 4 . Jadi

output port 4 adalah dalam satu negara dan yang lain adalah nol negara.

Ara 3.2 Blok Diagram Multiplexer

3.5 HASIL DAN PEMBAHASAN SIMULASI

Bagian ini tesis bercerita tentang hasil multiplexer

dan encoder menggunakan Mach - Zehnder Inferometer untuk semua - optik

 

logika . Proyek ini disimulasikan dalam OPTSim 4.7.1 ditentukan dalam

Modus blok yang membawa komponen yang berbeda untuk menghasilkan

sirkuit yang diperlukan yang memberikan hasil akhirnya .

3.6 SISTEM URAIAN MULTIPLEXER

Ini diberikan di bawah angka merupakan diagram skematik

semua - optik logika multiplexer oleh MZI switch. Karena mengandung

dua generator gelombang sinus yang memiliki frekuensi 10 GHz yang

bertindak sebagai pembangkit sinyal diikuti oleh Direct Modulated Laser ,

sebagai laser yang mengubah sinyal listrik menjadi sinyal cahaya dan

Output dari kedua laser diumpankan ke coupler optik yang berisi

dua port yang disebut sebagai bar pelabuhan dan lintas pelabuhan , sekarang dari setiap lengan

dari coupler diumpankan ke MZI s1 MZIs2MZIs3 .

Semikonduktor penguat optik dan akhirnya pergi ke

coupler optik sebagai coupler optik diikuti oleh semikonduktor

penguat optik disebut Mach Zehnder Beralih dan berbeda

output dari coupler optik diumpankan ke Spectrum Analyzer .

Signal Generator menghasilkan 10 sinyal GHz dalam bentuk sinusoidal

yang diumpankan ke laser DM . Direct Mode Laser blok menunjukkan

disederhanakan gelombang kontinu ( CW ) laser. Fase kebisingan adalah

diperhitungkan dengan menghasilkan sinyal generator yang

FWHM (Full Width Half Maximum ) ditentukan oleh Laser

parameter . Dalam model dianggap pusat has193.42 THz

frekuensi emisi , 1550 nm panjang gelombang , 1.650 nm

panjang gelombang , 0dBm CW Power, 1mW CW Power, Laser yang ideal

bandwith noise , lebar garis 10 FWHM dan laser acak

fase . Skrup optik , juga disebut sebagai skrup opt , adalah

perangkat terkenal digunakan cahaya langsung dari satu sumber cahaya

untuk anggota yang menerima cahaya . Sebuah coupler optik pasif

perangkat untuk percabangan atau kopling sinyal optik . umumnya ,

coupler terpusat dengan bergabung dua serat bersama-sama sehingga

bahwa cahaya dapat melewati dari unit pengirim ke dua

 

penerima , atau yang lain itu dapat dilakukan dengan menyandingkan dua

" receiver " serat yang kemudian akan selaras dan diposisikan sehingga

sebagai untuk menghadapi " pengirim " serat .

Semiconductor amplifier optik adalah amplifier yang menggunakan

semikonduktor untuk menyediakan media gain . itu

semikonduktor amplifier optik adalah ukuran kecil dan

elektrik dipompa .

SOA memiliki noise yang lebih tinggi , keuntungan yang lebih rendah , polarisasi moderat

ketergantungan dan nonlinier tinggi dengan cepat sementara waktu .

Ini berasal dari nanodetik pendek atau kurang negara atas

seumur hidup , sehingga gain bereaksi cepat terhadap perubahan pompa atau

kekuatan sinyal dan perubahan keuntungan juga menimbulkan fase

perubahan yang dapat mengganggu sinyal .

Kinerja Sistem Komunikasi Serat Optik Menggunakan MZI Switching

103

Ara 3.3 Skema Diagram Multiplexer ( A = 0,1 ? ? E , B = 0,0 ? ? E )

Ara 3.4 Panjang gelombang spektrum A = 0,1 ? ? E & B = 0,0 ? ? E

Diagram di atas menunjukkan spektrum panjang gelombang

diperlukan logika pada output port 1 . Sebagai spektrum bahwa baik

sinyal input dan sinyal kontrol memiliki berbeda

panjang gelombang jadi kita gunakan untuk sinyal kontrol adalah 1550 um

sedangkan sinyal yang masuk terdiri dari panjang gelombang 1650 um

sehingga memiliki amplitudo maksimum pada panjang gelombang kontrol

sinyal .

Kasus 1 : ? ? ? ? ? ? Ketika A = f1 f , B = e0 f , & EN = e1 f ?

Dalam diagram skematik ini encoder tiga gelombang sinus

generator digunakan untuk menghasilkan pulsa sinusoidal yang secara langsung

diumpankan ke laser termodulasi langsung yang bekerja di berbagai

panjang gelombang sinyal masukan tertentu , sebagai encoder ini memiliki

tiga sinyal input dan mengaktifkan sinyal pada panjang gelombang yang berbeda

dari sinyal input dan mengaktifkan sinyal ini diumpankan langsung ke

 

lengan input dari coupler dari MZI beralih dengan beam splitter

dan memberikan logika yang diperlukan .

Skema diagram dari encoder ( A = 0,1 ? ? E , B = .1 ? ? E )

Panjang gelombang spektrum A = 0,1 ? ? E & B = .1 ? ? E

Kasus 2 : ? ? ? ? ? ? Ketika A = e1 f , B = e1 f & EN = e1 f ?

Dalam skema ini sebagai diagram terakhir mewakili dua sinusoidal

Generator 10 GHz diikuti oleh laser Direct Mode yang

mengubah sinyal listrik menjadi sinyal optik atau sinyal cahaya

dan output dari laser langsung dimasukkan ke lengan input

coupler yang melewati sinyal pada bar pelabuhan sebagai tergantung

pada sinyal kontrol . Di sini sinyal kontrol dimasukkan ke dalam

sirkuit di tingkat ketiga saklar MZI karena terdiri dari

dua semikonduktor amplifier optik di pelabuhan kedua

coupler optik pada input dan hal yang sama diikuti di

output dari saklar.

Jadi di sini dalam sinyal kontrol sirkuit ini diterapkan untuk semua yang

masukan dari encoder tetapi menurut prinsip MZI

menonaktifkan input diterima di pelabuhan bar coupler ketika

sinyal kontrol hadir sehingga kami menerapkan dua kontinu

sinyal pada masukan dari kedua laser sehingga pada output tahap pertama adalah

diterima di pelabuhan bar dari coupler optik 2 sesuai dengan

sehingga output dari coupler optik diumpankan ke input dari

coupler optik 3 dan pada saat yang sama masukan ketiga juga FBI untuk

input dari coupler optik 3 sekarang lagi sama

fenomena hadir sebagai sinyal kontrol menjadi output dari

coupler optik 2 dan berkesinambungan sinyal gelombang diperlakukan sebagai masukan

output sehingga sama sama dari coupler optik diumpankan ke

input coupler optik 5 dan juga sinyal kontinyu yang sama

diumpankan ke coupler optik 4 dan sekarang output dari optik

coupler 3 & 5 diproses dengan benar .

Skema diagram dari encoder ( A = 0,1 ? ? E , B = 0,0 ? ? E , EN = ? ? E1 ? E )

 

Panjang gelombang spektrum ( A = 0,1 ? ? E & B = 0,0 ? ? E , EN = ? ? E1 ? E )

Sekarang output dari coupler optik 4 & 6 diumpankan ke salah satu

masukan lengan 7,9 serat optik , 11 , 13 dan kemudian output

dari ini coupler optik yang diperlukan adalah pergi ke optik

coupler 8 , 10 , 12 , 14 sampai lewat dengan Semiconductor yang

penguat optik ( SOA ) . Pada kedua port dari optik

spektrum analyzer coupler dihubungkan untuk mengukur

spektrum panjang gelombang yang lewat melalui tepat

channel seperti yang kita lihat sebelumnya jika kita menerapkan masukan pada kedua ujung

coupler salah satu dari terus menerus dan lainnya adalah sinyal kontrol

memiliki panjang gelombang yang berbeda dari gelombang kontinu

sinyal maka output diterima pada satu port coupler

sehingga dengan cara itu MZI beralih bekerja sebagai inverter logis sehingga

EN sini sama seperti bekerja sebagai inverter begitu berbeda spektrum

diterima di output dari coupler optik tapi benar

cara output diterima di spektrum analyzer 10 dan

ditampilkan di ara 2.10 yang menunjukkan spektrum panjang gelombang

sinyal yang diterima dari semua logika optik encoder dalam bentuk

.1 ? E .

International Journal Soft Computing dan Teknik ( IJSCE )

ISSN : 2231-2307 , Volume - 2 , Edisi - 3 , Juli 2012

104

Ara 3.5 Skema diagram dari encoder ( A = 0,1 ? ? E , B = .1 ? ? E , EN =

.1 ? ? E )

Gambar 3.6 Panjang gelombang spektrum A = 0,1 ? ? E & B = .1 ? ? E , EN = 0,1 ? ? E

3.7 KESIMPULAN

Kami telah disimulasikan logika berbasis Multiplexer semua - optik dan

Encoder menggunakan MZ Inferometer . Fungsi logika yang berbeda

dapat direalisasikan hanya dengan menyesuaikan dua komponen yaitu

multiplexer dan encoder. Metode simulasi memiliki

potensi untuk beroperasi diatas 40 GB / s .

 

IV OPTICALTIME DIVISION MULTIPLEXING MENGGUNAKAN

MZI SWITCHING

Banyak aliran data yang lebih rendah kecepatan dapat di-multiplexing ke

satu kecepatan tinggi aliran dengan cara pembagian waktu Optical

multiplexing ( OTDM ) , sehingga setiap saluran masukan mentransmisikan

data dalam slot waktu yang ditetapkan . Tugas ini dilakukan

oleh sebuah saklar multiplexer cepat ( mux ) . The routing yang berbeda

data stream pada akhir link TDM dilakukan oleh

beralih demultiplexer ( demux ) dan demultiplexer ini

bekerja menggunakan MZI beralih karena terdiri semikonduktor

penguat optik ( SOA ) dan coupler optik . Dalam bab ini

empat saluran OTDM disimulasikan pada 40 Gbit / s dan selanjutnya adalah

meneliti dampak dari kekuatan sinyal , lebar pulsa dan

mengontrol kekuatan sinyal pada BER .

4.1 PENDAHULUAN

Kapasitas transmisi jaringan optik bisa

diperpanjang dengan cara yang sederhana dengan memasang serat tambahan (spasi

division multiplexing atau SDM ) . Karena ini sangat mahal ,

metode telah dikembangkan untuk penggunaan yang lebih efisien dari

bandwidth yang tersedia dalam jaringan serat yang ada . A pertama

solusi adalah untuk meningkatkan bit rate dalam jaringan , yang

membutuhkan elektronik kecepatan tinggi di node jaringan .

Interleaving dapat dilakukan pada bit - by- bit dasar , seperti ,

atau pada packet - by- packet basis . Sebagai kecepatan data menjadi lebih tinggi

dan lebih tinggi , menjadi lebih sulit untuk bagian-bagian elektronik

( switch ) dalam sistem untuk menangani data dengan benar . A. Cheng

et al . Disajikan 40 Gb / s demultiplexing OTDM menggunakan

semua - optik merdu delay line dan elektro - penyerapan

modulator . Penundaan serat optik terus menerus untuk saluran

Temukan direalisasikan menggunakan empat - gelombang - pencampuran dan panjang gelombang

tergantung delay kelompok . Ken Morito et al . disajikan seragam

 

kekuatan output dan rasio kepunahan tinggi untuk Mach - Zehnder

Jenis interferometer semua switch optik dengan asimetris

amplifier bias dan shifter fase ditemukan di dinamis

analisis untuk kontrol pulsa sempit dan beralih dioptimalkan

windows. Kinerja pembagian waktu optik

multiplexing ( OTDM ) sistem dibatasi oleh kompleks

kombinasi suara . Dalam makalah ini kami menyajikan teoritis

kerangka kerja untuk penerima optik dalam sistem OTDM berdasarkan

fungsi generasi saat . Jianfeng Zhang et . al .

disajikan model receiver yang diusulkan menunjukkan lebih

akurat dalam memprediksi tingkat kesalahan bit ( BER ) kinerja

dari yang bekas [ 47 ] Masalah ini dapat diatasi dengan .

routing data melalui domain optik, yang dilambangkan

waktu sebagai optik division multiplexing ( OTDM ) . Kecepatan

hari ini sistem OTDM eksperimental dalam urutan

10 Gb / s ( saluran tunggal ) , dan sebagian besar dibatasi oleh kecepatan

unsur-unsur non - linear dan pengaruh efek fisik

seperti dispersi kromatik pada pulsa optik dalam

serat dipekerjakan . Mach - Zehnder interferometer dengan

terintegrasi SOA ( SOA - MZI ) sangat menarik sebagai

kecepatan tinggi gerbang optik . Mereka memiliki energi beralih rendah ,

kekompakan tinggi dan stabilitas , serta potensi

integrasi optik lebih lanjut. Dalam bab ini kita simulasi empat

saluran saluran OTDM pada kecepatan 40 Gbit / s untuk BER dengan

lebar pulsa , Pengendalian Sinyal Daya .

4.2 WAKTU DIVISION MULTIPLEXING

TDM adalah proses digital yang memungkinkan beberapa sambungan ke

berbagi bandwidth tinggi link . Alih-alih berbagi a

sebagian dari bandwidth seperti pada FDM , waktu bersama . masing-masing

koneksi menempati sebagian waktu di link . TDM adalah

Teknik multiplexing digital untuk menggabungkan beberapa tingkat rendah

 

saluran menjadi satu tingkat tinggi satu . Pada TDM , data rate dari

link n kali lebih cepat , dan durasi unit n kali lebih pendek .

4.3 OPTIK WAKTU DIVISION MULTIPLEXING

Multiplexing elektronik pada kecepatan tersebut tetap sulit

dan menyajikan pembatasan pada pemanfaatan bandwidth

link serat single-mode . Strategi alternatif untuk meningkatkan

bit rate sistem serat optik digital di luar

kemampuan bandwidth dari drive elektronik yang dikenal sebagai

waktu optik division multiplexing ( OTDM ) . Di mulai dari

pengiriman data serat optik saluran digital listrik

sinyal elektrik telah sampai multiplexing untuk maksimal yang

Data agregat tingkat menyusul hirarki data yang telah ditetapkan .

Sinyal listrik agregat ini dikonversi

elektro - optik ke dalam domain optik hanya untuk

transmisi . Untuk demultiplexing , yang ditransmisikan optik

sinyal diubah ke dalam domain listrik dan

demultiplexed dalam domain listrik . Prinsip ini

Teknik adalah untuk memperpanjang waktu division multiplexing dengan optik

menggabungkan sejumlah kecepatan yang lebih rendah baseband elektronik

saluran digital . Gambar menunjukkan multiplexing optik dan

Rasio demultiplexing adalah 1:4, dengan tingkat saluran baseband

bit rate yang dibutuhkan .

Sistem dapat disebut sebagai sistem OTDM empat channel .

Empat pemancar pada gambar didorong oleh umum 40

Jam GHz menggunakan kuartal bit jangka waktu penundaan . modus Dikunci

Kinerja Sistem Komunikasi Serat Optik Menggunakan MZI Switching

105

sumber laser semikonduktor yang diproduksi pendek optik

pulsa yang digunakan pada pemancar untuk memberikan tugas rendah

pulsa siklus stream multiplexing untuk waktu berikutnya . data

dikodekan ke sungai pulsa ini menggunakan optik terintegrasi

 

MZ modulator yang memberi RZ output pemancar di 10

Gbit / s . perangkat IO ini digunakan untuk menghilangkan laser

kicauan akan menghasilkan dispersi dari pulsa ditransmisikan sebagai

mereka disebarkan dalam mode serat tunggal , sehingga membatasi

jarak transmisi dicapai .

Empat 40 Gbit / s sinyal data digabungkan dalam pasif

power combiner optik tapi , pada prinsipnya , sebuah switching yang aktif

elemen dapat dimanfaatkan . Meskipun empat sumber optik

bekerja , mereka semua dipancarkan pada panjang gelombang optik yang sama

dalam toleransi ? ? } 0,2 nm dan karenanya 40 Gbit / s Data

stream sedikit disisipkan untuk menghasilkan 160 Gbit / s baseband

komponen dalam demultiplexer yang terdiri dari dua tingkat .

Sekali lagi perangkat Waveguide IO digunakan untuk memberikan switching

berfungsi pada setiap tingkat . Pada tingkat pertama saklar IO didorong

oleh sinusoid pada 80 GHz untuk demultiplex yang masuk 160

Gbit / s stream ke 80 Gbit / s sinyal . Oleh karena itu tunggal panjang gelombang

160 Gbit / s transmisi optik diperoleh dengan elektronik

yang hanya dibutuhkan bandwidth maksimum sekitar 25 GHz ,

sebagai kembali ke nol pulsa bekerja .

4.4 DEMUX OPERASI MENGGUNAKAN MZI - SOA SWITCH

The MZI - SOA beralih semua optik ditunjukkan pada Gambar 5.4

terdiri dari dua simetris 2x2 multimode interferometer

( MMI ) splitter untuk membagi dan menggabungkan pulsa data, dua

skrup untuk memperkenalkan pulsa kontrol, dua SOA ? es untuk

memberikan pergeseran fasa , dan shifter fase ( PS ) untuk menyesuaikan

mengimbangi fase . Sinyal data diinjeksikan dari port input

diarahkan ke lintas pelabuhan atau port bar tergantung pada

perbedaan fasa antara dua es SOA ? . Dengan menyuntikkan

control pulsa 2 dengan penundaan waktu tertentu dan energi yang tepat

perbedaan terhadap pulsa kontrol 1 , perubahan fase lambat

terkait dengan pemulihan gain lambat dalam dua SOA ? es yang

 

benar-benar dihilangkan . Hal ini menimbulkan beralih pendek

windows. Dengan menyesuaikan kali suntikan dua control

pulsa , salah satu sinyal pulsa data yang multiplexing dapat

turun ke port bar dan sinyal lain dapat

ditransmisikan ke lintas negara . Berikut kontra menyebarkan data yang

dan pulsa kontrol diasumsikan .

Ara 4.1 DEMUX menggunakan MZI saklar

4.5 SETUP SIMULASI UNTUK OTDM

Setup sistem tertentu dari OTDM ditunjukkan pada gambar

( 5.5 ) . Komponen yang digunakan dalam gambar ( 5.5 ) dipilih dari

Optsim Ver.4.7.0 palet komponen perpustakaan dan ditempatkan sesuai

persyaratan di bidang desain editor Optsim . kemudian

berbagai parameter simulasi ditetapkan . pemancar

terdiri dari urutan biner pseudo- acak atau PRBS

Generator , modus terkunci dioda laser , sebuah generator listrik ,

empat kali pergeseran blok , sebuah MUX optik dan optik

menormalkan . Beberapa saluran dari MLLD adalah RZ

dimodulasi dengan pola PRBS yang berbeda . The PRBS block

menghasilkan beberapa output pola , masing-masing berbeda dari

lainnya dan pada bit rate yang sama . Semua saluran dari MLLD berada di

panjang gelombang yang sama 1650nm dan kekuatan yang sama . sebelum

menjadi multiplexing bersama-sama masing-masing saluran konsekuen adalah

tertunda oleh 1/4 dari waktu window berturut-turut . Jumlah kekuatan

semua saluran dikendalikan oleh menormalkan optik, yang

menentukan output daya rata-rata sinyal OTDM sebelum

propagasi lebih panjang serat . Para perjalanan sinyal OTDM

melalui serat optik dari 100 km panjang dan maka itu adalah

de - multiplexing pada akhir penerima . Penerima terdiri dari

empat identik beralih SMZ ( tapi dengan penundaan waktu yang berbeda ) ,

masing-masing terdiri dari generator denyut nadi kereta api ( dengan pengulangan yang sama

Tingkat sebagai pemancar ) , normalizer optik blok , splitter pulsa

 

dan dua blok waktu tunda dan SMZ beralih dengan dua keluaran

port . The BER meter terhubung pada kedua output dan

tercermin pelabuhan untuk mendapatkan hasil . Semua saklar SMZ adalah

terhubung pada output dari fiber.Schematic nonlinear

Diagram OTDM menggunakan MZI beralih

4.6 HASIL DAN PEMBAHASAN

Sinkronisasi antara pemancar dan penerima di OTDM

merupakan isu penting untuk kinerja optimal dari sistem . dalam hal ini

kertas, pemancar dan penerima telah disinkronkan

dengan penambahan delay optik dalam sinyal kontrol . itu

delay optik bervariasi sebagai multiple integer dari 1/4 dari

lebar pulsa dalam sebuah diharapkan terikat. Pola yang

muncul dari variasi tersebut menentukan optimum optik

delay yang dibutuhkan untuk setiap saluran . Pengaruh kebisingan dan

distorsi sangat terkenal di transmisi digital . kebisingan

menyebabkan kesalahan bit di gerbang keputusan penerima dan

distorsi menyebabkan perubahan pada bentuk pulsa sehingga antar

gangguan simbol ( ISI ) , yang juga menghasilkan kesalahan bit .

Parameter utama selain bandwidth, yang

ciri link optik digital , adalah BER . Jadi efek

kekuatan sinyal ( Psignal ) , sinyal kontrol listrik ( Pcontrol ) , dan

lebar pulsa pada BER diselidiki . Gambar menunjukkan variasi

BER dengan perubahan dalam kekuatan sinyal . Seperti disebutkan sebelumnya

normalizer optik mengontrol daya keluaran rata-rata

Sinyal multiplexing .

BER untuk kanal 1 adalah di kisaran 10-21.10-28 untuk

Psignal nilai 5 dan 10 dBm , masing-masing.

Ara 4.2 BER versus daya sinyal input dengan dispersi

Jadi teramati bahwa dengan peningkatan kekuatan sinyal

( Psignal ) BER mengalami perbaikan .

Demikian pula untuk saluran 2 dan 4 variasi ini berada di kisaran

 

10-22.10-25 dan 10-22.10-25 untuk nilai Psignal dari 5 dan 10

International Journal Soft Computing dan Teknik ( IJSCE )

ISSN : 2231-2307 , Volume - 2 , Edisi - 3 , Juli 2012

106

dBm , masing-masing. Sangat menarik untuk dicatat bahwa BER untuk

saluran 2 dan 3 adalah sama untuk semua nilai Psignal . BER dari

penerima optik berbanding terbalik dengan SNR , yang di

menghidupkan tergantung pada daya optik sinyal . Jadi BER

menurun dengan peningkatan kekuatan sinyal .

Ara 4.2 BER vs lebar pulsa dengan dispersi

Selanjutnya pada Gambar . 5.8 efek perubahan lebar pulsa pada BER

diselidiki . Lebar pulsa dari sinyal input bervariasi

dalam batas-batas 5e - 12.12e - 12 m dan variasi dalam BER

diamati . Seperti yang terlihat pada gambar untuk saluran 1 BER pada

5e - 12m adalah 10-140 dan dengan peningkatan denyut nadi lebar itu

menurun menjadi 10-27 untuk lebar pulsa 12e - 12 m . sekali lagi

ada tumpang tindih dalam kurva untuk saluran 2dan 3 dan

variasi untuk BER adalah 10-140 ke 10-167 untuk

Variasi tersebut di atas lebar pulsa . Untuk channel 4 yang

nilai BER bervariasi 10-140 hingga 10-162 untuk

yang disebutkan di atas variasi lebar pulsa . hasil

menunjukkan peningkatan kinerja penerima dengan

peningkatan lebar pulsa . Peningkatan ini dapat dikaitkan dengan

pengurangan lebar pulsa distorsi Menampilkan signifikan

degradasi kinerja penerima ketika sinyal kontrol

listrik meningkat secara bertahap melampaui 22 dBm .. Jadi dalam kasus

channel 1BER pada sinyal kontrol 22dBm adalah 10-35 dan meningkat

untuk 10-4 pada 26 dBm . Saluran 2 dan 3 sekali lagi menunjukkan

pola BER identik dan variasi dalam

Ara 4.3 BER vs sinyal kontrol listrik dengan dispersi

Rentang 10-35.10-4 pada 22 dan 26 dbm , masing-masing. itu

 

variasi BER untuk kanal 4 adalah di kisaran 10-35.10-2

untuk variasi yang disebutkan di atas dalam kekuatan sinyal kontrol. itu

dipahami bahwa prinsip operasi dari sebuah switch adalah MZI

berdasarkan interferensi antara sinyal melewati dua

kaki dari MZI . Sinyal kontrol mempengaruhi perubahan

indeks bias bahan semi - konduktor . Perubahan

indeks bias pada gilirannya memperkenalkan pergeseran fasa pada input

sinyal . Ini menunjukkan efek Pcontrol pada BER tanpa

dispersi untuk semua channels.the dua sinyal gangguan pada

output dan output yang dihasilkan tergantung pada mereka

pergeseran fase relatif . Dengan demikian , sinyal dapat mengganggu baik

konstruktif atau destruktif . Dari grafik jelaslah

bahwa peningkatan sinyal kontrol melampaui 22 dB m memperkenalkan

pergeseran fasa , yang menurunkan kinerja penerima dan

BER terus meningkat dengan peningkatan sinyal kontrol

kekuasaan. Gambar . 5.12 menggambarkan diagram mata untuk channel1 , di

Nilai Pcontrol dari 22 dan 26 , masing-masing. ada

degradasi di tingkat keputusan

nilai offset dari 1,5 X 10-5 sampai 7 X 10-6 dengan peningkatan

Nilai Pcontrol 22-26 . Pengamatan ini mendukung

kesimpulan yang ditarik dari ayat-ayat BER sinyal Pcontrol .

Ara 4.4 Diagram Eye

4.6 KESIMPULAN

A 160 - Gb / s transmisi OTDM dengan semua - channel

modulasi dan semua -channel demultiplexing simultan memiliki

berhasil disimulasikan untuk pertama kalinya . The MUX dan

DEMUX penggunaan MZI beralih ketat menjaga waktu tunda

antara saluran yang berdekatan dan suhu tinggi - tawaran

stabilitas karena mereka hibrida terintegrasi pada MZI saklar ;

mereka akan , oleh karena itu, menjadi kunci untuk transmisi OTDM masa depan

sistem .

 

V. KESIMPULAN DAN ASPEK MASA DEPAN

5.1 KESIMPULAN

Dalam tesis ini skema menghasilkan logika optik

dilaksanakan oleh MZ Inferometer seperti yang dibahas dalam bab 2 kaleng

digunakan untuk tujuan yang berbeda . Skema ini dapat dengan mudah dan

berhasil diperpanjang dan dilaksanakan untuk lebih tinggi

jumlah masukan angka oleh penggabungan yang tepat dari MZI berdasarkan

switch optik , ekstensi vertikal dan horizontal pohon

dan oleh seleksi cabang sesuai. Sekali lagi seluruh operasi

paralel di alam , yaitu hasil usaha yang berbeda

antara data yang diperoleh pada suatu waktu . Di sini kita bisa

melaksanakan beberapa instruksi beberapa tipe data

operasi dengan baik . Operasi aritmatika dapat dilakukan

sini antara dua data yang besar berbentuk . Salah satu yang diusulkan

bit skema perbandingan digital juga berhasil memanfaatkan

bahan non - linear berbasis struktur pohon untuk operasi . sekarang

Penting untuk dicatat bahwa diskusi di atas didasarkan pada

model sederhana . Dalam simulasi ini beberapa parameter berjalan harus

dipertimbangkan sehingga sifat dispersi , polarisasi

serat , nilai-nilai yang telah ditentukan dari intensitas / panjang gelombang

sinar laser untuk kontrol dan sinyal masuk , pengenalan

filter , kerugian intensitas karena splitter balok / fiberskrup dll Seperti dalam tesis ini panjang gelombang gelombang terus menerus sinar laser adalah 1550 ƒÊm dan sinyal berdenyut panjang gelombang dari 1650 ƒÊm dapat digunakan sebagai masuk dan kontrol sinyal , masing-masing. Intensitas kerugian akibat skrup dan splitter dalam tahap interkoneksi tidak mungkin menciptakan banyak kesulitan dalam memproduksi bit optik yang diinginkan pada output sebagai keseluruhan sistem adalah salah satu digital dan output tergantung pada keberadaan dan ketiadaan cahaya . di Kinerja Sistem Komunikasi Serat Optik Menggunakan MZI Switching

107

skrup tahap interkoneksi serat dapat digunakan sebagai pengganti splitter balok . Empat channel 4 X 40 Gb / s sistem OTDM ( semua saluran ) dengan modulator Mach - Zehnder , MZI switching dan panjang serat dari 100 Km , telah eksperimental dan berhasil diverifikasi . Hasil penelitian mengungkapkan bahwa BER menurun dengan meningkatnya kekuatan sinyal dan peningkatan lebar pulsa . Seperti dalam tesis BER ini meningkat dengan peningkatan kekuatan sinyal kontrol dengan dispersi dalam serat mode tunggal . Hal ini juga disimpulkan bahwa kinerja sistem OTDM dapat

 

ditingkatkan dengan menggunakan dispersi kompensasi serat. 5.2 ASPEK FUTURE logika All- optik adalah penelitian terbaru di bidang komputasi optik sebagai skema ini juga memberikan ide memori optik jika kita merancang sebuah flip-flop optik yang menyimpan data sebagai pulsa optik . Seperti FTTH memiliki banyak keuntungan atas semua teknik transmisi sehingga , Penyedia bisa menggunakan ATM , SONET , Ethernet atau Analog termodulasi pembawa RF sebagai teknologi lapisan data link mereka . Karena semua pengguna dilayani oleh splitter yang sama . Combiner pada PON tepi jalan ( dan oleh Remote Node yang sama dalam sebuah arsitektur Bintang Aktif ) harus dilayani oleh teknologi lapisan data - link yang sama . Infrastruktur FTTH yang berteknologi dan kompetitif netral ; di mana penyedia suara , video dan layanan data dapat memilih dan menggunakan teknologi pilihan mereka untuk mendukung layanan mereka berencana untuk menawarkan . FTTH juga menyediakan layanan tambahan di atasnya seperti UWB ( Ultra wide band ) , WCDMA , Radio atas serat , begitu banyak layanan lain seperti jaringan akan menggunakan jaringan FTTH sebagai interface untuk jaringan akses . Fokus telah dimasukkan pada sistem PON masa depan-bukti memiliki gigabit simetri dalam bandwidth antara downlinks atas dan bawah. Telah terbukti bahwa OCDMA mampu memberikan gigabit - atau bahkan multi- gigabit - per - detik untuk setiap pengguna baik di up- dan downlinks , dan OCDMA over WDM PON bisa menjadi salah satu arsitektur yang paling menjanjikan yang dapat menembus yang terakhir / pertama mil bottleneck .

REFERENSI

[ 1 ] Jitendra Nath Roy , ? ? \ Mach - Zehnder interferometer berbasis arsitektur pohon untuk semua - optik logika dan operasi aritmatika . , Optik Int Cahaya Elektron Opt . (2009) .

[ 2 ] Koji Igarashi dan Kazuro Kikuchi , ? ? \ Optical Signal Processing oleh Phase Modulation dan selanjutnya spektral Filtering Bertujuan Aplikasi untuk ultrafast Sistem Komunikasi Optik . , IEEE jurnal topik yang dipilih dalam elektronik kuantum , Vol . 14 , No 3 , Mei / Juni .

[ 3 ] K. Uchiyama , H. Takara , K. Mori , T. Morioka , ? \ 160 Gbit / s all- optical time-division demultiplexing memanfaatkan dimodifikasi multiple- output yang OTDM demultiplexer ( MOXIC ) . , Electron . Lett . 38 ( 2002) 1190,1191 .

[ 4 ] I. Shake, H. Takara , I. Ogawa , T. Kitoh , M. Okamoto , K. Magari , T. Ohara , S. Kawanishi , 0,160 - Gbit / s channel penuh optical time-division de- multiplexer berdasarkan SOA -array terpadu PLC dan aplikasi untuk OTDM percobaan transmisi . , IEICE Trans . Commun . 53 ( 1 ) ( 2005 ) 20,2096 .

[ 5 ] H. Le - Minh , Z. Ghassemlooy , W.P. Ng , ? ? \ Penindasan crosstalk dalam simetris Mach.Zehnder semua - optik ( SMZ ) beralih dengan menggunakan pulsa kontrol dengan kekuatan yang tidak sama , Prosiding Simposium Internasional tentang Telekomunikasi 2005 ( IST 2005 ) . , Vol . 1 , Shiraz , Iran 2005 , hlm 265,268 .

[ 6 ] M. Heid , S. Spalter , G. Mohs , A. Farbert , W. Vogt , H. Melchior , ? \ 160 - Gb / s demultiplexing didasarkan pada interferometer Mach.Zehnder monolithically terintegrasi , Proceedings of the European konferensi Optical Komunikasi (ECOC 2001) " , Amsterdam , Belanda , September 30.October 4 , 2001.

[ 7 ] Haijiang Zhang , Pengyue Wen , dan Sadik Esener , . Cascadable inverter semua - optik

berdasarkan vertikal - rongga semikonduktor amplifier optik nonlinier . , Opt . Lett . 32 , 1884-1886 ( 2007) .

[ 8 ] Yanming Feng , Xiaofan Zhao , Li Wang , dan Caiyun Lou , ? ? \ Kinerja tinggi semua - optik OR / NOR gerbang logika dalam semikonduktor amplifier optik tunggal dengan keterlambatan gangguan filtering , Appl .. , Opt . 48 , 2638-2641 (2009) .

 

? ? [ 9 ] Jitendra Nath Roy dan Dilip Kumar Gayen , \ Terpadu logika semua - optik dan arithmeticoperations dengan bantuan perangkat interferometer berbasis Toad - pendekatan alternatif MF Lane , DZ Chen , dan D. Kokkinos , Mengelola Fiber Connections di NGN dan Aplikasi , di Konferensi Fiber Optic Nasional . , OSA Teknis Digest Series ( CD ) ( Optical Society of America , 2007) , kertas NThA1 .

[ 10 ] Petrantonakis , P. Zakynthinos , D.Apostolopoulos , A.Poustie , G. Maxwell , dan H.

Avramopoulos , . All- Optical Empat - Wavelength Burst Mode Regenerasi Menggunakan Integrated Quad SOA - MZI Arrays . , IEEE Photonics TEKNOLOGI LETTERS , VOL . 20 , NO . 23 , 1 Desember 2008 .

[ 11 ] Colja Schubert , Jorn Berger , Stefan Diez , Hans Jurgen Ehrke , Reinhold Ludwig , Uwe Feiste , Carsten Schmidt , Hans G. Weber , Gueorgui Toptchiyski , Sebastian Randel , dan Klaus Petermann , ? ? \ Erbandingan Interferometric Switches All- Optical untuk Demultiplexing Aplikasi di Internet Kecepatan Tinggi OTDM Systems. , JOURNAL OF TECHNOLOGY Lightwave , VOL . 20 , NO . 4 , April 2002 .

[ 12 ] K. Kitayama , T. Kuri , JJ Vegas Olmos , dan H. Toda , ? ? \ Fiber - Wireless Networks dan Radio - over - Fiber Teknik , di Konferensi Laser dan Electro-Optics/Quantum Elektronika dan Laser Science Conference dan fotonik Aplikasi Sistem Teknologi . , OSA Teknis Digest ( CD ) ( Optical Society of America , 2008) , kertas CThR4 .

[ 13 ] R. Llorente , T. Alves , M. Morant , M. Beltran , J. Perez , A. Cartaxo , dan J. Marti , . Distribusi Optical dari OFDM dan Impulse - Radio UWB di FTTH Networks . , Di National 74 Fiber Optic Engineers Conference , OSA Teknis Digest ( CD ) ( Optical Society of America , 2008) , kertas JWA109 .

[ 14 ] Kuniharu Himeno , Shoichiro Matsuo , Ning Guan , dan Akira Wada , Low- Bending - Rugi Single- Mode? ? \ Serat untuk Fiber- to-the -Home . , J. Cahaya gelombang Technol . 23 , 3494 - (2005)

[ 15 ] D. Iazikov , C. Greiner , dan TW Mossberg , ? ? \ Filter Apodizable Terpadu untuk Rambut Kasar WDM dan FTTH -Type Aplikasi . , J. Cahaya gelombang Technol . 22 , 1402 - ( 2004)

[ 16 ] MF Lane , DZ Chen , dan D. Kokkinos , . Mengelola Fiber Connections di NGN dan Aplikasi . , Di Fiber Optic Nasional Engineers Conference , OSA Teknis Digest Series ( CD ) ( Optical Society of America , 2007) , kertas NThA1 .

 

0121: Anto Satriyo Nugroho dkk. TI-15

PENGEMBANGAN SISTEM COMPUTER AIDED DIAGNOSIS BERBASIS FREE OPEN SOURCE SOFTWARE

Anto Satriyo Nugroho'*, Made Gunawan', Ismail Ekoprayitno Rozi2, Vitria Pragesjvara',

Miranti Jatnia Riski', Desiani', Nuke Dewi Kania, Theda Lukito, Agung Riyadi, Yuni Arti, Ninon Nurul Faiza

'Pusat Teknologi Informasi & Komunikasi, Badan Pengkajian & Penerapan Teknologi

Gedung Teknologi 3 Lantai 3 Puspiptek Serpong, Tangerang 15314

Tel. (021) 75791260 Ext.236 Fax. (021) 75791248

2Divisi Bioinformatika, Lembaga Biologi Molekuler Eijkman

Jl. Diponegoro 69, Jakarta 10340

Tel. (021) 3917131 Fax. (021) 3147982

*e-Mail: anto.satriyo@bppt.go.id, asnugroho@gmail.com

Disajikan 29-30 Nop 2012

ABSTRAK

Malaria merupakan penyakit tropis utama yang hingga kini masih banyak ditemukan di berbagai daerah di Indonesia. Penyakit Malaria disebabkan oleh infeksi protozoa pada darah yang disebut plasmodia. Parasit ini terdiri dari 4 spesies: Plas¬modium vivax, Plasmodium falciparum, Plasmodium ovale dan Plasmodium malariae. Penyakit Malaria diindikasikan dari berbagai gejala seperti menggigil, demam tinggi, sakit kepala, anemia dan pembesaran Limfa. Di Indonesia diperkirakan jumlah penderita Malaria mencapai sekitar 15 juta, dengan angka kematian 30 ribu per tahun. Angka ini mengindikasikan bahwa Malaria merupakan salah satu penyakit tropis di Indonesia yang harus mendapatkan penanganan serius dan cepat. Salah satu faktor penting dalam penanganan kasus Malaria adalah diagnosa dini Malaria yang akurat. Badan Pengkajian Penerapan Teknologi telah bekerja sama dengan Lembaga Biologi Molekuler Eijkman mengembangkan sistem Computer Aided Diagnosis berbasis teknik pengolahan citra yang mampu membantu proses diagnosa agar berlangsung cepat dan akurat. Dalam periode pertama penelitian ini (2011), telah dikembangkan prototype sistem Computer Aided Diagnosis memanfaatkan teknologi pengo-lahan citra, untuk mengidentifikasi status penyakit Malaria dari citra apusan tipis darah pasien Malaria, sedangkan pada tahun kedua (2012), sistem ini disempurnakan dengan memasukkan modul klasifikasi komponen darah. Sistem yang dikembangkan merupakan transformasi dari pengetahuan yang dimiliki oleh ekspert microscopist ke dalam rule-based classifier dikombinasikan dengan Bayes Decision. Dalam uji coba dengan data citra mikroskopis darah yang dikumpulkan dari berbagai daerah di Indone¬sia, sistem ini menunjukkan hasil yang relatif bagus. Akurasi identifikasi tiap komponen: erythrocyte 94.75%, leucocytes 90%, platelets 41.28% dan erythrocyte yang terinfeksi oleh parasit 84.5%, sedangkan akurasi secara keseluruhan sekitar 77.53%.

Kata Kunci: Computer aided diagnosis, malaria, microphotograph, apusan tipis

 

I. PENDAHULUAN

Malaria merupakan penyakit tropis utama yang hingga kini masih banyak ditemukan di berbagai daerah di Indonesia. Penyakit Malaria disebabkan oleh infeksi protozoa pada darah yang disebut plasmodia. Parasit ini terdiri dari 4 spesies: Plasmodium Vivax, Plasmodium Falciparum, Plasmodium Ovale dan Plas-modium Malariae. Indikasi penyakit ini terlihat dari berbagai gejala seperti menggigil, demam tinggi, sakit kepala, anemia dan pembesaran Limfa. Di Indone¬sia diperkirakan terdapat sekitar 15 juta penderita, de-ngan angka kematian 30 ribu per tahun. Angka ini mengindikasikan bahwa Malaria merupakan salah satu 

 

penyakit tropis di Indonesia yang harus mendapatkan penanganan serius dan cepat.[1]

Salah satu faktor penting dalam penanganan ka-sus Malaria adalah diagnosa dini Malaria yang akurat. Dalam praktek medis, dikenal berbagai teknik diagnosa seperti PCR, diagnosa klinis dan diagnosa mikroskopi. Di antara teknik tersebut, diagnosa mikroskopi meru-pakan teknik gold standard dan relatif paling murah. Teknik ini memanfaatkan mikroskop untuk menghi-tung parasit, di mana sampel darah diberikan pewar-naan melalui proses kimia yang disebut (Giemsa) stain¬ing. Selain murah, diagnosa mikroskopi juga bersi-fat sensitif, yaitu dapat menentukan kepadatan para 

 

 

Prosiding InSINas 2012

 

TI-16 0121: Anto Satriyo Nugroho dkk.

 

sit dari darah pasien. Diagnosa ini juga informatif, yaitu dapat dipakai untuk menentukan jenis plasmod¬ium Malaria, yang merupakan salah satu pertimban-gan dalam memberikan terapi kepada pasien. Akan tetapi diagnosa mikroskopi ini juga memiliki berba-gai kelemahan. Akurasi diagnosa sangat tergantung kepada keahlian dan pengalaman mikroskopis (orang yang bertugas membaca slide apusan darah). Ter-lebih lagi, prevalensi Malaria di Indonesia cukup ren-dah, sehingga slide yang dibaca umumnya negatif, maksudnya tidak terjangkit Malaria. Padahal untuk memastikan bahwa seorang pasien darahnya negatif jauh lebih sulit daripada menentukan bahwa seorang pasien statusnya positif terjangkit Malaria. Untuk itu, Badan Pengkajian Penerapan Teknologi bekerja sama dengan Lembaga Biologi Molekuler Eijkman dalam pe-ngembangan sistem Computer Aided Diagnosis berba-sis teknik pengolahan citra, yang mampu membantu proses diagnosa, agar berlangsung cepat dan akurat.

Dari kajian literatur, beberapa pihak melakukan upaya serupa untuk kasus Malaria di luar negeri, akan tetapi terbatas pada kasus dengan skala kecil. Misal-nya Premaratne memakai artificial neural network yang dikhususkan pada spesies Plasmodium falciparum. Model tersebut memakai citra sel yang telah dinor-malisasikan sebagai input bagi proses training maupun testing neural network.[2] Studi lain dilakukan oleh Ross yang memakai warna, tekstur dan ciri geometri dari parasit dan sel sebagai vektor fitur, yang selan-jutnya diklasifikasikan ke salah satu spesies plasmod¬ium dalam dua tahap.[3] Akan tetapi studi yang mereka lakukan hanya terbatas pada data berskala kecil, 5 sam-pel untuk training dan 15 sampel untuk testing, sedang-kan karakteristik parasit sangat kompleks. Dapat dis-impulkan bahwa hingga kini, belum ada studi kom-prehensif yang dilakukan untuk mengembangkan com¬puter aided diagnosis terhadap kasus Malaria, khusus-nya di Indonesia berskala besar.

Dalam periode pertama penelitian ini (2011), telah dikembangkan prototype sistem Computer Aided Di¬agnosis memanfaatkan teknologi pengolahan citra, un-tuk mengidentifikasi status penyakit Malaria dari citra apusan tipis darah pasien Malaria. Beberapa algoritma baru pengolahan citra telah dikembangkan oleh tim peneliti dan bekerja sama dengan ahli mikroskopi di Lembaga Biologi Molekuler Eijkman, sehingga hasilnya dapat diverifikasi langsung oleh praktisi lapangan.[4, 5] Pada tahun pertama penelitian, algoritma tersebut telah dievaluasi pada data citrayang telah terinfeksi oleh Plasmodium Falciparum, dengan hasil cukup baik. sen¬sitivity 92.59% dan specificity of 99.65%, Positive Pre¬dictive Value (PPV) of 67.56%, and F1 score of 0.7812.[5]

Dalam tahun kedua penelitian ini (2012), kegiatan penelitian difokuskan pada penyempurnaan prototype system sehingga bisa dipakai untuk mendeteksi keber- 

 

adaan berbagai spesies parasit selain Plasmodium fal-ciparum (P. malariae, P. ovale, P. vivax) , pada citra de-ngan berbagai pada kualitas lapangan.

II. METODOLOGI

A. Persiapan Data Microphotograph

Sample darah pasien Malaria dipersiapkan oleh tim peneliti Lembaga Biologi Molekuler Eijkman, diperoleh dari berbagai daerah di Indonesia. Sample tersebut ter-diri dari 5 spesies: Plasmodium falciparum, Plasmod¬ium vivax, Plasmodium ovale, Plasmodium Malariae, dan Plasmodium Knowlesi.

Setiap slide dilakukan proses staining memakai standard Giemsa staining protocol. Slide kemudian diperiksa di bawah mikriskop dengan pembesaran 10100, dan citra ditangkap dengan 5-megapixel Nikon digital sight DS 5 Mc, yang didesain khusus untuk mikroskop cahaya. Data set tersebut kemudian dianal-isa secara manual, dan dilakukan perhitungan jumlah parasite, serta fase dan tipe masing-masing sel.

B. Sistem Computer Aided Diagnosis

Sistem yang dikembangkan merupakan transformasi pengetahuan yang dimiliki oleh microscopist analyst Lembaga Biologi Molekuler Eijkman ke dalam algo-ritma pengolahan citra. GAMBAR 1 memperlihatkan bagan sistem yang dibangun untuk melakukan iden-tifikasi citra microphotograph darah pasien. Sistem itu terdiri dari fase preprocessing untuk memperbaiki kualitas citra dan menghapus noise, dilanjutkan dengan fase segmentasi komponen darah dan klasifikasi kom-ponen darah. Tujuan dari dua fase ini adalah menge-nali jenis berbagai komponen darah, seperti erythro-cyte, leucocyte, thrombocyte, dan artefacts.

Erythrocyte sendiri dibagi dua: erythrocyte sehat dan suspected erythrocyte, yaitu erythrocyte yang di-curigai terinfeksi oleh parasit. Selanjutnya Bayes Clas-sifier dipakai untuk menentukan apakah erythrocyte tersebut sehat (bercak hitam itu hanya sekedar artefak) ataukah erythrocyte tersebut benar-benar terinfeksi par-asit.

 

GAMBAR 1: Rancangan sistem Computer Aided Diagno¬sis Malaria

 

 

Prosiding InSINas 2012

 

0121: Anto Satriyo Nugroho dkk. TI-17

 

B-1. Prepocessing

Dalam tahap ini, citra dikonversikan ke gray-scale image, karena informasi warna tidak diperlukan untuk melakukan klasifikasi komponen sel darah. Noise pada citra dihapus memakai Median filter. Pada GAMBAR 2 diperlihatkan gambar berbagai komponen darah: (1) erythrocyte (sel darah merah), (2) leucocyte (sel darah putih) dan (3) platelet dan (4) erythrocyte yang mengan-dung parasit.

 

GAMBAR 2: sampel citra darah dengan berbagai kompo-nen yang diberi nomor

B-2. Segmentasi komponen darah

Tahap ini bertujuan untuk melakukan segmentasi, yaitu memisahkan tiap citra komponen darah dari citra latar belakang. Pada citra input, intensitas terbagai ke dalam 3 level: paling terang, menengah dan pa¬ling gelap. Citra thrombocytes, leucocytes dan artefacts umumnya terbentuk dari pixel dengan warna gelap, se-dangkan erythrocytes tersusun dari pixel dengan in-tensitas medium (abu-abu), dan latar belakang paling terang. Karena ada 3 intensitas yang dominan, citra in-put didekomposisikan ke dalam 3 citra memakai Otsu Dual Thresholding. Citra yang diperoleh terdiri dari 3 dan disebut mask. Mask pertama mengidentifikasikan posisi seluruh komponen darah dan dinotasikan de-ngan A.

Mask kedua adalah citra yang terdiri dari pixel

dengan warna paling gelap. Pixel pada mask

ini mengindikasikan posisi artefact, nucleus parasit, thrombocytes dan leucocytes. Mask kedua diidenti-fikasikan dengan Λ.

Mask yang terakhir mengindikasikan posisi erythro-cytes, yaitu yang tersusun dari pixel dengan warna medium. Mask ini dinotasikan dengan E Karena ery-throcytes memiliki struktur cekung, pada saat dual thresholding dilakukan, akan terbentuk lobang di ten- 

 

gah. Agar proses segmentasi berjalan benar, hasil seg-mentasi tersebut diperbaiki dengan proses morfologi hole-filling yang dikembangkan pada penelitian ter-dahulu.[4] GAMBAR 3-4 menunjukkan proses morfologi hole filling yang diterapkan pada sebuah citra, sehingga diperoleh hasil dengan lubang tertutup.

 

GAMBAR 3: Fast Hole Filling

B-3. Klasifikasi tahap pertama: klasifikasi komponen berbasis aturan

Setelah 3 mask file diperoleh, komponen darah diiden-tifikasi berdasarkan aturan berikut: A mewakili semua komponen dari citra darah,  mewakili komponen dengan pixel gelap, E mewakili komponen dengan intensitas piksel sedang. Lima kelas didefinisikan sebagai berikut: ω1 erythro¬cyte sehat, ω2: leucocyte ω3: thrombocyte, ω4: Thrombocyte menumpang di erythrocyte ω5: suspected erythrocyte. Selan-jutnya klasifikasi dilakukan berdasarkan aturan-aturan seba-gai berikut:[6]

 

GAMBAR 4: kiri: mask file sebelum dilakukan hole filling kanan: setelah hole filling

(i) X  A s.t. X / X  E X  ω1

(ii) X  A s.t. X  φX  E  φ X  ω3

 

 

Prosiding InSINas 2012

 

TI-18 0121: Anto Satriyo Nugroho dkk.

 

(iii) X  A s.t. (X  Λ = 0)  (X  E = 0) (τ1  area(X)  τ2)  X  ω4

(iv) X  A s.t. (X  Λ = 0)  (X  E = 0) (area(X) > τ2)  X  ω2

(v) X  A s.t. (X  A = 0)  (X  E = 0) (area(X) < τ1)  X  ω5

B-4. Klasifikasi tahap kedua

Apabila sebuah erythrocyte diklasifikasikan ke dalam ω5 (suspected erythrocyte), maka sel tersebut akan diidenti-fikasikan lebih lanjut untuk memastikan apakah benar-benar terinfeksi oleh parasit atau bercak hitam itu hanya sekedar artefact. Verifikasi ini dilakukan dengan mempartisi sel ke dalam 3 area berdasarkan Otsu dual thresholding (lihat GAM-BAR 5), kemudian menghitung rasio white/black pixel seba-gaimana diperlihatkan pada GAMBAR 6, sebagai acuan untuk membuat keputusan dengan Bayes classifier.[5]

 

GAMBAR 5: hasil dual thresholding terhadap erythro¬cyte yang terinfeksi oleh parasit dan erythrocyte yang ditumpangi oleh artefacts

 

GAMBAR 6: rasio white/black pixel sebagaimana se-bagai acuan untuk membuat keputusan dengan Bayes classifier[5]

III. HASIL DAN PEMBAHASAN

Computer Aided Diagnosis system yang dibangun selan-jutnya dievaluasi dengan 40 citra microphotograph. Tingkat 

 

keberhasilan yang dicapai secara keseluruhan adalah 77.63%, dengan akurasi masing-masing: erythrocyte 94.75%, leuco-cytes 90%, platelets 41.28% dan erythrocyte yang terinfeksi oleh parasit 84.5%.[6] Analisa terhadap error menunjukkan bahwa umumnya hal itu disebabkan oleh kegagalan segmen-tasi. Contoh dari kegagalan segmentasi ini diperlihatkan pada GAMBAR 7, di mana thrombocyte masih tersambung dengan erythrocyte, sehingga diidentifikasikan sebagai para-sit. Beberapa kesalahan lain disebabkan oleh komponen yang tumpang tindih (overlapping). Saat ini tengah dikaji solusi dari masalah tersebut dengan memanfaatkan teknik segmen-tasi watershed.[8]

 

GAMBAR 7: Error yang disebabkan kegagalan segmen-tasi. Thrombocyte masih tersambung dengan erythro¬cyte, sehingga diidentifikasikan sebagai parasit

Eksperimen juga dilakukan terhadap spesies yang lain. Eksperimen ini memakai 5 slide P. Falciparum dan 7 slide P.vivax pada fase asexual. Tiap slide diambil data sebanyak 150 citra microphotograph. Hasil eksperimen ini menun-jukkan bahwa sistem mampu mendeteksi fase aseksual P. fal-ciparum dan P. vivax relatif baik. Untuk P. falciparum, tingkat sensitifitas sistem 73.59% sedangkan specifisitas 98.54%, PPV (Positive Predicted Value) 23.97% dan score F1 sebesar 29.54%. Adapun P. vivax, sensitifitas sistem sebesar 79.15%, spesifisi-tas 98.42%, PPV 7.62%, dan score F1 sebesar 12.69%.[7]

IV. KESIMPULAN

Studi ini merupakan bertujuan untuk mengembangkan sistem Computer Aided Diagnosis, yang membantu praktek klinis di lapangan agar bisa melakukan diagnosa dini Malaria dengan cepat dan akurat. Sistem ini memakai citra micropho¬tograph apusan tipis darah pasien Malaria yang telah dilaku-kan pewarnaan dengan Giemsa staining. Sistem yang diban-gun merupakan transformasi dari pengetahuan ekspert di-agnosa mikroskopis yang sering disebut microscopist pada Lembaga Biologi Molekuler Eijkman, ke dalam algoritma pe-ngolahan citra. Hasil yang diperoleh dibandingkan dengan hasil diagnosa yang dilakukan secara manual oleh para mi¬croscopist tersebut.

Salah satu masalah utama dalam studi ini adalah melakukan klasifikasi komponen darah, dan menentukan ada tidaknya sel darah merah yang terinfeksi. Klasifikasi kom-ponen darah berhasil dilakukan dengan membagi citra darah ke 3 sub citra secara otomatis berdasarkan kemiripan intensi-tasnya, dan selanjutnya mendefinisikan class tiap komponen

 

 

Prosiding InSINas 2012

 

0121: Anto Satriyo Nugroho dkk. TI-19

 

lewat klasifikasi berbasis aturan (rule based classifier). Secara keseluruhan, algoritma yang dikembangkan mampu mende-teksi 77.53% class tiap komponen darah, dengan akurasi masing-masing: erythrocyte sehat 94.75%, suspicious erythro¬cytes 84.50%. Suspicious erythrocytes ini selanjutnya akan diidentifikasikan dengan Bayes rule, apakah memang benar-benar positif parasit ataukah hanya artefak yang kebetulan posisinya di erythrocyte.

Prototype piranti lunak ini sedang dikembangkan lebih lanjut dengan integrasi pada hardware, yaitu microscope camera dan mikroskop dan disambungkan pada laptop yang ringkas dan dapat dipakai secara portable untuk melakukan identifikasi status Malaria pada daerah rawan. Tahun 2013 di-rencanakan untuk mengevaluasi sistem di kawasan Indonesia timur, yang prevalensi Malaria yang cukup tinggi.

DAFTAR PUSTAKA

[1] US Namru-2 Jakarta (2008), Panduan Pelatihan Diagno¬sis Mikroskopi Malaria, Departemen Parasitologi Medis,

[2] S.P. Premaratne, N.D. Karunaweera, S. Fernando, W.S.R. Pererab, R.P. Asanga, S. Rajapaksha (2003), A Neu¬ral Network Architecture for Automated Recognition of Intracellular Malaria Parasites in Stained Blood Films, APAMI & CJKMI-KOSMI Conference 2003

[3] N.E. Ross, C.J. Pritchard, D.M. Rubin, A.G. Duse, (2006), Automated Image Processing method for the diagnosis and classification of Malaria on thin blood smears, Med¬ical and Biological Engineering and Computing, Vol.44, No.5, pp.427-436, 2006

[4] D. Anggraini, A.S. Nugroho, C. Pratama, I.E. Rozi, A.A. Iskandar, R.N. Hartono (2011), Automatic Status Iden-tification of Microscopic Images Obtained from Malaria Thin Blood Smears, Proc. of 3rd International Conference on Electrical Engineering and Informatics (ICEEI 2011), CDROM A3-2, Institut Teknologi Bandung, Bandung, In-donesia, July 17-19, 2011

[5] D. Anggraini, A.S. Nugroho, C. Pratama, I.E. Rozi, V. Pragesjvara and M. Gunawan (2011), Automated Sta¬tus Identification of Microscopic Images Obtained from Malaria Thin Blood Smears using Bayes Decision: A study case in Plasmodium Falciparum, Proc. of Interna¬tional Conference on Advance Computer Science and In¬formation System 2011, Jakarta, Indonesia, 17-18 Decem¬ber 2011

[6] N.D. Kania, T. Lukito, A.S. Nugroho, I.E. Rozi, M. Gu-nawan, V. Pragesjvara, D. Anggraini (2012), Blood Com-ponent Classification for Malaria Computer Aided Diag¬nosis from Thin Blood Smear Microphotographs, Proc. of 5 International Symposium on Computational Science (ISCS), pp.199-206, 15-16 May 2012

[7] T. Lukito, N.D. Kania, I.E. Rozi, A.S. Nugroho, M. Gunawan, V. Pragesjvara, D. Anggraini (2012), Semi-automated Computer-aided Diagnosis for Malaria Multi species Parasite Detection from Thin Blood Smear Mi¬crophotographs”, Proc. of 5th International Symposium on Computational Science (ISCS), pp.296-302, 15-16 May 2012

[8] A. Riyadi, M.J. Riski, N.N. Faiza, Y. Arti, A.S. Nugroho, M. Gunawan, I.E. Rozi (2012), Evaluasi Algoritma Wa¬tershed pada Pemisahan Citra Apusan Tipis Sel Darah

 

Prosiding InSINas 2012

 

Procedural Guide for Ph.D. Students Human Development and Family Science

Auburn University

A PROCEDURAL GUIDE FOR

DOCTOR OF PHILOSOPHY STUDENTS

DEPARTMENT OF HUMAN DEVELOPMENT AND FAMILY SCIENCE

AUBURN UNIVERSITY

updated September 2021

TABLE OF CONTENTS

Introduction

Appointing a Temporary Advisor

Selecting a Major Professor

Forming an Advisory Committee

HDFS Graduate Faculty

Plan of Study

Graduate School Course Requirements for the Plan of Study

Professional Socialization Guidelines

Lines of Authority

Annual Evaluation

Steady Progress toward the Degree

Evaluation Process

Evaluation Outcomes

HDFS Assistantship Policy

M.S. Ph.D. Students

Ph.D. Students

Graduate Assistant Evaluation Policy

Doctoral Competencies

The Doctoral Qualifying (General/Preliminary) Examination

Guidelines for the Qualifying (General/Preliminary) Exam

The General Oral Exam (Defense of the Written Exam)

Dissertation and Proposal Guide

Timing of the Dissertation Proposal Meeting

Guidelines for Working with Your Committee

Requirements for a Proposal

Elements of a Dissertation

Final Examination (Dissertation Defense)

 

1

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

INTRODUCTION

The Doctor of Philosophy is conferred in recognition of the mastery of a special field of learning as shown by the satisfactory completion of a prescribed course of study and investigation, the successful passing of general examinations covering the major and minor fields, the preparation of an acceptable dissertation reflecting high achievement in scholarship and independent investigation, and the passing of a final examination on the dissertation and related subjects. The degree is a research degree not conferred merely upon fulfillment of technical requirements, but awarded in recognition of the ability to think and work independently, originally, and creatively in a chosen field.

(Italicized material throughout the text of the guide has been selectively drawn (or edited) from the graduate school web page and indicates stated policies or definitions of the Auburn University Graduate School. For more complete information go to http://www.grad.auburn.edu/.)

During your first year in the Ph.D. program in the Department of Human Development and Family Science (HDFS) you will make a number of important decisions. These decisions may influence not only the direction of your doctoral training but also the course of your early career. First, you will select your major professor and the advisory committee that will direct your studies and examine your progress at several critical points during your program. Second, you will develop your tentative plan of study to identify the courses you will take as electives and supporting emphases. Third, you will begin planning and pursuing the specific skills and experiences needed to prepare for your career. This guide is intended to clarify the purposes of, and procedures associated with, each of these decisions. Ultimately, you are responsible for learning and following HDFS and Graduate School policies and procedures.

APPOINTING A TEMPORARY ADVISOR

If you have entered the doctoral program in HDFS without coming through our master’s program and you have not specified a faculty member as your preferred major professor, the graduate program officer (GPO) will serve as your temporary advisor. As temporary advisor, the GPO will advise you regarding your course of study in the first semester. During the first semester, we encourage you to identify a faculty member to serve as your major professor. The process of selecting a major professor and advisory committee is important and addressed in the following sections.

SELECTING A MAJOR PROFESSOR

You may select a major professor at any time during your first semester of study; it is important to select your major professor no later than the second semester of your first year. Your major professor must be a member of the Graduate Faculty and eligible to direct doctoral students. In making your selection, you should set individual appointments with at least two faculty members who appear to have professional and research interests congruent with your own. During the appointments, you should evaluate the fit between your own interests and the relevant professional and research interests of each potential major professor. In addition, you should discuss with each professor their style of mentoring. Beyond questions you arrive at on your own or in conversations with other students who have already made these decisions, consider seeking answers to the following questions:

* How many students is the faculty member working with? (Sometimes having several students means an advisor will have limited time; sometimes it means the advisor is especially effective as a mentor. Get the names of some of the professor’s advisees and talk with them before committing to an advisor.)

2

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

* How connected are students to the professor’s research program? (Put differently, will you work with the professor on his/her own research program or will you be allowed [required] to pursue an interest that is entirely your own? There are advantages and disadvantages of each approach. Talk about them. Talk with current advisees of the professor you are considering.)

* How available is the professor to students?

* What kinds of opportunities are available to students working with the professor? (To the extent that you know what you want to do with your degree, you will also be able to identify [with the assistance and advice of your major professor and advisory committee] experiences that will promote your capacity to be competitive for that kind of career. Will you have opportunities to get those experiences while working with the professors you are considering? Talk about the kinds of teaching experiences you will get while working with the professor. What kinds of research experiences will you get? Will there be publication opportunities? Grant writing opportunities? What kinds of opportunities will you get to apply your knowledge in the service of others or in a policy context?)

We encourage you to choose your major professor carefully. Before making the decision, consider all appropriate faculty prospects and talk with other students who have worked with the professors you are considering. You can identify possible major professors by reviewing information about faculty research interests on the HDFS web page. In addition, the introductory seminar and department events should facilitate your awareness of well-matched professors. Finally, a discussion with the department head or GPO can be helpful in directing you to faculty whose interests are aligned with yours.

Doctoral students sometimes wish to change their major professor, or major professors may want to discontinue their role with a student. On such occasions, the student and advisor are encouraged to meet to discuss their wishes and attempt to arrive at a mutually acceptable arrangement. The GPO and department head are available to help if concerns arise. An important point is that students cannot progress in the program without a major professor. Hence, discontinuation from one major professor must be accompanied by realignment with another major professor for the student to continue in the program.

FORMING AN ADVISORY COMMITTEE

You should form an advisory committee after selecting a major professor. The advisory committee must be

selected in consultation with your major professor. We encourage you to select the advisory committee during the

first year of the doctoral program.

When selecting an advisory committee, consider the theoretical/methodological fit between your major professor and the prospective committee members, the general interest domains and specific content areas of the

prospects, and the degree to which you and the major professor work effectively with the prospective members. Your major professor’s input is critical in the selection of the advisory committee.

The Graduate School indicates that: The advisory committee is responsible for developing the student's plan of study and conducting the doctoral general and final examinations. It should consist of at least four members of the graduate faculty. A majority of the Auburn University affiliated committee members, including the major professor, must be members of the Graduate Faculty, Level 2. The major professor must also be a graduate faculty member of the program granting the degree. Faculty eligible for directing doctoral students as major professor have considerable experience advising thesis or dissertation research, they have considerable research experience themselves, and they remain active in the process of research. Although the majority of advisory committee members must share this highest standing on the Graduate Faculty (Level 2), the primary criterion for the selection

3

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

of advisory committee members is the fit between the student’s dissertation interest and the faculty member’s expertise. The goal is to ensure that each doctoral student has the best combination of advisors and examiners available given his/her research interests. The student and major professor should collaborate in identifying the committee that fulfills this goal.

HDFS Graduate Faculty:

Francesca Adler-Baeder, Ph.D.

Katrina Akande, Ph.D.

Cory Cobb, Ph.D.

Adrienne Duke, Ph.D.

Mona El-Sheikh, Ph.D.

Stephen Erath, Ph.D.

Cynthia Frosch, Ph.D.

Thomas Fuller-Rowell, Ph.D.

Ben Hinnant, Ph.D.

Jennifer Kerpelman, Ph.D.

Scott Ketring, Ph.D.

Kyle Kostelecky, Ph.D.

Mallory Lucier-Greer, Ph.D.

Julianne McGill, Ph.D.

Joshua Novak, Ph.D.

Lauren Ruhlmann, Ph.D.

Diana Samek, Ph.D.

Wendy Troop-Gordon, Ph.D.

Brian Vaughn, Ph.D.

Silvia Vilches, Ph.D.

Angela Wiley, Ph.D.

Elif Dede Yildirim, Ph.D.

Some students may wish to have members on their advisory committee from outside Auburn University, such as a scholar from another university. Graduate school policy indicates that: Affiliate faculty and other scholars not employed by the University may serve on individual advisory committees with the approval of the Dean of the Graduate School. Such approval does not imply Graduate Faculty membership. The department head or other appropriate administrative officer should forward a recommendation for approval along with the scholarly credentials of the individual to the Graduate School for review and subsequent action.

The formal appointment of the advisory committee occurs when the Committee, Transfers, Exceptions, and Candidacy (CTEC) form (http://graduate.auburn.edu/current-students/committee-selection/) is approved by the Graduate School. Any change in the membership of a student’s advisory committee must be reported to the Graduate School via email with copies to the former and new committee members as well as the committee chair.

PLAN OF STUDY

Following selection of the major professor and advisory committee, the next step is to submit the CTEC form (CTEC; http://graduate.auburn.edu/current-students/committee-selection/). The purpose of the form is to identify the student, committee chair/major professor, committee members, and any transfer courses or curriculum

4

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

exceptions or substitutions. Transfer courses and curriculum exceptions or substitutions should be discussed and approved in advance by the committee chair/major professor. Doctoral students must submit and obtain approval of the CTEC form prior to submitting the Request for the Report on the General Oral Examination (i.e., before scheduling the qualifying exam defense with the Graduate School).

Graduate School Course Requirements for the Plan of Study

The minimum number of hours in a doctoral program is 60 semester hours earned through instruction beyond the bachelor’s degree, including 1) a minimum of 30 semester hours of graded (e.g., A, B) graduate course work (6000-level and above) and 2) a minimum of 30 semester hours of additional graduate course work (6000-level and above) that may include ungraded courses, 7990 (Thesis Research) and 8990 (Dissertation Research) and must include at least 10 hours of 8990. Some programs require more than 60 semester hours, and requirements may vary according to a student’s background and interests.

The total number of credit hours that may be transferred from another accredited institution toward a doctoral degree varies by program but must be less than 50% of the credit hours listed on the Plan of Study. Such transfer credit 1) must fall within the time limits of the degree, and 2) must be approved by the advisory committee and the Dean of the Graduate School. A maximum of 4 hours of 7990 from a completed master’s program may be counted. You may enroll in 8990 at any time during your program that your advisor considers appropriate. However, the number of 8990 hours taken in any given semester should reflect the amount of time being spent that semester on the dissertation and the degree to which university resources are being utilized. No grade is assigned to 8990 hours.

PROFESSIONAL SOCIALIZATION GUIDELINES

Working toward the Ph.D. is qualitatively different from working on an undergraduate or even a master’s degree. You are expected to work more independently (i.e., in a more self-motivated, self-directed way) and to perform at a higher level, producing better-quality work. At this stage in your scholarly development you are being trained as a professional colleague. Thus, over your doctoral program you should increasingly identify yourself as a professional rather than a student. Success in the Ph.D. program in HDFS, and later in the larger professional world, requires more than just completing a specified number of courses, passing exams, and completing a dissertation. It involves learning to work with other professionals, developing leadership skills, and gaining knowledge of professional protocols, as well as building research and academic skills. To promote these competencies, your committee will recommend experiences and provide feedback regarding research, teaching, and service. To develop professional relationship skills and to acquire knowledge of basic professional protocol, you will have opportunities to observe and experience various professional situations, again at the recommendation of your advisor and committee.

Good time management is essential to accomplishing all that will be required of you. You will need considerable time to focus on your coursework, but time to interact with fellow classmates and faculty outside the classroom is also important. Much of your learning will be informal and spontaneous and some of it will be vicarious as you observe the efforts and outcomes of other students in the program. Doctoral students are strongly encouraged to attend all HDFS public defenses of qualifying exams and dissertations when possible. Observing other students is valuable, as it familiarizes you with the process and provides performance models with both strengths and limitations. Although much of the feedback you will receive from faculty and fellow students regarding your development will be informal and spontaneous, the department also has a formal evaluation process that will provide direct feedback to you on an annual basis. The goal of both formal and informal feedback is improvement

5

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

and professional socialization.

The HDFS faculty want graduates of our program to be strong and competitive new professionals on the job market. Our evaluations of your performance in this department, in classes and assistantships but also in the development of your scholarly specialization and in other aspects of your professional socialization, will be reflected in letters of reference, recommendations for awards, etc. We want you to be successful.

LINES OF AUTHORITY

There may be times when you have a concern regarding a course, some aspect of the doctoral program, doctoral requirements, your assistantship, or some other issue. For concerns regarding a course or assistantship, we encourage you to first discuss the concern with the faculty member who is teaching the course or supervising the assistantship. If the concern is not satisfactorily addressed, then you should bring the concern to the GPO or Department Head. If the concern does not deal with a course or assistantship and the student does not know where to begin in order to address the concern, contact the GPO or Department Head.

ANNUAL EVALUATION

As a doctoral student, your progress in the doctoral program will be evaluated on an annual basis.

Steady Progress toward the Degree 

There is so much to do as you prepare for your career that it is easy to lose track of time. Students must understand that their success in the program is defined fundamentally in terms of the progress they are making toward the degree. Progress involves completion of the core coursework, electives, the MS thesis (if relevant to the student), the qualifying exam, and the dissertation, as well as the skill-building experiences recommended for students by the advisor and committee.

The Ph.D. is conceptualized as a 5-or-6-year degree program. This timeframe assumes that the student will take the full doctoral core of coursework including that which is in common with the MS degree. Therefore, the timeframe applies not only to students who come to the program with a bachelor’s degree and intend to proceed all the way to the Ph.D., but also to those students who complete an MS degree elsewhere and come to Auburn to pursue the Ph.D. All doctoral graduates will complete the full core unless they can show a transcript and syllabus that indicates the content of a graduate class taken elsewhere covers the content of a given class. The teaching faculty member decides whether the class content has been adequately addressed.

The 5-or-6 year timeframe is important as a standard against which to measure a student’s progress toward the degree. In order to accomplish the degree within this timeframe, students must keep their sights on completing the intermediate steps required by the degree. If adequate progress is not made, the assistantship support available to the student will be curtailed. Departmental assistantships and university fellowships are not available to students who are not making adequate progress toward the degree. Inadequate progress may or may not result in expulsion from the program, but if not, financial support will be the responsibility of the student not advancing at pace. Students for whom assistantship support is important to the ability to pursue graduate education must ensure that they progress appropriately.

6

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

Evaluation Process

Early in the spring semester, the GPO will direct you to complete a form outlining the progress you have made toward your degree goals. As part of this process, you will also declare your goals for the coming year in terms of research, teaching, and outreach/service. There will also be an opportunity for you to identity longer-range goals and to engage in self-evaluation in the context of your program. Along with this form, you will submit an updated curriculum vita. Other supporting documents such as teaching evaluations, extension/outreach products, conference presentations, and published papers that were generated during the evaluation year should be submitted with the vita. Evaluations of your assistantship performance (submitted by your assistantship supervisor) are considered in the review as are comments from your instructors and faculty who have had other exposure to you as a student in the program.

The goal of the evaluation process is to assess each student’s progress and accomplishments and to identify concerns or areas for improvement so that supportive action can occur. For students who are on track, suggestions about suitable opportunities may also be raised, and student accomplishments are brought to the attention of the faculty. Students have been nominated for significant student awards as a result of the annual review.

Evaluation Outcomes

Students will receive one of three overall ratings from the review.

• “Satisfactory progress” indicates the student is progressing as expected or better.

• “Concerns with progress” means the faculty have identified specific issues that must be addressed in the coming year. Students who receive a rating of "concerns" are funded as TAs or RAs only if their major professor is willing to provide funding for them or vouch for their progress the next year. Students for whom the faculty have concerns in consecutive years receive the rating “unsatisfactory” in the second year.

• “Unsatisfactory progress” indicates the faculty believe previously identified concerns have not been adequately addressed and/or the student is seriously off track in the program. Students with this rating will not be eligible for assistantship support for the coming year. A student rated "unsatisfactory" in one year must attain a rating of "satisfactory" during the following year to continue in the program.

Following the faculty meeting, your major professor will compile faculty feedback in a letter and share it with you. The letter becomes a part of your permanent record in the program.

HDFS ASSISTANTSHIP POLICY

All assistantship funding in the Department of Human Development and Family Science is contingent upon remaining in good standing in the program, which means that the student is making satisfactory progress toward the degree. Furthermore, all funding is dependent upon departmental need, the availability of funds, and the approval of the fiscal year budget by the Auburn University Board of Trustees.

Assistantship stipends are paid on different scales for M.S. versus Ph.D. students.

M.S. - Ph.D. Students

For students who originally entered the M.S. program and transitioned to the Ph.D. program, the pay scale remains at the M.S. level until the semester following the successful defense of the master’s thesis.

7

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

Because third-year M.S. students tracking into the doctoral program are taking doctoral core courses while completing their thesis, they are not considered off track in the doctoral program as long as they complete the thesis by the end of their third year in the program. A maximum of six years of departmental support (including any provided while an M.S. student in HDFS) is available to students whose annual reviews indicate that they have remained on track throughout their program.

Ph.D. Students

Students who joined the doctoral program with an M.S. degree from another program receive a stipend at the doctoral level of pay. They are required to take the full doctoral core unless they are able to transfer core classes from their master’s program but may transfer electives and may not need to complete a thesis (e.g., if they have already completed a thesis); thus, a maximum of five years of departmental support is available to students whose annual reviews indicate they have remained on track.

GRADUATE ASSISTANT EVALUATION POLICY

Each semester you hold a graduate assistantship, you should expect to meet with the supervisor of the assistantship at the beginning of the semester to discuss your duties and the advisor’s expectations. You should also expect at least one meeting, typically at the end of the semester, during which your supervisor reviews your assistantship performance. Typical items in the evaluation include whether you were: timely in your performance of assistantship duties, thorough and thoughtful in your role, responsive to feedback, able to take the lead in appropriate times and ways, a team player in relation to the people you work with in your duties, and developing appropriate competencies related to the assistantship.

There is a separate evaluation form for each type of assistantship. You should expect to see your supervisor’s evaluations and have the opportunity to discuss them in a meeting with her/him. These forms become part of your permanent record in the program and are also part of your annual evaluation. If you have concerns about your performance in an assistantship, you should approach your supervisor for feedback when your concerns arise rather than waiting to the end of the semester.

If you are the instructor of record for a class, the teaching evaluations completed by your students will also be part of your evaluation. They should be included with the material you provide for the annual review.

DOCTORAL COMPETENCIES

Doctoral students join the HDFS program with different goals, so the competencies or skills that they hope to develop in the program are likely to vary to some degree. However, the HDFS program has some goals in common for all doctoral students. We want all doctoral graduates to have the ability to perform competently as teachers, researchers, and providers of professional service or outreach. Thus, in addition to performing well in class, you will be asked to address the competencies below in your annual evaluation as you plan and implement your program and professional goals.

Teaching Competencies

1. Create appropriate materials for an HDFS course including course syllabi, homework, and tests.

2. Plan lectures relevant to an HDFS course and/or appropriate material for an entire HDFS course.

8

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

3. Implement academic honesty policies and ethical treatment of students.

4. Teach effectively.

Research Competencies

1. Conceptualize an HDFS-relevant research question and develop an appropriate research design.

2. Collect data.

3. Apply statistical software packages to independently to analyze data.

4. Develop an analysis strategy and interpret research data independently.

5. Prepare and submit manuscripts for publication.

Service Competencies

1. Participate in university, college, department, or community outreach.

2. Provide professional service including membership in professional organizations.

THE DOCTORAL QUALIFYING (GENERAL/PRELIMINARY) EXAMINATION

Before doctoral students can be admitted to candidacy for the Ph.D. degree, they must pass the written and oral phases of the qualifying (or general or preliminary) examination. The qualifying exam process may begin early in the student’s program but the qualifying exam cannot be defended until the student has completed all required (core) coursework in the doctoral program. Each doctoral student in HDFS takes a core of coursework intended to provide breadth in the student’s understanding of our multi-disciplinary field and to integrate its many threads. In addition, a prominent aspect of doctoral education is the development of depth in a specialization area that will serve as the foundation of the student’s professional career and contribution to the field. Both this breadth and depth occur as students pursue their coursework and their varied assistantship, practicum, and independent learning experiences working with other students and professors in the department. The qualifying (general/preliminary) examination in HDFS represents an opportunity for students to demonstrate the breadth and depth of their knowledge in an independent fashion as they take the lead in producing one of three research products (described below) in collaboration with their major professor or another professor (with the approval of the major professor). This product then becomes the primary focus of the oral component of the qualifying exam. Only after both written and oral parts of the qualifying (general/preliminary) exam have been passed is a student admitted to candidacy for the Ph.D. degree. Students must be admitted to candidacy before their dissertation proposal meeting.

The following is edited from the Graduate School’s policy regarding the general oral exam:

A general examination, often called the “preliminary examination,” is required of all applicants for the degree of doctor of philosophy. It consists of written and oral testing by the student’s advisory committee (or by an examination committee designated by the student’s academic program) in the student’s major and minor. The written portion of the examination does not require approval in advance by the Graduate School. The oral portion, however, does require such approval. Arrangements for the oral examination must be made by application to the Graduate School at least one week in advance of the examination. The primary purpose of the general examination is to assess the student’s understanding of the broad body of knowledge in a field of study. The examination also affords the advisory committee an opportunity to review the student’s proposed research and understanding of research methods and literature in the chosen field. If the general examination reveals deficiencies in any of these areas, the advisory committee may recommend remedial work, re-examination, or discontinuation of doctoral

9

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

study.

The general oral examination should be conducted immediately after the successful completion of the written examination and well before the final examination. At least one complete semester must intervene between the general oral and final examinations. The two examinations thus cannot be taken either in the same semester or in consecutive semesters. Successful completion of the oral examination requires unanimous support of the student’s advisory committee. If the general oral examination is failed, a re-examination may be given on recommendation of the committee and approval by the Dean of the Graduate School. Further examinations require exceptional circumstances and approval by the Graduate Council.

The student becomes a candidate for the degree on successful completion of the general examination. Guidelines for the Qualifying (General/Preliminary) Exam 

The qualifying exam serves as an assessment of the student’s developing research competencies and involves generating one of the following scholarly products: a critical review paper of publication quality, an empirical paper of publication quality (that is not the master’s thesis), or a grant proposal of sufficient quality to submit for funding. Students are encouraged to submit the scholarly product for publication in a peer-reviewed journal or to a funding agency; however, submission is not required as part of the qualifying exam process. The target journal or funding mechanism should be appropriate for a new professional in the field and should be specified for the committee by the student when the written proposal is submitted. More than one possible journal or funding mechanism may be specified.

The student’s choice of research topic and research product should be considered in close consultation with the major professor and should represent the student’s area of specialization. The scholarly product should also be consistent with the student’s career goals. For example, the product may address basic or applied research topics and may present implications for future research, practice, or policy. Grant proposals may be written for submission to a public or private research agency or to a service foundation.

After choosing a research topic and product in close consultation with the major professor, the student completes the written proposal, written exam, and oral exam, each described below.

Written proposal

The written proposal serves as a research proposal as well as an initial assessment of the student’s research competencies. It allows the advisory committee to identify strengths and areas that may require further development in the student’s doctoral training and to provide recommendations about the written exam.

The written proposal challenges the student to think and write independently, though reflective guidance from the major professor is encouraged. In other words, at this stage of the qualifying exam process, the student generates ideas (e.g., about theoretical frameworks, hypotheses, measures, planned analyses), and the major professor helps the student understand the strengths and limitations of the ideas. The major professor also provides feedback about the content and style of writing, but the student writes the entire proposal.

The written proposal introduces the topic and its significance, identifies the key theories and studies, and outlines the innovations and questions or hypotheses of the proposal or paper. This introductory section should span approximately the same number of pages as the Introduction section of a standard journal publication (i.e., 5-10

10

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

pages). In addition, unless the student proposes a critical review paper, the written proposal includes a standard Method or Approach section with appropriate sub-sections (e.g., participants, procedures, measures, and planned analyses).

Upon completion of the written proposal, the student arranges a meeting with the advisory committee and submits the written proposal to the advisory committee at least one week before the meeting. The student begins the meeting with a brief presentation of the proposal (e.g., 10-15 minutes), followed by a period during which the advisory committee asks questions and provides recommendations for the written exam. Within one week following the proposal meeting, the student should write a brief summary of the discussion with an emphasis on decisions or modifications to the original proposal and email the summary to the committee for their records and reference when reviewing the full written exam.

Written exam 

Preparing the written exam is an important learning and professional development experience. The written exam serves as a research product as well as a reflection of the student’s research competencies. Along with the oral exam, the written exam helps the advisory committee determine the student’s readiness for advancement to doctoral candidacy and the dissertation.

As compared to the written proposal, the major professor may play a more active role in the development of the written exam (e.g., flesh out the approved study plan, assist with troubleshooting analyses, and assist with processing discussion points), though it remains essential that the student produces the written exam in the lead role. Providing feedback through verbal or written comments is preferred over text editing and writing. Given the opportunity for increased input from the major professor, the benefit of feedback from the advisory committee and other colleagues, and additional time for independent learning and skill-building, the written exam should demonstrate progress in the student’s thinking and writing beyond the written proposal.

In addition to the scholarly product of choice, the written exam may be accompanied by a memo explaining any revisions to the full written product that deviate significantly from the approved proposal.

Upon completion of the written exam, the student submits the written exam to the advisory committee and tentatively schedules the oral exam with the advisory committee no less than three weeks from the submission of the written exam. The advisory committee reviews the written exam within one week. Possible outcomes of the committee review include: accept without revision or with minor revisions (which requires unanimous agreement) or major revisions required. The student is not permitted to proceed to the oral exam unless every member of the advisory committee accepts the written exam without revision or with minor revisions. Major revisions must be specified clearly by committee members through feedback about limitations as well as recommendations about how the exam could be improved. The student should prepare a revision for the committee following the assessment by even one advisory committee member that major revisions are required. Up to two rounds of revision and committee review are possible. Each revision should be accompanied by a detailed response memo that lists each committee member's questions and suggestions and an explanation about how they were addressed. If two rounds of revisions do not yield a product that is considered acceptable without revision or with minor revisions by all committee members, the written exam is considered failed and the student’s graduate program is concluded.

11

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

The General Oral Exam (Qualifying Exam Defense)

The oral defense of the written exam occurs within one month of receiving an acceptable outcome on the written exam. The Graduate School requires the student to apply for the oral exam (Request for the Report on the General Oral Examination) at least two weeks in advance of the oral exam. Thus, the oral exam should be scheduled with the advisory committee no less than three weeks from the submission of the written exam to the advisory committee, allowing sufficient time for the student to submit the application for the oral exam to the Graduate School (if the written exam is accepted by the advisory committee) or to reschedule the oral exam with the advisory committee (if the advisory committee requires revisions to the written exam before proceeding to the oral exam). The time and place of the oral exam, as well as the members of the advisory/examination committee, are identified on the application form. The application is available at the following website: http://graduate.auburn.edu/current-students/forms-directory/.

When the time and location of the defense is finalized (i.e., scheduled through the Graduate School), the major professor should announce the defense to HDFS faculty and graduate students. Students preparing for the oral exam are strongly encouraged to consult with their major professor, their committee members, or any other relevant resource.

The general oral exam has two phases. First is an approximately 45-minute, professional-level presentation of the written exam to the HDFS faculty and graduate student body. This presentation is followed by a question-and-answer period during which the student fields questions from the audience about the presentation. Approximately one hour should be allocated for the presentation and Q-A session. The audience is then dismissed and the second part of the exam proceeds. (Note, however, that Graduate School policy authorizes any graduate faculty member to attend the second part of the defense.) Although the second part of the oral exam typically continues to address the written exam, any aspect of the student’s specialization or any area of the Ph.D. core or program is legitimate examination material. The purpose of the oral examination is not simply to assess the student’s knowledge base, but also to test the student’s capacity to think extemporaneously.

At the end of the meeting, the student is typically asked to leave the room for a few minutes while the committee discusses the student’s performance and considers their votes. The votes are ‘pass’ or ‘fail’. To pass the exam, the committee vote must support passing unanimously. Committee members register their vote through the Graduate School’s online form, which is automatically delivered to committee members around the time of the exam. Although there is no official notice, students who have passed the general oral examination can change their designation from ‘doctoral student’ to ‘doctoral candidate’.

In the event of a failure of the general oral exam, students are allowed one retake. Before a failed oral exam ends, the committee’s expectations for the student’s performance in a retake and their recommendation for the timeframe should be clearly specified. One (and only one) retake of the oral exam is permitted, and the student should have a minimum of four weeks to prepare for it. When the time and place for the retake are defined, the student must reapply to the Graduate School for the exam. Only the student and committee reconvene for the retake. The committee should specify at the end of the first oral exam whether a presentation will be part of the second oral defense.

A second failure of the oral exam results in dismissal from the program.

12

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

DISSERTATION AND PROPOSAL GUIDE

Timing of the Dissertation Proposal Meeting 

You must pass the General Oral Examination before the dissertation proposal meeting.

Guidelines for Working with Your Committee 

The HDFS Graduate Advisory Committee has developed the following guidelines to help you in the preparation of your proposal and dissertation.

The purpose of a dissertation is to help you further develop skills in all aspects of the research process, which may include conceptualizing a research question, developing a research design, collecting data, coding and analyzing data, and interpreting results. The dissertation provides a structure through which these skills may be improved. Your committee, therefore, will expect you to demonstrate skill in each of these areas. It is the prerogative of the committee to ask questions regarding any or all of these skills.

You should seek advice from each member of your advisory committee when preparing and writing your dissertation proposal. Each committee member should receive a draft of your proposal at least one week before the meeting with your doctoral advisory committee at which you discuss and defend your proposal. Different committees may need more time, and it is their prerogative to request it.

As you proceed through the process of completing the dissertation, it is your responsibility to check in the Graduate School Bulletin for the graduate school deadlines (http://graduate.auburn.edu/calendar/). The format-check deadline is early in the semester. The deadline for submitting the final dissertation is usually around mid-semester for students intending to defend their dissertation and graduate in the same semester.

Requirements for a Proposal

A dissertation proposal is the plan of action that results in a final dissertation. As such, the Introduction and the Review of Literature will remain largely the same from proposal to dissertation. In most cases, the Method section will be changed only by conversion from future to past tense. Thus, the proposal is not a draft that will be polished up later. Students should understand, however, that many dissertations will confront unanticipated problems that will necessitate substantive changes to the proposal before the final draft of the dissertation is submitted.

You should consider the proposal defensible (i.e., ready for an examination meeting with the advisory committee) only when the Introduction, Literature Review, Method, References, and appropriate Appendices are finalized in the judgment of your major professor.

All research with human subjects must be approved in advance by the Institutional Review Board for the Use of Human Subjects in Research (IRB). The department head must approve protocols before they are forwarded to the IRB.

Elements of a Dissertation

The APA Publication Manual provides guidance for all of the sections that will appear in a dissertation with the exception of the Review of the Literature and Proposed Analysis sections (or see http://www.apastyle.org for the

13

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

same information). In the following paragraphs, we provide a brief description of elements of a typical dissertation and note some of the ways in which the AU Thesis guidelines differ from APA practice. Please use the APA Manual (http://www.apastyle.org) and the AU Guide to Preparation and Submission of Theses and Dissertations (http://graduate.auburn.edu/current-students/electronic-thesis-dissertation-guide/) throughout preparation of your document.

A dissertation normally consists of:

Prefatory Material: This material given in the opening pages of the dissertation is stipulated by the Auburn University Thesis and Dissertation Guide (certificate of approval, title page, table of contents, etc.)

Introduction: The Introduction is a brief statement of the question and an overview of the logic for the development of the question (including supporting literature). It may help to think of the Introduction as a section that could (with little editing) become the introduction to a published article. Specific hypotheses may be presented in the Introduction or may be held for the Literature Review section (a decision you should reach with your major professor).

Review of the Literature: The Review of the Literature is an extensive review of extant literature on all topics directly related to your dissertation question (e.g., theoretical and empirical background, a discussion of conceptual issues related to measures, any critique of previous literature and/or methodologies). The Literature Review should build logically to a statement of the specific research hypotheses or questions that will guide the analytic strategy for your study.

Method: The Method section is organized by subheadings that provide information about the following topics: Participants, Procedure, Measures, and Proposed Analyses. A thorough description of participants in your study is presented in the Participants section. This description includes all attributes of the sample that will be relevant to the analysis (e.g., gender and racial composition, age, socioeconomic status of subjects) but also relevant attributes of subjects that may not be included as variables in the analysis (e.g., region of country where subjects reside, number of subjects contacted to participate, number refusing). The Procedure details the steps undertaken to collect the data that you will analyze. The purpose of the Procedure is to permit a critical review of your data collection strategy and to provide enough information for others to replicate your study. The Measures section includes information about how each variable is measured. Estimates of reliability and validity for all measures are important pieces of information and should be included for each variable where relevant. Where possible, these estimates should be based in the data collected for your study. Furthermore, since many studies use previously existing measures for some or all variables in the analysis, published reports of the reliability and validity of these measures based in previous research is also legitimate and important information. The Proposed Analysis specifies each step in the analysis that will be used to answer the questions or hypotheses of the study. The presentation of these analyses, therefore, should link directly to the research questions or hypotheses.

Results: At a minimum, the Results section is a presentation of the outcome of all the proposed analyses. It may be organized (a) by research questions or hypotheses, (b) by issue or content area, or (c) some other logical criterion. Sometimes the planned analyses raise other interesting questions, and thus, the Results section may contain analyses beyond those originally planned.

Discussion: The Discussion is a brief presentation of each finding or group of findings together with explanations and alternative explanations of their meaning. The Discussion should be related to the

14

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

Introduction by connecting the findings with the driving issues that originally sparked the project. Furthermore, implications, limitations, and caveats of the results should be considered.

References: Students should use APA guidelines for listing references.

Footnotes: According to the Graduate School, footnotes may be placed at the bottom of the relevant page or the end of the section in which the footnote appears.

Tables and Figures: Tables and figures are integrated into the text near their first citation for AU dissertations and theses. See AU Dissertation Guide and APA for guidance on Table and Figure construction.

Appendices: Any pertinent material (e.g., informed consent letter, copies of questionnaires or surveys, interview protocol, observational coding forms, additional analyses not presented in Results) should be contained within separate appendices identified alphabetically at the end of the thesis/dissertation.

Multi-Study Dissertation Option. A multi-study (multi-publication) dissertation is an alternative to the single-study dissertation. In addition to the prefatory material described above and standard-length sections of a scientific article (Abstract, Introduction, Method, Results, Discussion, References, Tables and Figures), the multi-study dissertation requires General Introduction and General Discussion sections that situate the dissertation studies within an overarching topic area and explain how the studies separately and jointly advance knowledge on the overarching topic. The General Introduction and General Discussion sections span approximately the same number of pages as standard Introduction or Discussion sections of scientific articles (i.e., 5-10 pages). Typical journal space guidelines limit the extensiveness (breadth) of the literature reviews in the multi-study dissertation, but this option raises the challenge of integrating the dissertation studies. A paper that is published between the dissertation proposal and the dissertation defense is permissible as part of the multi-study dissertation as long as the paper is student-led and the research plan for the published paper was proposed and approved as part of the dissertation proposal process.

The dissertation may be accompanied by a memo explaining any revisions that deviate significantly from the proposal.

Final Examination (Dissertation Defense)

After a draft of your dissertation has been completed and approved by your advisory committee, it is submitted to the Graduate School. The representative of the Graduate School (the university reader) then reviews the dissertation. This review process adds weeks to the timetable for dissertation completion and should be accounted for as students plan their final semester. (It is common, but not required, for the university reader to be recruited by the student and/or major professor well in advance of the completion of the dissertation. Identifying the university reader in advance can reduce the time between your committee’s acknowledgment of the readiness of your dissertation to be defended and your actual defense.) When the university reader has approved the dissertation, you should apply for the final examination at least one week prior to the planned meeting.

Like the oral defense of your general written exam, your final examination will involve a public presentation of your dissertation followed by a meeting of your committee (although University policy dictates that any graduate faculty member may attend); therefore, select a time when all members of your examination committee can allocate a 2.5-hour block. When the time and location of the defense is scheduled, your major professor will announce the meeting to HDFS faculty and graduate students. The exam will begin with a professional-level

15

Revised 10/2018

 

Procedural Guide for Ph.D. Students Human Development and Family Studies

Auburn University

presentation of the dissertation to the HDFS faculty and graduate student body. This presentation (approximately 50 minutes) is followed by a question and answer period (approximately 10 minutes) during which you will field questions from the audience about your study. The audience is then dismissed and the second part of the exam proceeds. This second part of the exam usually focuses on your dissertation study, but any aspect of your major or minor area is legitimate examination material.

The representative of the Graduate School (the university reader) also attends the defense and participates, but does not vote on your performance or sign your dissertation approval form. Only oral examinations receiving unanimous approval of the committee are successful.

If the final examination is failed, a re-examination may be given on recommendation of the advisory committee and approval by the Dean of the Graduate School.

In addition to successful completion of all examinations, the oral and written approval forms must be completed, and the final dissertation must be uploaded (i.e., published) to the Auburn University Electronic Thesis and Dissertation (AUETD) website before the degree is conferred.

 

16

Revised 10/2018

 

Ulasan Elemen Reka Bentuk Antara Muka bagi Meningkatkan

Keterlibatan Pelajar terhadap E-Pembelajaran di Institusi Pengajian

(Reviews on Elements of Interface Design to Improve Student Engagement towards E-Learning in

Educational Institutes)

Hazwani Nordin & Dalbir Singh

Universiti Kebangsaan Malaysia (UKM)

Fakulti Teknologi dan Sains Maklumat,

Universiti Kebangsaan Malaysia, 43600 UKM Bangi, Selangor Darul Ehsan

No. Telefon: 012-9375730, No. Faks: 03-89256732

waninordin@gmail.com, dalbir@ukm.edu.my

Abstrak

E-pembelajaran telah lama diimplementasikan dalam sistem pendidikan di Malaysia untuk meningkatkan kualiti pengajaran dan pembelajaran. Perubahan sistem pendidikan ini membolehkan komunikasi secara aktif dikalangan pelajar dan tenaga pengajar. Namun, keterlibatan dan komitmen pelajar terhadap epembelajaran masih lagi ditahap yang lemah. Pelajar lebih menerima kaedah pengajaran konvensional dan keberatan untuk menggunakan e-pembelajaran. Reka bentuk antara muka memainkan peranan penting dalam menggalakkan keterlibatan dan komitmen pelajar terhadap e-pembelajaran. Reka bentuk antara muka yang lemah merupakan penyebab ketidakselesaan pelajar untuk terus menggunakan epembelajaran. Reka bentuk antara muka e-pembelajaran yang tidak memenuhi kehendak pelajar menyebabkan mereka tidak mengambil bahagian dalam e-pembelajaran walaupun ianya boleh membantu dalam pelajaran mereka. Oleh itu, kertas ulasan ini meneliti isu berkaitan dengan elemen reka bentuk antara muka yang sentiasa berubah mengikut keperluan pengguna. Isu keterlibatan pelajar turut dibincangkan dalam kertas ulasan ini bagi memahami maksud dan situasi keterlibatan pelajar yang berbeza mengikut pelbagai tempat dan keadaan. Dapatan ulasan ini boleh digunakan dalam pembangunan antara muka e-pembelajaran yang boleh meningkatkan keterlibatan dan komitmen pelajar semasa menggunakan e-pembelajaran. Akhir sekali, hasil ulasan ini diharapkan dapat menyumbang kepada peningkatan interaksi antara pengajar dan pelajar serta turut meningkatkan kualiti pengajaran dan pembelajaran.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 89

 

Abstract

E-learning has been implemented in the Malaysian educational system to improve the quality of teaching and learning. The changes in the educational system has allowed active communication among students and lecturers. However, student engagement and commitment towards e-learning is still poor. Students are more accepting of conventional teaching method and reluctance to use e-learning. Interface design plays an important role to encourage student engagement and commitment towards e-learning. Poor interface design causes discomfort among students to continue in using e-learning. E-learning interface design that does not meet the student’s needs may hinder student’s participation toward e-learning although it can help them in their studies. Therefore, this article review the issues related to interface design elements that changes based on the user needs. Student engagement issues are also discussed in this review article to understand the meaning and different student engagement situations which changes based on various places and environments. The findings from this review could be used in the development of e-learning interface design that would increase the student’s engagement while using e-learning. Lastly, the result of this review is hoped to contribute in enhancing interaction between student and lecturer while improving the quality of teaching and learning.

1. PENDAHULUAN

Perkembangan teknologi maklumat dan komunikasi pada hari ini memberi peluang kepada pelajar serta tenaga pengajar untuk menyampaikan aktiviti pengajaran dan pembelajaran interaktif melalui e-pembelajaran. Penggunaan e-pembelajaran dalam sistem pendidikan hari ini dapat membantu pelajar untuk memahami pengetahuan yang kompleks dengan lebih mudah dan membolehkan pelajar mencapai pembelajaran yang bermakna (Katuk, 2013) serta membolehkan pengajaran dan pembelajaran dijalankan walaupun setiap pelajar dipisahkan oleh tempat yang berlainan dan pada waktu yang berbeza (Sithole, B. D., & Onyari, 2012). Ini menyebabkan banyak institusi pengajian telah mengimplementasikan e-pembelajaran bagi menukar strategi pengajaran dan pembelajaran daripada pembelajaran dalam bilik kuliah semata-mata kepada pembelajaran bersepadu dan secara dalam talian sepenuhnya. Namun begitu, penerapan teknologi maklumat dan komunikasi dalam sistem pendidikan masa kini menyebabkan ramai pengguna mahukan reka bentuk antara muka yang berkualiti tinggi ketika berinteraksi dengan sistem e-pembelajaran. Reka bentuk antara muka merupakan teras dan komponen bersepadu bagi keseluruhan sistem e-pembelajaran (Reyna, 2009). Ini kerana reka bentuk antara muka e-pembelajaran perlu ditentukan berdasarkan bagaimana pelajar

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 90

 

ingin belajar dan tugasan yang pelajar perlukan untuk menjalankan e-pembelajaran (Guralnick 2006; Reyna 2009).

Reka bentuk antara muka yang baik dapat meningkatkan keterlibatan pelajar terhadap e-pembelajaran (Zaharias 2008; Wang et al. 2010; Faghih et al. 2013). Ini kerana reka bentuk antara muka memainkan peranan penting dalam meningkatkan rupa dan rasa (look and feel) e-pembelajaran dan ia boleh mempengaruhi cara pelajar berinteraksi dengan e-pembelajaran (Reyna, 2009). Rupa dan rasa reka bentuk antara muka e-pembelajaran boleh mempengaruhi pelajar sama ada ingin terus berinteraksi dengan e-pembelajaran atau sebaliknya. Komunikasi yang efektif dapat meningkatkan keterlibatan dan motivasi pelajar serta menjadikan interaksi pelajar dengan e-pembelajaran menjadi lebih menarik (Wang et al., 2010). Selain itu, keterlibatan pelajar juga boleh meningkatkan kepuasan pelajar ketika berinteraksi dengan e-pembelajaran dan seterusnya membantu pelajar meningkatkan pencapaian pelajar dalam bidang akademik (Essam & Al-Ammary, 2013).

Penglibatan pelajar secara berterusan dalam e-pembelajaran penting kerana ia menjurus kepada pembelajaran yang lebih baik. Kebanyakan institusi pengajian telah mengimplementasikan e-pembelajaran tersendiri bagi memenuhi keperluan pengajaran dan pembelajaran pelajar. Selain daripada e-pembelajaran yang dibangunkan oleh pihak institusi, terdapat e-pembelajaran yang ditawarkan secara percuma kepada pelajar seperti ‘google classroom’ dan ‘open learning’. Tenaga pengajar juga memainkan peranan dalam memastikan keterlibatan pelajar dalam e-pembelajaran dengan memberi markah kepada pelajar yang menyertai perbincangan di forum, dan memberikan latihan secara berterusan secara melalui e-pembelajaran. Namun begitu, aspek reka bentuk antara muka e-pembelajaran tidak boleh diabaikan kerana antara muka merupakan titik permulaan anggapan pelajar terhadap e-pembelajaran di mana kejayaan atau kegagalan e-pembelajaran bergantung kepada reka bentuk antara muka (Faghih, Azadehfar & Katebi 2013). Tambahan pula, reka bentuk antara muka e-pembelajaran adalah sangat kritikal kerana keberkesanan pembelajaran dan reka bentuk antara muka mempunyai perkaitan yang sangat ketara (Guralnick, 2006). Oleh itu, kertas ulasan ini menumpukan kepada reka bentuk antara muka untuk meningkatkan keterlibataan pelajar dalam e-pembelajaran di institusi pengajian.

2. KETERLIBATAN PELAJAR TERHADAP E-PEMBELAJARAN

Istilah keterlibatan pelajar (student engagement) semakin banyak diguna pakai dalam bidang pendidikan sejak beberapa tahun kebelakangan ini. Keterlibatan pelajar dilihat sebagai petunjuk kepada kejayaan pengajaran dan pembelajaran pelajar dalam e-pembelajaran dan ia sangat dihargai sebagai sebahagian daripada aktiviti penambahbaikan dalam bidang akademik. Antara faktor utama keterlibatan pelajar penting dalam e-pembelajaran adalah disebabkan oleh kadar keciciran pelajar yang tinggi dalam e-pembelajaran di universiti (Angelino, Williams & Natvig 2007; Backs 2012). Selain itu, pendidik turut memberi penekanan bagi meningkatkan keterlibatan pelajar dalam e 

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 91

 

pembelajaran sebagai jalan untuk menangani pencapaian yang rendah, kebosanan ketika sesi pembelajaran dan pengasingan pelajar (Fredricks et al. 2011).

Beberapa penyelidik terdahulu telah mengkaji mengenai keterlibatan pelajar dalam e-pembelajaran dan telah mengemukakan pendapat dan definisi yang berbeza. Oleh itu, memahami takrifan keterlibatan pelajar dengan lebih mendalam adalah penting walaupun ramai penyelidik mengemukakan definisi yang berlainan namun ia mampu membantu kajian ini bagi memahami maksud keterlibatan pelajar dengan lebih mendalam.

3. DEFINISI KETERLIBATAN PELAJAR

Istilah keterlibatan pelajar pertama kalinya muncul pada akhir tahun 1980-an. Ketika itu, kebanyakan penyelidik cenderung untuk melihat sebab musabab keterlibatan pelajar melalui satu set faktor di mana demografi dan latar belakang sosial dikaitkan dengan pelajar seperti keadaan keluarga dan pengaruh rakan sebaya. Namun, konsep keterlibatan pelajar mula beralih makna hasil daripada peningkatan perhatian dan pengaruh sekolah terutamanya hubungan antara institusi pendidikan dengan pengalaman keterlibatan pelajar (Dunleavy & Milton, 2009). Keterlibatan pelajar mempunyai pelbagai maksud kepada ramai penyelidik dalam kajian yang berbeza serta ia boleh ditakrifkan secara berbeza mengikut tempat. Sebagai contoh, konsep keterlibatan pada sesebuah sekolah boleh diperhatikan seperti menghadiri kelas, perubahan kerja dalam jangka masa yang diberikan dan mengikut peraturan sekolah. Sementara itu, di sekolah lain konsep keterlibatan difahami dari segi dalaman seperti semangat, rasa ingin tahu, keyakinan, motivasi dan faedah (Ching, 2012).

Istilah keterlibatan memberi maksud lebih daripada penglibatan di mana ia memerlukan perasaan dan membuat pertimbangan serta aktiviti (Trowler, 2010). Menurut Trowler (2010) penglibatan tanpa perasaan diibaratkan seperti berlakon tanpa perasaan dan hanya terlibat untuk patuh kepada skrip. Welch, & Bonnan-White (2012) menyatakan bahawa keterlibatan pelajar bermaksud masa dan usaha pelajar yang tertumpu kepada aktiviti-aktiviti yang secara empirikal dikaitkan dengan hasil yang dikehendaki oleh institusi dan apa yang dilakukan oleh institusi untuk mendorong pelajar menyertai aktiviti-aktiviti ini seperti belajar, berinteraksi dengan pengajar dan rakan sebaya, melibatkan diri dalam aktiviti kampus dan memenuhi jangkaan akademik yang tinggi.

Dunleavy, & Milton (2009) menyatakan bahawa definisi keterlibatan pelajar adalah berbeza dan bergantung kepada pendekatan yang digunakan. Beliau menyatakan bahawa keterlibatan pelajar melibatkan dua dimensi iaitu keterlibatan sosial dan keterlibatan akademik. Beliau menakrifkan keterlibatan sosial sebagai gabungan semangat kekitaan di kalangan pelajar di institusi pengajian, penerimaan mereka terhadap matlamat institusi, perasaan yang berhubung dan diterima rakan-rakan dan pengalaman berhubungan dengan orang dewasa yang menunjukkan minat mereka sebagai individu. Pada awal tahun 1990-an, Newmann, Wehlage & Lamborn (1992) menakrifkan keterlibatan akademik sebagai pelaburan psikologi pelajar dan usaha yang menjurus kepada pembelajaran,

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 92

 

pemahaman atau penguasaan pengetahuan dan kemahiran bagi menggalakkan prestasi akademik. Beliau turut meneruskan kajian bagi memahami apa yang mendorong pelajar untuk berjaya dalam akademik seperti menyiapkan latihan dan kerja sekolah, menghadiri kelas serta mengambil bahagian di dalam bilik darjah (Dunleavy, & Milton 2009; Ching 2012). Selain itu, ramai penyelidik lain yang berpendapat bahawa takrifan keterlibatan pelajar adalah bergantung kepada dimensi yang berbeza seperti tingkah laku (behavioural), emosi (emotional) dan kognitif (cognitive).

Terdapat banyak definisi mengenai keterlibatan pelajar namun tiada definisi yang khusus mengenainya. Definisi yang dikemukakan oleh penyelidik lepas adalah berdasarkan kepada kajian yang dijalankan oleh mereka pada masa itu di mana ia dirasakan bersesuaian dengan situasi kajian mereka pada masa tersebut. Walaupun definisi yang dikemukakan berbeza antara satu sama lain tetapi perbezaan tersebut tidak terlalu ketara dan hampir kesemua definisi mempunyai matlamat yang sama iaitu menjurus kepada penambahbaikan dalam pendidikan pelajar. Oleh itu, definisi keterlibatan pelajar dalam kajian ini menjurus kepada menjalankan atau melakukan aktiviti yang disertakan dengan perasaan dalaman yang tinggi yang mana ia boleh mendorong pelajar ke arah yang lebih baik. Oleh kerana kebanyakan definisi yang dikemukakan melibatkan tugasan atau aktiviti maka definisi yang sedia ada telah menepati kehendak kajian.

4. FAKTOR YANG MEMPENGARUHI KETERLIBATAN PELAJAR

Kejayaan dan kegagalan e-pembelajaran bergantung kepada keterlibatan pelajar. Berdasarkan kajian lepas beberapa faktor yang boleh mempengaruhi keterlibatan pelajar dalam e-pembelajaran telah dikenal pasti. Antara faktor tersebut adalah seperti berikut:

a) Kandungan kursus

Di kebanyakan institusi pengajian, kandungan kursus disediakan oleh pengajar. Ladbrook, & Parr (2015) menyatakan bahawa kandungan kursus menyumbang kepada penglibatan pelajar secara aktif dalam e-pembelajaran. Beliau menyarankan agar kandungan kursus diberi perhatian dari segi reka bentuk kursus kerana ia mampu meningkatkan motivasi dan penglibatan pelajar dalam e-pembelajaran. Namun, ramai pelajar yang menarik diri daripada kursus yang diikuti dalam e-pembelajaran berbanding dengan pembelajaran dalam bilik kuliah (Angelino et al. 2007; Ali, & Smith 2015; Boton et al. 2015). Menurut Boton et al. (2015), sebab utama pelajar menarik diri daripada kursus berkenaan kerana kandungan kursus yang ditawarkan tidak sesuai dengan keperluan pelajar. Selain itu, tugasan yang diberikan tidak mencabar menyebabkan pelajar lebih cepat bosan (Dixson, 2010) dan tidak merasa cabaran ketika menyiapkan tugasan tersebut. Oleh itu, mereka bentuk kursus adalah tanggungjawab pengajar dalam meningkatkan keterlibatan pelajar dan motivasi pelajar untuk menggunakan e-pembelajaran sebagai medium pembelajaran mereka.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 93

 

b) Alatan yang ditawarkan

Alatan (tools) yang ditawarkan dalam e-pembelajaran turut mempengaruhi motivasi dan keterlibatan pelajar (Delialioǧlu 2012; Ladbrook, & Parr 2015). Alatan yang dimaksudkan seperti audio, visual dan teks adalah penting dalam membolehkan sesuatu komunikasi berlaku dalam e-pembelajaran. Audio dan visual yang berkualiti rendah boleh mengganggu komunikasi pelajar ketika sesi perbincangan dan secara tidak langsung boleh mempengaruhi motivasi dan keterlibatan pelajar untuk terus menggunakannya. Selain itu, teks yang digunakan mesti bersesuaian dari segi jenis tulisan dan saiz teks dengan cita rasa pengguna seperti pengguna yang berusia, remaja mahupun pelajar sekolah.

c) Budaya Pelajar

Budaya sering dikaitkan dengan keterlibatan pelajar dalam e-pembelajaran. Setiap pelajar terdiri daripada latar belakang budaya yang berbeza (Chakrabarty & Mohamed, 2013) namun perbezaan budaya ini tidak bermaksud malas atau tidak berusaha dalam pembelajaran tetapi ia bermaksud setiap pelajar mempunyai keutamaan yang berbeza dalam persekitaran pembelajaran mereka (Parsons & Taylor, 2011). Budaya terdiri daripada kepercayaan, andaian, pengetahuan, nilai-nilai atau amalan yang dikongsi dalam masyarakat (Benaida, 2014; Mcfarlane, 2011). Sebagai contoh, terdapat sesetengah budaya yang mana pelajar lebih menerima pembelajaran konvensional berbanding pembelajaran menggunakan e-pembelajaran (Zhang, 2013). Oleh itu, faktor budaya tidak boleh dipandang ringan dalam usaha meningkatkan keterlibatan pelajar menggunakan e-pembelajaran di institusi pengajian.

d) Reka bentuk antara muka

Reka bentuk antara muka pengguna tidak boleh dipandang remeh bagi meningkatkan keterlibatan pelajar dalam e-pembelajaran. Ini kerana reka bentuk antara muka pengguna adalah bahagian yang paling penting dan kritikal dalam e-pembelajaran kerana keberkesanan pembelajaran dan reka bentuk antara muka mempunyai perkaitan dan saling bergantung antara satu sama lain (Guralnick, 2006). Pelajar pada hari ini yang sentiasa terdedah dengan teknologi terkini berasa cepat bosan ketika menggunakan e-pembelajaran yang mempunyai antara muka yang tidak menarik (Parsons & Taylor, 2011) seperti antara muka yang mempunyai banyak teks, visual yang membosankan serta kesukaran mencari maklumat. Reka bentuk antara muka e-pembelajaran hendaklah ditentukan oleh bagaimana pelajar belajar dan bagaimana mereka menggunakan e-pembelajaran untuk melaksanakan sebarang tugasan (Reyna, 2009). Oleh itu, reka bentuk susun atur memainkan peranan penting dalam rupa dan rasa e-pembelajaran dan boleh mempengaruhi cara pelajar belajar dan berinteraksi menggunakan e-pembelajaran.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 94

 

5. DEFINISI REKA BENTUK ANTARA MUKA

Aplikasi teknologi hari ini adalah berpaksi kepada pembangunan antara muka kerana reka bentuk antara muka yang baik dapat melancarkan perjalanan tugasan tetapi berurusan dengan antara muka yang rumit dan mengelirukan boleh menyebabkan pengguna mengelak daripada menggunakan sistem tersebut (Barzegar, Shojafar, & Keyvanpour, 2009). Reka bentuk antara muka hendaklah dibangunkan dengan teliti dan menepati kehendak dan piawaian pengguna agar pengguna dapat berinteraksi dengan aplikasi sistem semaksima yang mungkin.

Pelbagai takrifan antara muka yang telah dikemukakan oleh penyelidik lepas. Galitz (2007) menyatakan antara muka adalah sebahagian daripada komputer dan perisiannya yang membolehkan pengguna mendengar, melihat, menyentuh, bercakap dan memahami secara langsung atau tidak langsung. Beliau mengkategorikan antara muka kepada dua komponen iaitu input dan output. Input adalah bagaimana pengguna berkomunikasi dengan komputer menggunakan komponen input seperti papan kekunci, tetikus, skrin sentuh dan suara. Output pula adalah bagaimana komputer menyampaikan hasil dan keperluan kepada pengguna.

Adnan, Ali & Ahmad (2015) menakrifkan antara muka sebagai satu alat atau program yang membolehkan pengguna berkomunikasi dengan sistem komputer. Antara muka didefinisikan sebagai medium perantaraan bagi komunikasi antara manusia dengan sistem komputer. Antara muka digunakan untuk menghantar maklumat daripada manusia untuk sistem yang dikenali sebagai interaksi. Dalam teknologi maklumat dan komunikasi (ICT), reka bentuk antara muka merujuk kepada permukaan skrin sesuatu alat seperti komputer, telefon pintar dan tablet yang membolehkan pengguna untuk mengakses dan memahami maklumat yang dikehendaki (Kamaruddin, Park & Nam 2012). Menurut Kamaruddin (2014), reka bentuk antara muka tidak hanya berkaitan dengan nilai-nilai estetika, rupa dan rasa yang dibentuk oleh elemen-elemen reka bentuk seperti warna, teks, imej dan susun atur tetapi merupakan medium komunikasi yang menjadi perantara antara pengguna dan sistem.

Pelbagai definisi antara muka telah dikemukakan oleh penyelidik lepas, namun kebanyakan kajian menyatakan bahawa antara muka adalah medium komunikasi antara manusia dan sistem yang diterjemahkan dalam bentuk permukaan skrin. Oleh itu, dalam kajian ini antara muka merujuk kepada permukaan skrin komputer bagi memudahkan interaksi dengan manusia.

6. KAEDAH DAN PROSEDUR PEMILIHAN ELEMEN ANTARA MUKA

Kajian ini adalah untuk menentukan elemen antara muka yang boleh mempengaruhi keterlibatan pelajar dalam e-pembelajaran. Jadual 1 di bawah menunjukkan kaedah yang digunakan dalam pemilihan elemen antara muka e-pembelajaran.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 95

 

Jadual 1 Pemilihan Prosedur

Prosedur Penerangan mengenai prosedur

Pemilihan artikel Mencari artikel yang berkaitan dengan reka bentuk antara muka.

Pemilihan elemen antara muka Memilih elemen antara muka yang sering digunakan dalam e-pembelajaran.

Perincikan elemen antara muka Memperincikan ciri-ciri setiap elemen antara muka yang telah dipilih.

6.1 Pemilihan Artikel

Bagi menentukan elemen antara muka yang boleh mempengaruhi keterlibatan pelajar, maklumat yang diperolehi daripada artikel, prosiding dan buku rujukan dikaji dengan lebih mendalam. Antara kriteria-kriteria yang digunakan ketika proses pemilihan artikel adalah seperti berikut:

• Artikel yang diterbitkan dari tahun 2000 hingga 2016 diperolehi daripada pangkalan data dalam talian. Jangka masa lebih dari satu dekad dipilih adalah untuk melihat sama ada terdapat perubahan elemen-elemen antara muka e-pembelajaran selama sedekad ini.

• Elemen reka bentuk antara muka: Artikel-artikel yang dikaji memfokuskan kepada elemen reka bentuk antara muka sahaja. Elemen antara muka yang boleh mempengaruhi pelajar untuk terus terlibat dalam menggunakan e-pembelajaran.

• Tempatan dan antarabangsa: Elemen antara muka e-pembelajaran dikaji secara meluas iaitu melibatkan artikel tempatan dan luar negara. Ini bertujuan untuk mengenal pasti persamaan dan perbezaan pemilihan elemen-elemen antara muka bagi dalam dan luar negara.

• Kata kunci: beberapa kata kunci digunakan semasa proses pencarian artikel bagi mendapatkan artikel yang berkaitan sahaja. Antara kata kunci yang digunakan adalah elemen antara muka, antara muka pengguna, keterlibatan pelajar, corak antara muka dan pilihan antara muka.

6.2 Pemilihan Elemen Antara Muka

Semua artikel yang berkaitan dikaji dan dianalisis dengan teliti bagi mengenal pasti apa fungsi setiap elemen dan ciri-ciri antara muka yang mendorong kepada kejayaan e-pembelajaran. Antara langkah-langkah yang diambil dalam memilih elemen antara muka adalah seperti berikut:

• Kenal pasti elemen antara muka: Menyenaraikan semua elemen antara muka yang dikaji oleh penyelidik lepas. Setiap fungsi elemen tersebut juga dikenal pasti.

• Kepentingan elemen antara muka: Setiap elemen antara muka dikenal pasti mesti sesuai kepada pengguna sasaran jika ia diimplemenkan ke dalam antara muka e-pembelajaran.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 96

 

• Kurangkan elemen yang tidak penting: Penggunaan elemen antara muka yang terlalu banyak boleh menyebabkan antara muka nampak terlalu sesak sehingga mengaburi objektif utama e-pembelajaran dibangunkan.

6.3 Perincian Elemen Antara Muka

Elemen antara muka yang telah dipilih mesti dikaji dengan lebih mendalam bagi memastikan ia memenuhi kehendak dan keperluan pengguna. Setiap elemen juga perlu dikenal pasti kefungsiannya. Perincian elemen antara muka juga dilaksanakan dengan mengadaptasikan ke dalam prinsip reka bentuk antara muka bagi membangunkan antara muka yang baik.

7. ELEMEN ANTARA MUKA MEMPENGARUHI KETERLIBATAN PELAJAR DALAM E-PEMBELAJARAN

Elemen antara muka penting dalam mereka bentuk antara muka. Elemen antara muka ditakrifkan sebagai sesuatu yang boleh dilihat (seperti gambar, carta atau animasi) yang digunakan untuk menjadikan antara muka tersebut lebih menarik dan mudah untuk difahami (Adnan et al., 2015). Masalah utama dalam e-pembelajaran adalah apabila pensyarah dan pembangun aplikasi lebih mengutamakan reka bentuk kursus dan tidak memberi perhatian kepada reka bentuk antara muka e-pembelajaran (Thomson, 2014). Akibatnya, pelajar hanya dapat melihat terlalu banyak teks, imej, pautan, grafik dan kurang keharmonian dalam reka bentuk antara muka e-pembelajaran. Ini menyebabkan pelajar sukar untuk memberi tumpuan, memahami dan belajar dengan cara yang diingin. Oleh itu, pelajar menjadi kurang berminat dengan pembelajaran menggunakan e-pembelajaran.

Elemen antara muka boleh memberi kesan ke atas rupa dan rasa reka bentuk antara muka e-pembelajaran dan sekaligus boleh mempengaruhi keterlibatan pelajar terhadap e-pembelajaran. Banyak elemen reka bentuk antara muka yang dikemukakan oleh penyelidik lepas dikategorikan kepada warna (Adnan et al., 2015; Arockiam & Selvaraj, 2013; Bakar & Long, 2013; Faghih et al., 2013; Khanum, Fatima, & Chaurasia, 2012; Reyna, 2009, 2013; Tomczyk, 2009), imej (Adnan et al., 2015; Faghih et al., 2013; Khanum et al., 2012), teks (Adnan et al., 2015; Arockiam & Selvaraj, 2013; Bakar & Long, 2013; Khanum et al., 2012; Reyna, 2009, 2013), susun atur (Khanum et al., 2012; Reyna, 2013), navigasi (Reyna, 2009), animasi (Faghih et al., 2013), bentuk (Adnan et al., 2015) dan pautan (Khanum et al., 2012). Jadual 2 menunjukkan kajian yang dijalankan oleh penyelidik lepas dan kekerapan elemen reka bentuk antara muka yang digunakan dalam kajian terdahulu.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 97

 

Jadual 2 Elemen antara muka pada kajian lepas

Reyna (2009) X X X X

Tomczyk (2009) X X

Kamaruddin (2012) X X X X

Ramakrisnan, Jaafar, X X X

Razak (2012)

Bakar & Long (2013) X X

Faghih et al. (2013) X X

Shahzad, Hafizi, & X

Golamdin (2014)

Yaa & Adzobu X X (2014)

Adnan et al. (2015) X X X

Rajah 1 merupakan hasil daripada sorotan susastera yang dijalankan yang mana ramai penyelidik lepas berpendapat bahawa warna, tipografi, imej, susun atur dan navigasi memainkan peranan penting dalam reka bentuk antara bagi meningkatkan keterlibatan pelajar dalam e-pembelajaran. Kajian yang dijalankan oleh Reyna (2009) mengkategorikan elemen ikon, simbol dan banner sebagai imej. Maka, elemen ikon, simbol dan banner dikategori dalam elemen imej untuk kajian ini. Kertas ulasan ini hanya berkisarkan mengenai elemen antara muka yang memberi kesan ke atas keterlibatan pelajar menggunakan e-pembelajaran. Berikut adalah antara elemen yang telah dikenal pasti oleh penyelidik lepas yang dikupas dengan lebih mendalam bagi meningkatkan keterlibatan pelajar menggunakan e-pembelajaran di institusi pengajian.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 98

 

 

Rajah 1: Elemen Reka Bentuk Antara Muka E-Pembelajaran

a) Warna

Warna menjadi elemen yang penting dalam antara muka e-pembelajaran. Ini kerana warna berperanan dalam menarik minat pelajar dan menjana keinginan untuk terus menggunakan e-pembelajaran. Terdapat beberapa warna yang boleh meningkatkan tindak balas pelajar dan turut memberi kesan yang menenangkan kepada pengguna (Thomson, 2014). Menurut Müller (2014), warna bukan hanya sekadar warna tetapi boleh mewujudkan emosi, perasaan (Reyna, 2013) dan kenangan yang boleh mempengaruhi minda dan badan manusia melalui fizikal dan juga reaksi psikologi kepada warna tertentu. Kombinasi warna yang betul boleh menekankan organisasi skrin, memudahkan diskriminasi komponen skrin, menyerlahkan perbezaan dan dapat membuatkan paparan lebih menarik. Jika warna tidak digunakan dengan betul boleh menyebabkan keletihan visual, pening, sakit mata, sakit kepala, hilang tumpuan atau motivasi (Reyna, 2009) dan boleh merosakkan kebolehgunaan sistem (Galitz, 2007). Jadual 3 menunjukkan kajian yang dijalankan oleh Galitz (2007) mengenai kombinasi warna latar belakang dan latar depan yang sesuai.Berikut adalah senarai kombinasi warna latar belakang (background) dengan warna latar depan (foreground) yang sesuai seperti dalam Jadual 3.

Jadual 3 Contoh kombinasi warna yang sesuai

Warna Latar Belakang (Background) Warna Latar Depan (Foreground)

Hitam Cyan gelap Cyan terang

Kuning gelap Kuning terang

Putih Magenta terang Hijau terang

Biru Hijau gelap Hijau terang

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 99

 

Kuning gelap Cyan terang

Putih Kuning terang

Hijau Hitam Kuning terang

Biru gelap Putih

Merah Hijau terang Kuning terang

Cyan terang Putih

Kuning Hitam Merah gelap

Biru gelap

Putih Hitam

Biru gelap

Cyan Hitam Kuning terang

Biru gelap Putih

Magenta Hitam Kuning

Cyan terang Putih

Selain itu, penggunaan warna dalam e-pembelajaran turut bergantung kepada pengguna. Jika pengguna e-pembelajaran adalah kanak-kanak, maka warna terang adalah paling sesuai digunakan. Kajian yang dijalankan oleh Adnan et al. (2015) menyatakan bahawa warna biru adalah warna yang paling popular dikalangan kanak-kanak. Sementara itu, kajian yang dijalankan oleh Müller (2014) ke atas pelajar di universiti di Afrika Selatan menyatakan bahawa warna biru, merah, ungu dan hitam adalah warna pilihan pelajar di universiti tersebut.

Selain daripada warna latar belakang, penekanan warna grafik, teks, butang menu dan pautan juga harus diberi perhatian. Ini kerana kesilapan terhadap kombinasi warna yang sesuai boleh mengundang kepada penurunan penggunaan e-pembelajaran dikalangan pelajar. Tambahan pula, terdapat warna yang boleh mengganggu keterlibatan pelajar dan tidak sesuai untuk digunakan dalam e-pembelajaran (Thomson, 2014). Ini kerana ia bertentangan dengan budaya sesuatu masyarakat. Menurut Kim, & Kuljis (2010) warna mempunyai maksud yang berlainan bagi budaya yang berlainan. Sebagai contoh, di Jepun warna putih melambangkan kematian manakala di Egypt melambangkan kegembiraan. Oleh itu, penggunaan warna perlu dipilih dengan teliti dan cermat agar tidak menyinggung mana-mana budaya.

b) Imej, grafik, logo dan ikon

Elemen imej sering diperkatakan oleh penyelidik lepas sebagai salah satu elemen yang mampu mempengaruhi keterlibatan pelajar dalam e-pembelajaran. Penggunaan imej dalam antara muka e-pembelajaran dapat membantu pelajar memahami isi kandungan tanpa perlu membaca teks. Imej pada

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 100

 

antara muka terdiri daripada ikon, simbol dan gambar (Adnan et al., 2015). Setiap imej yang digunakan mempunyai maksud yang tersendiri yang difahami oleh para pengguna tanpa perlu membaca teks atau label. Penggunaan imej yang sesuai serta menarik boleh memikat lebih ramai pelajar untuk belajar menggunakan e-pembelajaran, memudahkan pembelajaran dan menguatkan ingatan (Adnan et al., 2015). Ini kerana, imej lebih mudah untuk dikenal pasti dan diingati berbanding dengan penggunaan teks semata-mata.

Walau bagaimanapun, penggunaan imej yang banyak juga boleh mempengaruhi minat pelajar untuk menggunakan e-pembelajaran. Imej yang banyak dalam e-pembelajaran boleh mengelirukan pelajar walaupun imej yang digunakan berkaitan dengan kursus yang diambil serta boleh menyebabkan pelajar hilang fokus terhadap apa yang ingin mereka pelajari daripada e-pembelajaran (Thomson 2014; Adnan et al. 2015). Selain itu, saiz imej juga harus diberi perhatian ketika membangunkan antara muka e-pembelajaran. Ini kerana saiz imej yang terlalu besar atau terlalu kecil boleh menjadi isu pelajar tidak mahu menggunakan e-pembelajaran. Namun, terdapat pelajar yang tidak terganggu dengan bilangan dan saiz imej dalam antara muka e-pembelajaran. Menurut kajian yang dijalankan oleh Kim, & Kuljis (2010), negara Korea Selatan lebih gemar jika antara muka mempunyai banyak imej berbanding dengan United Kingdom yang lebih gemar jika penggunaan imej adalah minima. Perbezaan ini dikaitkan dengan latar belakang budaya yang berbeza yang menyebabkan kehendak setiap pelajar adalah berbeza.

Bagi meningkatkan keterlibatan pelajar, imej yang diletakkan pada e-pembelajaran perlu diminimumkan dan berstruktur. Selain itu, imej pada antara muka hendaklah dalam keadaan bersusun bagi memudahkan pelajar mencari maklumat dan tidak sesat ketika melayari e-pembelajaran (Galitz, 2007). Menurut Thomson (2014) meminimakan penggunaan grafik selari dengan teks adalah salah satu cara untuk mengekalkan minat, motivasi dan keterlibatan pelajar terhadap e-pembelajaran. Oleh itu, ruang putih yang banyak pada antara muka e-pembelajaran dibenarkan bagi membolehkan pelajar lebih fokus kepada satu-satu tugasan.

c) Teks / Tipografi

Elemen tipografi penting dalam reka bentuk antara muka e-pembelajaran sebagai medium untuk menyampaikan maklumat. Ia berfungsi untuk menghuraikan visual seperti label untuk ikon yang mana menjadi pilihan apabila simbol atau ikon tidak dapat menyampaikan mesej yang ingin disampaikan (Faghih et al. 2013; Adnan et al. 2015). Tambahan pula, tipografi merangkumi jenis dan saiz teks yang mempunyai kesan ke atas kebolehgunaan sesuatu sistem (Khanum et al., 2012). Teks yang digunakan pada antara muka bukan hanya menulis kandungan tetapi dimanipulasikan sebagai pautan, butang dan animasi yang berperanan dalam navigasi, interaksi dan penyampaian e-pembelajaran (Adnan et al., 2015). Namun, penggunaan teks semata-mata dalam antara muka membuatkan antara muka menjadi kurang menarik (Zhang 2013; Faghih et al. 2013). Tambahan pula,

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 101

 

dalam membantu proses pembelajaran pelajar penyediaan antara muka berasaskan teks sahaja tidak mencukupi untuk menjadikan pembelajaran berkesan, cekap dan membuatkan pelajar terus melibatkan diri (Bart, 2012). Penggunaan teks yang terlalu banyak dalam antara muka boleh mengelirukan pelajar dan mereka tidak dapat membezakan apa yang relevan dengan pembelajaran mereka sama ada penting atau tidak (Thomson, 2014) kerana kekurangan penstrukturan teks boleh menjejaskan kepentingan minat pelajar dalam menggunakan e-pembelajaran. Tambahan pula, bentuk atau jenis fon dalam e-pembelajaran perlu dipilih dengan teliti. Fon yang digunakan hendaklah sesuai dengan tujuan e-pembelajaran diwujudkan dan tidak terlalu mencolok dan mewah (Thomson, 2014) sehingga mampu mengalih perhatian pelajar ketika proses pembelajaran berlangsung. Fon yang digunakan perlu membuatkan antara muka nampak estetik bagi meningkatkan keterlibatan pelajar dalam e-pembelajaran. Selain itu, penggunaan fon mestilah sesuai dengan peringkat umur pelajar sasaran kerana jenis fon bagi yang digemari bagi kanak-kanak, orang dewasa serta orang tua adalah tidak sama antara satu sama lain (Tomczyk, 2009).

Selain itu, penggunaan warna teks juga penting dalam menarik lebih ramai pelajar untuk menggunakan e-pembelajaran. Penggunaan warna teks dan warna latar belakang perlu kontra seperti warna latar belakang yang cerah dan warna teks yang gelap agar pengguna mudah untuk membaca teks. Selain itu, saiz teks juga perlu diberi perhatian kerana teks yang terlalu kecil atau besar boleh menjejaskan tumpuan pelajar ketika menggunakan e-pembelajaran. Oleh itu, pembangun antara muka e-pembelajaran perlu mengetahui kehendak dan keperluan pelajar dalam sesebuah antara muka e-pembelajaran agar pembelajaran dapat berjalan dengan sempurna dan lancar.

d) Susun atur (Layout)

Susun atur antara muka merupakan elemen yang paling penting kerana pengguna memberikan perhatian ketika berkomunikasi dengan e-pembelajaran. Susun atur ditakrifkan sebagai bagaimana elemen-elemen antara muka seperti imej, butang, pautan, logo, skim warna dan seni bina maklumat diagihkan pada halaman antara muka (Tomczyk 2009; Reyna 2013). Ramai penyelidik lepas yang menyatakan bahawa susun atur fizikal objek visual pada skrin memainkan peranan dalam persepsi estetika pengguna terhadap antara muka (Altaboli & Lin, 2011) dan susun atur antara muka penting untuk mendapatkan kepercayaan daripada pengguna ketika menggunakan e-pembelajaran (Alias, Zakariah, Ismail, & Aziz, 2012).

Susun atur yang konsisten amat penting dalam antara muka e-pembelajaran kerana ia melibatkan kredibiliti dan pengajaran serta pembelajaran secara professional. Susun atur e-pembelajaran yang konsisten mempunyai halaman dengan susun atur bentuk dan corak yang sama di semua halaman e-pembelajaran (Tomczyk 2009; Reyna 2013). Susun atur yang tidak konsisten memberi kesan terhadap navigasi pengguna (Lim, Ayesh, & Chee, 2013). Sebagai contoh, butang menu diletakkan di bahagian atas pada halaman pertama e-pembelajaran tetapi butang menu

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 102

 

diletakkan pada sebelah tepi pada halaman berikutnya. Ini menyebabkan pengguna menghabiskan banyak masa mencari butang yang sama pada halaman berlainan.

e) Navigasi

Navigasi adalah elemen penting yang perlu diambil kira ketika membangunkan antara muka e-pembelajaran. Navigasi penting bagi membolehkan pergerakan yang lancar dalam halaman e-pembelajaran. Navigasi yang baik akan membolehkan pelajar terus terlibat dalam mengendalikan tugas ketika menggunakan e-pembelajaran (Kamaruddin, 2010) dan dapat menggalakkan pelajar untuk menerokai e-pembelajaran secara keseluruhan manakala navigasi yang tidak dibangunkan dengan baik akan menyebabkan pelajar terperangkap dan bergerak di ruang yang sama (Reyna, 2009). Ini menyebabkan pelajar akan mengambil masa yang lama dan berasa kecewa ketika cuba berinteraksi menggunakan e-pembelajaran.

Kekurangan navigasi yang lancar menyukarkan para pelajar untuk menggunakan e-pembelajaran. Navigasi yang lancar adalah berpandukan kepada bantuan navigasi seperti anak panah, ikon dan butang yang perlu direka dengan baik, mudah dilihat dan senang difahami dan dilayari (Thomson, 2014). Walau bagaimanapun, antara muka e-pembelajaran yang mengandungi terlalu banyak bantuan navigasi akan menyukarkan pelajar dalam mengenali bentuk bantuan yang ditawarkan. Oleh itu, navigasi dalam antara muka e-pembelajaran adalah penting agar pelajar mudah untuk mencari maklumat dan menggunakan e-pembelajaran. Selain itu, navigasi yang lancar dapat memberi motivasi serta meningkatkan keterlibatan pelajar meneroka dan menggunakan e-pembelajaran.

8. PERBINCANGAN: HUBUNG KAIT REKA BENTUK ANTARA MUKA E 

PEMBELAJARAN DAN KETERLIBATAN PELAJAR

Reka bentuk antara muka dan keterlibatan pelajar merupakan dua aspek yang berbeza tetapi ia saling berkaitan antara satu sama lain kerana keberkesanan antara muka e-pembelajaran ditentu berdasarkan penglibatan pelajar. Ulasan di atas telah membincangkan mengenai elemen antara muka yang mempengaruhi keterlibatan pelajar dalam e-pembelajaran yang diimplementasikan di institusi pengajian. Oleh itu, kertas ulasan ini adalah untuk mengulas mengenai elemen reka bentuk antara muka yang memainkan peranan dalam memberi kesan ke atas keterlibatan pelajar dalam e-pembelajaran di institusi pengajian.

Berdasarkan ulasan di atas, terdapat beberapa elemen antara muka yang boleh memberi kesan ke atas keterlibatan pelajar dalam e-pembelajaran. Namun begitu, terdapat elemen antara muka yang dikemukakan oleh penyelidik lepas berbeza antara satu sama lain. Sebagai contoh, kajian yang dijalankan oleh Adnan et al. (2015) menekankan elemen imej, bentuk, warna dan teks sebagai elemen yang mampu membuatkan pelajar terlibat secara konsisten dalam e-pembelajaran. Beliau menyatakan

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 103

 

bahawa elemen warna merupakan elemen yang penting kerana ia mampu menarik minat pelajar untuk terus terlibat dalam e-pembelajaran dan elemen bentuk pula menjurus kepada penggunaan simbol bagi memudahkan pelajar memahami ikon atau logo tanpa perlu menggunakan teks yang banyak. Sebaliknya, Faghih et al. (2013) pula menekankan elemen imej, muzik dan animasi dalam antara muka e-pembelajaran bagi meningkatkan keterlibatan pelajar. Perbezaan elemen antara muka disebabkan oleh keperluan dan kehendak pengguna antara muka yang berbeza antara satu sama lain. Selain itu, latar belakang sosio budaya pelajar juga memainkan peranan dalam menentukan elemen antara muka e-pembelajaran. Ini kerana elemen yang tidak menepati kehendak pelajar menyebabkan mereka keberatan untuk terus terlibat dalam menggunakan e-pembelajaran.

Berdasarkan elemen antara muka yang dikemukakan oleh penyelidik lepas, elemen warna sering kali disebut oleh kebanyakan penyelidik seperti Jadual 2. Pemilihan warna yang sesuai adalah penting untuk memastikan kualiti ketepatan dan penyampaian maklumat dalam reka bentuk antara muka. Tambahan pula, pemilihan warna yang baik juga berkaitan dengan aspek psikologi manusia. Ini kerana warna berperanan mempengaruhi persepsi manusia (Ibrahim, Ashaari, & Wook, 2013). Penggunaan warna yang tidak sesuai boleh menyebabkan pengguna memberi pelbagai reaksi dan perasaan yang berbeza boleh memberi kesan terhadap keterlibatan pelajar. Kombinasi warna yang menarik pada antara muka boleh menimbulkan minat pelajar untuk menggunakan e-pembelajaran secara berterusan.

Selain itu, penyelidikan mengenai keterlibatan pelajar dalam komunikasi manusia dan komputer masih kurang. Keterlibatan pelajar sering kali dikaji dari aspek kehidupan harian pelajar seperti menghadiri kelas, aktif bertanya dan menyiapkan kerja sekolah tanpa melibatkan hubungan antara manusia dan komputer. Namun, definisi keterlibatan pelajar yang telah dibentangkan boleh dijadikan sebagai titik permulaan bagi mengimplementasikan keterlibatan pelajar dalam reka bentuk antara muka e-pembelajaran. Penyelidikan dalam menghasilkan reka bentuk antara muka yang menarik minat pelajar diharap dapat menyelesaikan masalah e-pembelajaran yang sering diabaikan tanpa digunakan semaksima mungkin.

9. KESIMPULAN

Reka bentuk antara muka e-pembelajaran adalah penting kerana ia berperanan dalam interaksi pelajar dengan e-pembelajaran. Faktor reka bentuk antara muka yang lemah adalah faktor utama yang menyebabkan pelajar kecewa untuk menggunakan e-pembelajaran dan seterusnya mengurangkan keterlibatan dan komitmen pelajar terhadap e-pembelajaran. Ramai penyelidik lepas mencadangkan agar memberikan lebih perhatian terhadap reka bentuk antara muka dalam menarik minat pelajar kerana reka bentuk antara muka sangat kritikal kerana ia berkait rapat dengan keberkesanan e-pembelajaran. Oleh itu, elemen antara muka memainkan peranan penting dalam membuatkan antara muka e-pembelajaran lebih menarik dan boleh digunakan. Dalam ulasan di atas, penulis telah

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 104

 

menyenaraikan lima elemen antara muka e-pembelajaran yang mampu meningkatkan keterlibatan pelajar terhadap e-pembelajaran iaitu elemen warna, imej, tipografi, navigasi dan reka letak. Berdasarkan tinjauan literatur, lima elemen yang dipilih adalah yang paling kerap muncul dalam penyelidikan lepas namun, elemen tersebut mungkin akan berubah setelah kajian empirikal dijalankan kelak.

Tambahan pula, penyelidikan mengenai keterlibatan pelajar dalam persekitaran bagi interaksi antara manusia dan komputer masih lagi kurang dan tidak mendapat perhatian seperti mana yang dibincangkan di atas. Oleh itu, mengimplementasikan budaya dalam reka bentuk antara muka e-pembelajaran merupakan satu jalan bagi meningkatkan keterlibatan dan komitmen pelajar terhadap e-pembelajaran. Ini kerana kehendak dan keperluan pelajar yang terdiri daripada latar belakang yang berlainan terhadap reka bentuk antara muka adalah berbeza antara satu sama lain dan menimbangkan perbezaan ini ketika membangunkan antara muka e-pembelajaran adalah penting bagi memastikan reka bentuk antara muka tersebut mampu menyampaikan pembelajaran yang berkesan kepada para pelajar. Selain itu, mengabaikan peranan budaya dalam antara muka boleh menyebabkan reka bentuk antaramuka tidak menepati kehendak pelajar dan seterusnya menolak daripada menggunakan e-pembelajaran. Terdapat pelajar yang tidak menggunakan e-pembelajaran jika reka bentuk antara muka tidak menggambarkan budaya mereka. Oleh itu, kajian ini berbaloi untuk diterokai dengan lebih mendalam dalam meningkatkan keberkesanan dan kejituan antara muka e-pembelajaran dan seterusnya meningkatkan keterlibatan pelajar menggunakan e-pembelajaran.

RUJUKAN

Adnan, A. S., Ali, M., & Ahmad, R. (2015). The Utilisation of Visual Elements on Interface Design

of e-learning . In International Conference on Information Technology & Society (pp. 273–279). Ali, A., & Smith, D. (2015). Comparing Social Isolation Effects on Students Attrition in Online

Versus Face-to-Face Courses in Computer Literacy, 12, 11–20.

Alias, N., Zakariah, Z., Ismail, N. Z., & Aziz, M. N. A. (2012). E-Learning Successful Elements for Higher Learning Institution in Malaysia. Procedia - Social and Behavioral Sciences, 67(November 2011), 484–489. doi:10.1016/j.sbspro.2012.11.353

Altaboli, A., & Lin, Y. (2011). Investigating effects of screen layout elements on interface and screen

design aesthetics. Advances in Human-Computer Interaction, 2011. doi:10.1155/2011/659758 Angelino, L. M., Williams, F. K., & Natvig, D. (2007). Strategies to Engage Online Students and

Reduce Attrition Rates. The Journal of Educators Online, 4(2), 1–14.

Arockiam, L., & Selvaraj, J. C. (2013). User Interface Design for Effective E-Learning based on Personality Traits. International Journal of Computer Applications, 61(14), 28–32.

Backs, A. (2012). Dragging the Y Generation Into E-Learning : A Study of Engaging Averse Online

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 105

 

Learners (pp. 30–40).

Bakar, Z. A., & Long, P. (2013). A Study of Visual Appeal Interfaces Based on Subjective Preferences, (November), 25–26.

Bart, M. (2012). Online Student Engagement Tools and Strategies.

Barzegar, S., Shojafar, M., & Keyvanpour, M. R. (2009). User interface in virtual learning environment. Internet Technology and Secured Transactions, 2009. ICITST 2009. International Conference for, 1–6. doi:10.1109/ICITST.2009.5402572

Benaida, M. (2014). Developing Arabic Usability Guidelines for E-Learning Websites in Higher Education.

Boton, E. C., Wales, N. S., Gregory, S., & Wales, N. S. (2015). Minimizing Attrition in Online Degree Courses. Journal of Educators Online, 62–90.

Chakrabarty, S., & Mohamed, N. (2013). Problem Based Learning: Cultural Diverse Students ’ Engagement , Learning and Contextualized Problem Solving in a Mathematics Class, 2013(Singaram 2010), 38–49.

Ching, G. (2012). Factors affecting student engagement: An analysis on how and why students learn, (December).

Delialioǧlu, Ö. (2012). Student engagement in blended learning environments with lecture-based and problem-based instructional approaches. Educational Technology and Society, 15(3), 310–322. doi:10.2307/jeductechsoci.15.3.310

Dixson, M. D. (2010). Creating effective student engagement in online courses: What do students find engaging? Journal of the Scholarship of Teaching and Learning, 10(2), 1–13.

Dunleavy, J., & Milton, P. (2009). What Did You Do in School Today? Exploring the Concept of Student Engagement and Its Implications for Teaching and Learning in Canada. Concept Paper May 2009.

Essam, S., & Al-Ammary, J. (2013). The Impact of Motivation and Social Interaction on the E 

Learning at Arab Open University, Kingdom of Bahrain. Creative Education, 04(10), 21–28. doi:10.4236/ce.2013.410A004

Faghih, B., Azadehfar, M. R., & Katebi, S. D. (2013). User Interface Design for E-Learning Software. The International Journal of Soft Computing and Software Engineering, 3(3), 786–794. doi:10.7321/jscse.v3.n3.119

Fredricks, J., McColskey, W., Meli, J., Mordica, J., Montrosse, B., & Mooney, K. (2011). Measuring student engagement in upper elementary through high school: a description of 21 instruments. Issues and Answers Report, 098, 26–27. Retrieved from http://ies.ed.gov/ncee/edlabs

Galitz, W. O. (2007). The Essential Guide to An Introduction to GUI Design Principles and Techniques. Xtemp01.

Guralnick, D. (2006). User Interface Design for Effective, Engaging E-Learning (pp. 1–10). Retrieved

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 106

 

from http://www.kaleidolearning.com/Effective E-Learning.pdf

Ibrahim, M. A., Ashaari, N. S. @, & Wook, T. S. M. T. (2013). Colour Impressions Framework. Procedia Technology, 11(Iceei), 305–311. doi:10.1016/j.protcy.2013.12.195

Kamaruddin, N. (2010). Challenges of Malaysian Developers in Creating Good Interfaces for

Interactive Courseware. The Turkish Online Journal of Educational Technology, 9(1), 37–42. Kamaruddin, N. (2012). Understanding Interface Design Role in Facilitating Teaching and Learning

Process: A Conceptual Framework, 1–11.

Kamaruddin, N. (2014). Understanding Interface Design Role in Facilitating Teaching and Learning Process: A Conceptual Framework, 1–11.

Kamaruddin, N., Park, J. Y., & Nam, Y. H. (2012). The Quality of Interface Design for Educational Courseware Development in Malaysian Educational Context, XXXIII(2), 81–87. doi:10.1007/s13398-014-0173-7.2

Katuk, N. (2013). Progressive assessment of student engagement with web-based guided learning.

Interactive Technology and Smart Education, 10(2), 116. doi:10.1108/ITSE-11-2012-0023

Khanum, M. A., Fatima, S., & Chaurasia, M. A. (2012). Arabic Interface Analysis Based on Cultural

Markers. International Journal of Computer Science Issues (IJCSI), 9(1), 255.

Kim, I., & Kuljis, J. (2010). Manifestations of Culture in Website Design, 125–132.

Ladbrook, J., & Parr, J. (2015). Motivation, Leadership and Curriculum design. In Motivation, Leadership and Curriculum design (pp. 161–172). doi:10.1007/978-981-287-230-2

Lim, Y. M., Ayesh, A., & Chee, K. N. (2013). Socio-Demographic Differences in the Perceptions OF Learning Management System. International Journal of Software Engineering and Applications, 4(5), 15–35. Retrieved from http://airccse.org/journal/ijsea/papers/4513ijsea02.pdf

Mcfarlane, P. S. (2011). Determining Student Preference for a Culturally Specific e-Learning Website.

Müller, R. (2014). Generation Y Students’ Product Colour Preferences. Mediterranean Journal of Social Sciences, 5(21), 69–74. doi:10.5901/mjss.2014.v5n21p69

Newmann, F. M., Wehlage, G. G., & Lamborn, S. D. (1992). Student Engagement and Achievement in American Secondary Schools. Student engagement and achievement in American secondary schools.

Parsons, J., & Taylor, L. (2011). Student Engagement: What do we know and what should we do ? The University Partners, (March), 1–59. Retrieved from https://education.alberta.ca/media/6459431/student_engagement_literature_review_2011.pdf

Ramakrisnan, P., Jaafar, A., Razak, F. H. A., & Ramba, D. A. (2012). Evaluation of user Interface Design for Leaning Management System (LMS): Investigating Student’s Eye Tracking Pattern and Experiences. Procedia - Social and Behavioral Sciences, 67, 527–537. doi:10.1016/j.sbspro.2012.11.357

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 107

 

Reyna, J. (2009). Developing quality e-learning sites: A designer approach. In AScilite 2009 (pp. 837–

838). Retrieved from http://www.ascilite.org/conferences/auckland09/procs/reyna-1-poster.pdf Reyna, J. (2013). The importance of visual design and aesthetics in e-learning. Australian Institute of

Training and Development, (40), 28–32.

Shahzad, A., Hafizi, W., & Golamdin, A. G. (2014). E-Learning Navigation Model Based on Student ’ s Learning Behavior: Case Study in UUM. International Journal of Business and Social Science, 5(5), 118–130.

Sithole, K., B. D., I., & Onyari, E. K. (2012). Influence of Generations’ Traits on teaching and Learning in an Open Distant Learning (ODL) Environment, 1–9.

Thomson, S. (2014). 9 Tips To Improve Online Learners’ Engagement. Retrieved from https://elearningindustry.com/9-tips-improve-online-learners-engagement

Tomczyk, L. (2009). Graphical User Interface in e-Learning System Dedicated for Seniors, 617–622. doi:10.1007/978-0-387-89024-1

Trowler, V. (2010). Student engagement literature review. Higher Education, (November), 1–15. Retrieved from http://americandemocracy.illinoisstate.edu/documents/democratic-engagement-white-paper-2_13_09.pdf

Wang, C., Ke, S., Chuang, H., Tseng, H., & Chen, G. (2010). E-learning system design with humor and empathy interaction by virtual human to improve students’ learning. In Proceedings of the 18th International Conference on Computers in Education. Putrajaya, Malaysia: Asia-Pacific Society for Computers in Education (pp. 615–622).

Welch, B. K., & Bonnan-White, J. (2012). Twittering to increase student engagement in the university classroom. Knowledge Management and E-Learning, 4(3), 325–345.

Yaa, N., & Adzobu, A. (2014). Design, Use and Evaluation of E-Learning Platforms: Experiences and Perspectives of a Practitioner from the Developing World Studying in the Developed World, 147–159. doi:10.3390/informatics1020147

Zaharias, P. (2008). Cross-cultural differences in perceptions of E-learning usability: An empirical

investigation. International Journal of Technology and Human Interaction, 4(3), 1–26.

Retrieved from http://www.scopus.com/inward/record.url?eid=2-s2.0-56249091448&partnerID=40&md5=2b4dab9138916809f8018a192115af9b

Zhang, Y. (LeaF). (2013). Power Distance in Online Learning: Experience of Chinese Learners in U . S . Higher Education. The National Review of Research in Open Distance Learning, 14(4), 239 – 254.

MyJICT - Malaysian Journal of Information and Communication Technology Vol 1 2016, Issue 2 | e-ISSN 0127-7812 108

 

An applicant for a driver license (DL) or identification card (ID) must present proof of lawful presence in the US. The table on the following pages describes the acceptable documents for each type of applicant attempting to verify lawful presence. All documentation must show the applicant’s name and date of birth. The applicant must validate a name change or other inconsistent information through additional documentation such as a marriage license, divorce decree or court order.

The department must verify applicable lawful presence documentation through the US Department of Homeland Security’s (DHS) Systematic Alien Verification for Entitlements (SAVE) Program. Verification through SAVE is often instantaneous, but when it is not, receipt of the DL/ID may be delayed for up to 30 days. If SAVE cannot verify on the first attempt, SAVE will permit two additional stages of verification. Each stage may require additional documentation from the applicant. After each stage, the applicant will receive instructions either verbally or by mail on how to proceed with the transaction. To avoid further delay, the applicant should comply with the instructions fully and as soon as possible. If the applicant provides timely responses, the process timeline generally occurs as follows.

Stage DLD receives response

from DHS DLD response to applicant

First Within a few seconds If verified, card issued

Second 3 to 5 business days Instruction letter issued within 48 hours after DHS response received by DLD

Third Up to 20 additional business days after second response received from DHS Instruction letter issued within 48 hours after DHS response received by DLD

Temporary Visitor/Limited Term Issuance

An applicant may be issued a limited term DL/ID if he or she is NOT:

  A US citizen;

A US national;

A lawful permanent resident;

A refugee; or

An asylee.

A limited term DL/ID will expire with the applicant’s lawful presence as determined by DHS. Commercial Driver Licenses

This guide does not apply to commercial driver licenses. A person who is a US citizen, US national, lawful permanent resident, refugee or asylee may apply for a commercial driver license. All others may apply for a nonresident commercial driver license, if eligible. Refer to http://www.dps.texas.gov/DriverLicense/CommercialLicense.htm or Chapter 522 of the Transportation Code for application and eligibility requirements.

(Rev. 7-13)

 

Category Acceptable Documents

U.S. Citizen v Birth certificate issued by the appropriate vital statistics agency of a U.S. State, a U.S. territory, or the District of Columbia indicating birth in U.S.

v Department of State Certification of Birth issued to U.S. Citizens born abroad (FS-240, DS-1350, or FS-545) or Consular Report of Birth Abroad

v Certificate of U.S. Citizenship

v Certificate of Naturalization

v U.S. Dept. of Justice – INS U.S. Citizenship Identification Card (I-197 or I¬179)

v Northern Mariana Card (I-873)

v U.S. passport book that does not indicate on the last page that "THE

BEARER IS A UNITED STATES NATIONAL AND NOT A UNITED STATES CITIZEN"

v U.S. passport card

U.S. National U.S. passport book that indicates on the last page that "THE BEARER IS A UNITED STATES NATIONAL AND NOT A UNITED STATES CITIZEN"

Kickapoo Traditional Tribe of Texas (“KIC”) (U.S. citizen) American Indian Card (form I-872) which indicates “KIC”

Kickapoo Traditional Tribe of Texas (“KIP”) (non-U.S. citizen) American Indian Card (form I-872) which indicates “KIP”

American Indian born in Canada (First Nations) An applicant may refer to the Jay Treaty, 8 U.S.C. § 1359, or 8 C.F.R. § 289.2 and may present a variety of documents. Issuance cannot occur without approval of the documents by Austin headquarters. DLD Personnel: make copies of documentation and seek approval through the chain of command.

Lawful Permanent Resident v Permanent Resident Card (I-551)

v Resident Alien Card (I-551) – card issued without expiration date

v Valid Immigrant Visa (with adit stamp) and unexpired foreign passport

v Unexpired foreign passport stamped with temporary I-551 language (adit stamp), "Approved I-551," or "Processed for I-551"

v I-94 stamped with temporary I-551 language (adit stamp), "Approved I¬551," or "Processed for I-551"

v Re-entry Permit I-327

Note: I-151, the predecessor to I-551, is not acceptable as proof of permanent resident status.

Immigrant Visa with Temporary I- 551 language A valid Immigrant Visa within one year of endorsement (i.e. stamped by Customs and Border Protection – adit stamp) and an unexpired passport

Conditional entrants Immigration documentation with an alien number or I-94 number indicating this status, which can include but is not limited to:

v I-94 or other document showing admission under Section 203(a)(7), “refugee conditional entry”

v I-688B coded 274a.12(a)(3)

v I-766 with category A3 or A03

 

2

 

Category Acceptable Documents

Asylee Immigration documentation with an alien number or I-94 number indicating this status, which can include but is not limited to:

v I-94 with annotation “Section 208” or “asylee”

v Unexpired foreign passport with annotation “Section 208” or “asylee”

v I-571 Refugee Travel Document

v I-688B coded 274a.12(a)(5)

v I-766 with category A5 or A05

Refugee Immigration documentation with an alien number or I-94 number indicating this status, which can include but is not limited to:

v I-94 with annotation “Section 207” or “refugee”

v Unexpired foreign passport with annotation “Section 207” or “refugee”

v I-571 Refugee Travel Document

v I-688B coded 274a.12(a)(3)

v I-766 with category A3 or A03

Temporary Protected Status (TPS) Immigration documentation with an alien number or I-94 number indicating this status or Employment Authorization Document (EAD) (I-766) with category A12 or C19

Applicant with Employment Authorization Document Employment Authorization Document (EAD)( I-766)

Applicants for adjustment of status

Note: These are individuals applying to become lawful permanent residents. Immigration documentation with an alien number or I-94 number

This can include but is not limited to a form I-797 indicating pending I-485 or pending application for adjustment of status.

Applicants for extension of status, change of status, petition for non-immigrant worker, with a pending I-918 application, or other pending category. Immigration documentation with an alien number or I-94 number

This can include but is not limited to a form I-797 indicating a pending application for an extension of status, change of status, petition for non-immigrant worker, or other pending category.

Citizens of the Republic of Palau Unexpired foreign passport or I -94 with annotation “CFA/PAL” or other annotation indicating the Compact of Free Association/Palau

OR

Employment Authorization Document (EAD)(I-766) with category A8 or A08

Citizens of the Republic of the Marshall Islands Unexpired foreign passport or I -94 with annotation “CFA/RMI” or other annotation indicating the Compact of Free Association/Republic of Marshall Islands

OR

Employment Authorization Document (EAD)(I-766) with category A8 or A08

 

3

 

Category Acceptable Documents

Citizens of the Federated States of Micronesia Unexpired foreign passport or I -94 with annotation “CFA/FSM” or other annotation indicating the Compact of Free Association/Federated States of Micronesia

OR

Employment Authorization Document (EAD)(I-766) with category A8 or A08

Cuban/Haitian entrants Immigration documentation with an alien number or I-94 number

This can include but is not limited to an I-94 with annotation “Cuban/Haitian entrant”

Lawful temporary residents Immigration documentation with an alien number or I-94 number

Self-petitioning abused spouses or children, parents of abused children, or children of abused spouses

(Applicants with Violence Against Women Act (VAWA) petitions) Immigration documentation with an alien number or I-94 number

This can include but is not limited to I-797 indicating approved, pending, or prima facie determination of I-360 or an approved or pending I-360 or an I-766 with category C31.

Parolees Immigration documentation with an alien number or I-94 number

This can include but is not limited to an I-94 with annotation “parole” or “paroled pursuant to Section 212(d)(5).”

Person granted deferred action Immigration documentation with an alien number or I-94 number

Persons granted deferred enforcement departure (DED) Immigration documentation with an alien number or I-94 number or Employment Authorization Document (EAD) (I-766) with category A11

Note: Individuals in this status may have been granted an extension to the

period of authorized stay that is not reflected on the current EAD. Notifications regarding any extensions to this category will be distributed by Austin headquarters.

Person granted family unity Immigration documentation with an alien number or I-94 number

Persons under an order of supervision Immigration documentation with an alien number or I-94 number

Persons granted extended or voluntary departure Immigration documentation with an alien number or I-94 number

 

4

 

Category Acceptable Documents

Persons granted withholding of deportation or removal Immigration documentation with an alien number or I-94 number

This can include but is not limited to an I-94 or passport with annotation “Section 243(h)” or a letter or order from USCIS or court granting withholding of deportation or removal.

Persons in removal or deportation proceedings Immigration documentation with an alien number or I-94 number

Persons granted a stay of deportation Immigration documentation with an alien number or I-94 number

Persons granted voluntary departure Immigration documentation with an alien number or I-94 number

A-1, A-2, and A-3 Unexpired foreign passport or I -94

Note: Issuance cannot occur unless applicant presents a letter from U.S. Department of State with original signature indicating ineligibility for

Department of State issued driver license or requesting issuance of a state

issued identification card.

B1/B2 Visa/BCC with I-94

(Border Crosser Card , DSP-150, or “laser visa”) All of the following:

¨ Unexpired foreign passport,

¨ Visa (border crosser card), and

¨ I -94

Note: Applicant must have an I-94 to be eligible because of the time and

distance from the border restrictions for applicants who do not obtain an I-94.

B-1, B-2, C-1, C-3, D-1, and D-2 Unexpired foreign passport or I -94

Note: The applicant may not be able meet residency/domicile requirements.

C-2

Alien in transit to U.N. Headquarters district. Travel limited to 25 miles radius of Columbus Circle in New York, NY This status is restricted to New York, NY and not eligible for a Texas driver license under the domicile/residency requirements.

E-1, E-2, and E-3 Unexpired foreign passport or I -94

E-2 CNMI

Treaty-investor and dependents in Commonwealth of the Northern Mariana Islands This status is limited to persons entering the Commonwealth of the Northern Mariana Islands (CNMI) and is not eligible for a Texas driver license (8 CFR § 214.2(3)(23)).

F-1

Foreign academic student Unexpired foreign passport or I -94 or I-20

 

5

 

Category Acceptable Documents

F-2

Dependent on F-1 Unexpired foreign passport or I -94

F-3

Commuter Student from Canada or Mexico This status is for commuters from Mexico or Canada and is not eligible for a Texas driver license under the domicile/residency requirements.

G-1, G-2, G-3, G-4, and G-5 Unexpired foreign passport or I -94

Note: Issuance cannot occur unless applicant presents a letter from US Department of State approving the issuance of a DL/ID.

H-113, H-1131, H-1C, H-2A, H- 213, H-2R, H-3, H-4, and I Unexpired foreign passport or I -94

]-1

Exchange visitor (may be student, trainee, work/travel, au pair, etc.) Unexpired foreign passport or I -94 or DS-2019

]-2

Dependent of J-1 exchange visitor Unexpired foreign passport or I -94

K-1, K-2, K-3, K-4, L-1, L-1A, L- 113, and L-2, Unexpired foreign passport or I -94

M-1

Non-academic student Unexpired foreign passport or I -94 or I-20

M-2

Dependents of non-academic students Unexpired foreign passport or I -94

M-3

Commuter Student from Canada or Mexico This status is for commuters from Mexico or Canada and is not eligible for a Texas driver license under the domicile/residency requirements.

N-1 through N-7 (NATO)

North American Treaty

Organization

Representatives and dependents Unexpired foreign passport or I -94

N-8, N-9, O-1, O-2, O-3, P-1, P 

2, P-3, P-4, Q-1, Q-2, Q-3, R-1, R-2, S-5, S-6, S-7, T-1, T-2, T-

3, T-4, T-5, TN-1, TN-2, TD, U-1, U-2, U-3, U-4, U-5, V-1, V-2, and V-3 Unexpired foreign passport or I -94

6

 

Category Acceptable Documents

WB*

Visitor for business (visa waiver program) Unexpired foreign passport with admission stamp annotated “WT/WB” or I -94 Note: The applicant may not be able meet residency/domicile requirements.

WT*

Visitor for pleasure (tourist in visa waiver program) Unexpired foreign passport with admission stamp annotated “WT/WB” or I -94 Note: The applicant may not be able meet residency/domicile requirements.

*Visa waiver program countries: Andorra, Australia, Austria, Belgium, Brunei, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Japan, Latvia, Liechtenstein, Lithuania, Luxembourg, Malta, Monaco, the Netherlands, New Zealand, Norway, Portugal, San Marino, Singapore, Slovakia, Slovenia, South Korea, Spain, Sweden, Switzerland, and the United Kingdom.

 

7

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

Challenges Facing the Development of Islamic Banking. Lessons

from the Kenyan Experience

Shamim Njeri Kinyanjui

Jomo Kenyatta University of Agriculture and Technology

Abstract

The purpose of the study was to investigate the challenges facing the development of Islamic banking in Kenya. The research designed for study was a case study approach focusing on four Islamic compliant banks in Kenya. These are: - First Community Bank; Gulf African Bank; Dubai Bank; Kenya commercial Bank Ltd and Barclays Bank Ltd. The population of the study consisted of 33 customers, who were the holders of accounts in the respective banks and 11 managers. Sample data collected by use of questionnaires administered by the researcher and a research assistant. Data analysis method used is based on the quantitative approach using descriptive statistics: mean, mode, and median. Frequency tabulations and cross tabulations were used to bring out the finding of the study. The study revealed Islamic banking compliant was driven by religious compliance and customers need being met. It also revealed that continuous review and improvement of shariah compliant products together with diversifying market niche will lead to drastic development and marketing of Islamic banking products. From the study the following conclusions were drawn: firstly, the factors that influence development of Islamic banking products in Kenya are purely religious compliance and customers need being met.

Keyword: Islamic Banking.

1.1 Introduction

There is no satisfactory definition of a bank and term bank, as such, may apply to an organization, a financial institution operated by an individual as sole proprietor, partnership, a corporation or other type of association, in modern use it is an establishment for the custody of money (Saddiqi, 1986).

One definition of an Islamic Bank is a bank that, by its own choice, opts to comply with two sets of law: the law of the Land (Jurisdiction); and the Islamic Law (Shari'ah). This is why Islamic bankers have two types of legal counsel: traditional ̏lawyers" and ̏Shari'ah Councils" (Al-Bahar, 1996). As Islamic finance is intertwined within its religion, the basis of the religion affects the finance in two important ways: ̏Islam aims at building a socio-economic order based on justice and considers economic activity as a means to an end and not an end in itself. It enjoins Muslims to harness natural resources, which are a trust from Allah( God) , for carrying out rightful activities; but abhors exploitation and man-made inequalities of income and wealth." ̏Islam is deeply concerned with the problem of economic development, but treats this as an important part of a wider problem, that of total human development. The primary function of Islam is to guide human development on correct lines and in the right direction. It deals with all aspects of economic development but always in the framework of total human development and never in a form divorced from this perspective" (Al-Harran, 1993).

Islamic banking has the same purpose as conventional banking except that it operates in accordance with the rules of Shari’ah, known as Fiqh al-Muamalat (Islamic rules on transactions). The basic principle of Islamic banking is the sharing of profit and loss and the prohibition of riba (usury/interest). Common terms used in Islamic banking include; profit sharing (Mudharabah), safekeeping (Wadiah), joint venture (Musharakah), cost plus (Murabahah), and leasing (Ijarah). In an Islamic mortgage transaction, instead of loaning the buyer money to purchase the item, a bank might buy the item itself from the seller, and re-sell it to the buyer at a profit, while allowing the buyer to pay the bank in installments. However, the bank's profit cannot be made explicit and therefore there are no additional penalties for late payment. In order to protect itself against default, the bank asks for strict collateral. The goods or land is registered to the name of the buyer from the start of the transaction. This arrangement is called Murabaha. Another approach is EIjara wa EIqtina, which is similar to real estate leasing. Islamic banks handle loans for vehicles in a similar way by selling the vehicle at a higher-than-market price to the debtor and then retaining ownership of the vehicle until the loan is paid (Kahf, 2004).

An interest free Islamic bank in relation to its clients plays the role of partner, investor and trader. The interest free modes of operation designed by Islamic financial institutions satisfy the criteria „God permits trading and forbid riba. Keeping in view of the Islamic injunctions against taking and giving of interest, the Islamic banks

 

94

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

have organized their operations on the basis of profit /loss sharing which is permitted in Islam. Islamic banks accept demand deposits and time deposits. Demand deposit are fully repayable on demand and do not get any return, holder of time deposits are given a share in the profit earned by the bank according to a profit sharing rates made known in advance.

1.2 Problem Statement

The study carried out by Gerrald in the year 1997 found that: Islamic religious belief and social responsibility are the two most important factors that determine bank selection. Cost benefit is the third most important factor considered in bank selection. Clients of conventional and Islamic banks share a number of motives, but they differ significantly on a few motives in relation to bank selection, the clients of Islamic banks are more familiar with the products/services that conform to the Shari’ah. However, the study by Erol and El-Bdour (1989) contradict the findings of other studies and argue that religious motivation did not appear to play a primary role in bank selection. Significant number of customers would withdraw their deposits if an Islamic bank did not generate sufficient profit to make a distribution in any one year hence motivated by higher dividend payments instead. Further, the results of the study by Oundo (2009) suggested that there was poor supply of Shari’ah-compliant products in Kenya’s financial institution. The challenges in the supply side of Shari’ah compliant financial services was illustrated by one bank that wrote Cheques to suppliers for their Muslim clients who were uncomfortable with receiving credit in cash against the Shari‟ah. There was high demand for Islamic financial products against a negligible supply of the same meaning that Islamic finance clients had few options (if any) and financially neglected by the microfinance sector. According to Ndungu (2010) Kenya was the first country in the East and Central African region to introduce Islamic banking in the year 2007. In this short period, two banks were licensed to exclusively offer Shari‟ah-compliant products with many other conventional banks establishing a window specifically for Shari‟ah-compliant products. He noted that the concept of shari‟ah complaint banking has emerged as an alternative vehicle for mobilization and supply of finance. For example, the two banks have already contributed in development agenda of the country by participating in Shari‟ah-compliant (Sukuks) components of infrastructure bonds issued by the Central Bank of Kenya on behalf of the Government of Kenya and “Structured Sukuk” is expected to cover the bonds and T-Bills market in future. However, although the concept of Islamic Finance has generated a lot of interest and overwhelming support from both Muslim and non-Muslim population in Kenya, as a regulator, CBK has faced by certain challenges which need to be addressed.

In their short period of existence, Islamic banking in Kenya has shown very commendable performance commanding combined market share of the banking sector in terms of gross assets of 0.8%. Currently there are two Islamic banks operating in Kenya: Gulf African and First Community bank, which had a loan portfolio of 4.9-billion shillings, deposits totaling 7.5billion shillings and 27270 deposit accounts. These indicators point to the tremendous potential of this market niche, which has been previously untapped, largely comprising Muslims estimated to make up at least 15% of Kenya's population of 36-million (Muriri, 2009). To address the above challenges highlighted by Central Bank of Kenya, this study seeks to establish factors influencing the development of Islamic banking in Kenya.

2.0 Literature Review

2.1 Islamic Banking Instruments

The Islamic banking model primarily relies on the instruments of mudarabah (joint venture) and musharakah (equity participation) to eliminate interest from the financial sector and economy. Other interest-free instruments such as murabaha (deferred payment sale), ijarah (leasing), bai Salam (advance payment) and bai istisna (procurement engagement) are also used to enhance the practical scope, diversification and risk management capabilities of the Islamic banking system.

2.1.1 Profit and loss sharing (PLS) Theory

Islamic scholars treat PLS instruments, mudarabah and musharakah as a central pillar of the Islamic banking model. In mudarabah banking, the Islamic bank accepts funds from depositors under risk-sharing arrangements. The Islamic bank either directly invests these funds in profitable investments or extends them to entrepreneurs on a risk-sharing basis. The Islamic bank shares the profit or loss made on mudarabah ventures with its depositors. In musharakah banking, the Islamic bank contributes the depositors' funds to a joint enterprise with

 

95

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

the client (an entrepreneur). Generally, the Islamic bank allows the client to manage all the affairs of a Musharakah business.

The Islamic bank and the client mutually share the profit or loss made on the Musharakah investment. In a typical PLS arrangement, an Islamic bank provides the risk capital to a firm in which professional managers are responsible for making strategic and operational decisions. The bank shares in profits and is liable to any financial loss. There is no serious problem with this arrangement if the bank is able, and is allowed, to monitor business operations of the firm. However, proper monitoring mechanisms are yet to be devised for PLS, especially in case of Mudaraba that does not provide any control rights to the financier (the Islamic bank in this case). Fiqh literature on this issue is quite out-of-date and needs serious reconsideration. For example, Saleh (1986) lists three rights and one responsibility of the financier in a Mudaraba arrangement. The rights include ensuring that the borrowing entrepreneur (firm) complies with the terms of the contract, sharing profits, and limited liability in case of loss. The sole responsibility is handing over the Mudaraba capital. He also outlines two rights and two responsibilities of the borrower.

The rights include conducting the business with an appropriate degree of freedom, and accounting decisions. The responsibilities are compliance with the terms of the contract, and liquidation of the Mudaraba business at the end of the contract. The modern use of Mudaraba as a mode of financing obviously requires more than such preliminary specification of rights and responsibilities. There is a need for construction of standardized PLS contracts, or bylaws, in the light of the legal frameworks of Muslim countries. A prominent feature of these bylaws should be definition of the rights and obligations of various officers or groups within the organizational structure. Similar bylaws should delineate the clauses related to performance of the borrowing firm compared with other firms in the same sector and, possibly, other firms.

2.1.2 Murabaha (deferred payment sale): theory and practice

Under the Murabaha arrangement, the client makes a promise to buy specified goods from the Islamic bank on a deferred payment basis. The Islamic bank purchases goods from the original supplier and sells them on to the client after adding its own profit margin. The legality of murabaha could not be established from the primary sources of Islamic Shari’ah, i.e. the holy Quran and Sunnah. The early Islamic jurists, such as Imam Malik (796) and Imam Shafi (820), approved murabaha sales but they did not refer to the increase in price in the case of deferred payment. Subsequently, certain other Islamic jurists, such as Sarakhsi (1091), Marghinani (1197) and Nawawi (1277), allowed the seller to charge a higher price in the deferred payment sale by characterizing it as a normal trade practice (Saadullah, 1994; Vogel and Hayes, 1998). Contemporary Islamic scholars have mixed opinions about the murabaha banking system. The majority of them have strong reservations about it because of its close resemblance to conventional banking practice.

2.1.3 Ijarah (lease financing): theory and practice

The features of Ijarah financing are very similar to those of conventional lease financing. However, unlike in the conventional lease contract, Shari’ah holds the leaser responsible for all damage, repairs, insurance and depreciation of the leased asset. The leaser should also bear the risk of uncertainty attached to the useful life of the leased asset. Islamic financial institutions mostly rely on leasing, known as Ijarah wa iqtina, for meeting financing needs in the real estate, retail, industry and manufacturing sectors. Leasing enjoys strong support from Shari’ah scholars and bears a close resemblance to conventional leasing (Iqbal, 2000).

2.1.4 Bai Salam (advance payment) and bai istisna (procurement engagement)

Bai salam and bai istisna are forward sale contracts in which the seller pays in advance the price of goods that are to be delivered to him at a specified future date. Bai salam was widely practiced in the Arabian agricultural sector long before the dawn of Islam. These instruments are best suited to meet the financing needs of the farming and manufacturing industries in the Islamic economy. Shari’ah stipulates that the terms and conditions of bai salam and bai istisna contracts, such as price, quantity and quality of goods, should be clearly determined without involving any features of interest, gharar (speculation) or dubious sale (Iqbal, 2000).

2.2 Challenges facing Islamic Banking in Kenya

Islamic banking industry has been trying for the last over two decades to extend its outreach to bring it at least to the level of conventional banking. But the absence of Shariah-compliant legal framework — needed to make interest-free banking acceptable (and create sound financial institutions) — is the major snag behind its low penetration in the financial market.

 

96

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

It is the time to take stock of challenges faced by the Islamic banks as they need a number of supporting institutions/arrangements to perform functions which are being carried out by various financial institutions in the conventional framework. Attempts should be made to modify the existing structure to provide better products and quality service within the ambit of Islamic laws.

Some of the most important challenges facing the Islamic banking industry are identified as follows.

Legal Support: Islamic law offers its own framework for execution of commercial and financial contracts and transactions. Nevertheless, commercial banking and company laws appropriate for implementation of Islamic banking and financial contracts do not exist. Islamic banking contracts are treated as buying and selling properties and hence are taxed twice. The commercial, banking and company laws contain provisions that are narrowly defined and prohibit the scope of Islamic banking activities within conventional limits. It is necessary that special laws for the introduction and practice of Islamic banking be put in place. The legal framework of Islamic banking and finances might include the following:

a. Islamic banking courts: The disputed cases of the Islamic banks are subject to the same legal system and are dealt with the same court and judge as the conventional one while the nature of the legal system of Islam is totally different. To ensure a proper, speedy and supporting Islamic legal system, amendments in existing laws, which are repugnant to injunctions of Islam, are required to promulgate Shariah compliant law for resolution of disputes through special courts.

b. Amendment of existing laws: Islamic banking has some kind of resemblance to universal banking, therefore, laws and regulations have to be amended accordingly to accommodate this new concept such as sections 7 (forms of business in which the banking company can engage) and 9 (prohibition of trade) of the Banking Companies Ordinance 1962 while Islamic banks are big or wholesale traders in reality.

c. Islamic banking law: In the absence of Islamic banking laws, the enforcement of agreements in courts may require extra efforts and costs. Therefore, banking and companies’ laws in several countries require suitable modifications to provide a level playing field for Islamic banks. Furthermore, international acceptance of Islamic financial contracts requires them to be Shariah compatible as well as acceptable under the major legal regimes such as Common law and Civil law systems.

d. Islamic banking balance sheet: Islamic banks do not show assets financed through Ijara, Murabah etc., on balance sheet because section 7 of Banking Ordinance 1962 does not allow a bank to own property or asset which section 9 prohibits to enter into any kind of trade. However, all the assets owned by Islamic banks be mentioned in their balance sheets.

e. Monthly payment agreement: The housing finance is executed on the basis of Diminishing Musharaka by the Islamic banks. Under this mode the house is jointly owned by the bank and the customer. The bank rents out its share to the customer on Ijara basis. The Islamic bank while executing Ijara with the partner/customer uses the term ‘Monthly Payment Agreement’ instead of having the Ijara agreement with the customer.

It is so named as to safeguard the bank’s interest in case of refusal by the customer to pay rentals. No legal cover is provided to the Islamic bank to overcome this risk.

f. PLS deposits: Deposits in Islamic banks are usually based on principle of profit and loss (Musharaka or Murabaha). If something happens and the bank suffers loss it has to be transferred to the depositor directly.

This fear of loss is the biggest barrier to deposit mobilisation in Islamic banks. In some cases, it leads to withdrawal of funds. The depositors should be provided with some kind of protection.

Islamic prudential regulations: Supervision of Islamic banks is equally important. At present, lack of effective prudential regulation is one of the weaknesses of the Islamic banking industry. For instance, leasing prudential regulations are applied to Ijara where the nature of both is different, such as taking advances. The bank is the owner in Ijara; so taking advances will render the contract of Ijara for conversion into Musharakah whereas the rules of Ijara are applied to it, which is illegal. And some of the Islamic banks are using the term of security, hence making the Ijara contract non-Shariah compliant as using the deposited sum under the heading of Ijara security (‘Rahn’) is nothing but Riba which is strictly prohibited by Islam.

Moreover, Ijara financing is subject to compulsory insurance which is essentially prohibited.

 

97

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

Risks: The nature of risk in Islamic banking is different from those of conventional banking and therefore some special prudential, accounting and auditing standards should be applied to them.

Benchmark: Taking the conventional interest based benchmarks (Kibor etc.,) as the base of pricing an Islamic financial product puts Islamic banks at the mercy of their conventional peers. A negative perception is created among the clientele that there is no prudent difference in Islamic bank products as these are also using the same interest based benchmark. The mechanism for long-term financing could be devised on the basis of prevailing renting system adopted by the private landlords while renting their assets/properties etc.

Shariah based product: All Islamic financial institutions offer the same basic products, (90 per cent Murabaha and Ijarah) but the problem is that each institution has its own group of Islamic scholars on the Shariah board to approve the product. Consequently, the very same product may have different features and will be subject to different kind of rules in these institutions.

Lack of standard financial contracts and products can be a cause of ambiguity and a source of dispute and cost. In addition, without a common understanding on certain basic foundations, further development of banking products is hindered.

Nature of Islamic banking: Islamic banks are offering only Murabaha and Ijarah while leaving the core and difference making Islamic financial instruments such as Musharakah and Murabah. It is necessary to enhance and facilitate the implementation of real Islamic banking activities i.e. promoting risk sharing through equity type facilities on the asset side and profit sharing investment accounts on the funding side.

Lender of last resort facility: Islamic banks are reluctant to enter into long-term transactions due the lack of availability of liquidation through secondary market. There is liquidity support in the form of lender of last resort facility.

There is no proper mechanism of transparency and disclosure to the public in order to ensure consumer protection as provided by Shariah.

Islamic future exchange: In conventional system, long-term finance is provided through long-term bonds and equities. Apart from the general public, the most important source of these long-term investments are investment banks, mutual funds, insurance companies and pension funds. Islamic banks do not deal with interest bearing bonds. Therefore, their need for equity markets is much higher. On the top of it, the most of the products in Islamic banks are based on goods and commodities while prices and currency rates go up and down frequently, creating a big risk for them being traders in reality especially in the case of Salam and Istisna'a. To hedge the risk, they are in need of derivative products and consequently of Future Exchanges.

Shortage of experts in Islamic banking: The supply of trained or experienced bankers has lagged behind the expansion of Islamic banking. The training needs affect not only domestic banks, both Islamic and non-Islamic, but foreign banks as well.

Absence of accounting (and auditing) standards pertinent to Islamic banks: Uncertainty in accounting principles involves revenue realization, disclosures of accounting information, accounting bases, valuation, revenue and expense matching, among others. Thus, the results of Islamic banking schemes may not be adequately defined, particularly profit and loss shares attributed to depositors.

Lack of uniform standards of credit analysis: Islamic banks have no appropriate standard of credit analysis. Similarly, there is a widespread training need involving related aspects such as financial feasibility studies, monitoring of ventures, and portfolio evaluation.

Potential conflicts with central banks: Islamic banks have been established as separate legal entities; therefore, their relationships with central banks and/or other commercial banks are uncertain. Problems may be further aggrivated when an Islamic bank is established in a non-Muslim nation like the case of Kenya, and is subject to that nation's rules and requirements.

Instruments that meet the demand of specific investment requirements: One of the biggest challenges facing institutions is the provision of short-term investment instruments. Several institutions have tried to develop high quality short-term instruments, but have been hampered by their ability to generate assets, by their credit ratings, and by liquidity.

 

98

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

3.0 Methodology

3.1 Design

The research design employed in this study was descriptive in nature in form of a survey. According to Chadran, (2004) a descriptive design enables the researcher to collect in depth information about the population being studied. Study by Gerrald (1997) have 27 used a similar research design where five significant factors were noted, the most relating to non Muslim being the desire to be paid higher interest on their savings.

3.2 Sampling

The study employed a combination of census and probability sampling. The census method was used to include all managers in the banks whose customers were sampled. The probability sampling was used to sample the customers; it was on a walk in basis. It was assumed that since the bank customers walk in at random, interviewing all willing customers was random enough. According to Kothari (2004), a sample is a representative group of the accessible population that is selected using various sampling methods. The sample size for the managers was 11 and that of the customers was 33. This was gotten from all the Islamic Banks located in Mombasa Island.

3.3 Data Collection

This study employed use of primary data, and obtained from bank employees and bank customers, by use of questionnaire. The use of questionnaire was justified because they 28 provide an effective way of collecting information from a large literate sample in a short period of time and at a reduced cost than other methods. Moreover, questionnaires facilitate easier coding and analysis of data collected.

3.3 Findings

The findings found out that 87.9% of the respondents were Muslims and only 12.1% were non-Muslims as shown in figure 3.3.1below.

 

Figure 3.3.1 religious population

 

99

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

To determine the factors that influenced the banks to start providing shariah compliant products and services, they were asked to rate the above factors on a five point likert scale where one represent no influence and five represent influence to a great extent. The findings shows that the managers were not sure about the influence from increased competition, Staff aggressiveness and commitment and having a conducive government regulations and support, all with a mean score of about three. The findings show that potential huge profit from Islamic banking and Service quality gap were the two major factors influencing entry. On the other hand corporate social responsibility (CSR), activity driven by public concern and high bank liquidity requiring investments, and diversification to cushion the risk of corporate borrowing does not have any influence. This is summarized in the table below.

Factors Mean Std. Dev

Increased competition in the traditional bank market 3.27 1.272

Potential Huge profit from Islamic banking 3.73 1.009

Service quality gap 3.91 0.831

High bank liquidity requiring investment 1.82 1.079

As CSR activity driven by public concern 2.36 0.924

Diversification to cushion the risk of public borrowing 2.18 0.982

Staff aggressiveness and commitment 2.64 1.206

Conducive government regulations and support 3.18 1.401

The statistics in the table below shows that the managers are not sure with laying down institutional policies and procedures to regulate Islamic products as a way of ensuring sustainability of Shari’ah compliant product/services for Commercial Banks. The managers also disagreed that separation of the funds between Islamic and Conventional funds could lead to sustainability of the Islamic compliant product or services. On the other hand, the managers agreed that the following were the factors that lead to sustainability of the products: increased number of customers; support from senior management and board; aligned Islamic banking products policy to the overall strategy of the bank; well versed shari’ah Board members; management and staff who understand 44

Islamic Banking Market dynamics; good customers’ services and establishment of accounting that comply with Islamic Shari’ah.

Factors Mean Std. Dev

Lowering Bank Charges 2.82 1.168

Increasing earnings on deposits and savings 2.09 1.136

Relaxing the collateral requirement during lending 2.91 1.044

Diversifying Islamic banking product 4.64 0.505

Continuous review and improvement of shariah compliant products 4.55 0.522

Bonus to marketing staff 2.73 1.555

Interest free lending 4.73 0.905

Profit loss sharing approach 4.64 0.505

5.0 Conclusion

Other than providing Islamic compliant products and services, factors such as: offering product and services that meet the customer’s needs, reduced borrowing cost, minimum requirements to open an account and service efficiency influences the development of Islamic banking. The Islamic banks should put these priorities at forefront and strengthening it. There should also be an amendment to the existing Banking Law. By doing so, they can also win a great number of customers from both Muslims and non-Muslims community. Above all, they should invest in branch expansion thus enabling the marketing Islamic banking products at the grassroots level.

Lastly, there are future market potentialities for the Islamic banking products. Many younger generations were seen going for these products and services. For this banking, system to develop and grow the banks should create awareness through seminars, workshops, and advertisement and prayer sermons quoting verses from Quran and sayings of the prophet (hadith) that concern Islamic Banking system.

 

100

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

6.0 Recommendations

In Kenya, all banks offering Islamic banking have established their own separate Shari’ah Board to supervise and offer guidance to their respective banks on Islamic banking system. In principle, Shari’ah Boards have the authority to impose their viewpoint, but logistic considerations do not permit timely vetting and/or monitoring of all banking operations. In view of these, I recommend CBK to put in place a policy to establish a universal Shari’ah Boards to oversee Islamic banking operation in Kenya. Conventional banking system in Kenya are compelled to have a uniform financial reporting standard thus mixing their funds that are both conventional and Islamic compliant. In view of this, Muslim customers are discouraged to bank with such conventional banks, as it is not acceptable in terms of shari’ah. Therefore, I recommend CBK to allow such conventional bank to have separate financial reporting standards. Lack of qualified Islamic banking personnel is another major bottleneck in Kenya. Further training and Islamic finance education should be given to bank personnel to up skill them to offer quality service and appropriate advice to bank customers. I recommend banks to organize regular training and workshops by inviting well-versed Islamic scholars to educate bank personnel about Islamic banking and financing.

References

Alexakis, C.&,Tsikouras, A, (2009). Islamic finance regulatory framework : challenges lying ahead, International Journal of Islamic and Middle Eastern Finance and Management, Vol. 2 Iss: 2, pp.90 – 104

Ahmad, S.M. (1952) “Interest and Unemployment” Islamic Studies, VIII I, pp. 9-46.

Al-Harran, S.A.S. (1993). Islamic Finance: Partnership Financing,Pelanduk Publications, Malaysia, pp. 15-157. Amin, H. (2008). Determination of choice criteria in Malaysia’s Retail Banking. Kuala Terrengganu.Malaysia. Anwar, M. (1989). Modeling interest free economy; Genuine publication. West, New Dhephi.

Al-Arabi,(1966) “Contemporary Banking Transactions and Islam’s Views Thereon” Islamic Review, May, pp. 10-16.

Ariff, (1982) Monetary and Fiscal Economics of Islam, International Centre for Research in Islamic Economics, Jeddah.

Bashir,(2000) Assessing the Performance of Islamic Banks: Some Evidence from the Middle East Paper retrieved on June 12, 2006 from http://www.luc.edu.orgs/meea/volume3/revisedbashir.pdf.

Bryman, C. (2001).Improving Internal Control: A Practical Guide for Microfinance Institutions.

Brunei. Khan, M. M. & Bhatti,M. I., (2008) Development in Islamic banking: a financial risk-allocation approach.VOL.1 PP 40-51. Khan, M. M., & Bhatti ,M. I. (2006). Why interest-free banking and finance movement failed in Pakista.Humanomics Vol 22, 3 pp 145-161. Australia. Kothari, C.R. (2004). Research methodology: methods & techniques. New Delhi: New age international (P) limited publishers.

Chandran, E. (2004). Research methods: A quantitative approach with illustrations from Christian ministries. Daystar University, Nairobi.

Chapra, (1982) “Money and Banking in an Islamic Economy”. In Ariff, M. Ed. Monetary and Fiscal Economics of Islam, Jeddah: International Centre for Research in Islamic Economics.

Cooper, D., & Schindler, P. (2000). Business research methods. (8th edn) New Delhi, Tata Mc Graw Hill. Ebrahim,

C. R. Kothari (1990) Research Methodology, Methods and Techniques. New Delhi, Wishwa.

 

101

 

European Journal of Business and Management www.iiste.org

ISSN 2222-1905 (Paper) ISSN 2222-2839 (Online) Vol.5, No.22, 2013

Erol, C. & El-Bdour, R. (1989). “Attitudes, behaviour and patronage factors of bank customers towards Islamic banks”. International Journal of Bank Marketing, Vol. 7 No. 6, p. 31.

Gafoor (2007) Islamic Banking Paper retrieved on June 27, 2007 from .

http://users.bart.nl/~abdul/chap4.html.

Gerrald, P. (1997).Division of Banking and Finance, Anyang business school.Nyanyang Technology University. ”. International Journal of Bank Marketing, 15 /16 pp204-216.

Husain, I.( 2010).Keynote address delivered at the Conference on Islamic banking and Finance : Islam and the Future of Economics .Karachi .

Iqbal, M. ed. 2000. Islamic Banking and Finance: Current Developments in Theory and Practice.Islamic Foundation and IAIE.

Iqhal (1991) Stabilization and growth in an open Islamic economy,Review of Islamic economy vol 1(2): 1-20 Kahf, M. (2004).Success factor of Islamic Banks, Symposium on Islamic Banking and Finance.

Kumar,M., Kee, F. T., & Charles .V. (2010).International Journal of Quality & Reliability Management, Volume 27, Number 3, 2010, pp. 352-378(27).Emerald Group Publishing Limited. Mingala, M. S. (2002). Financial Markets and Institutions. Addison Wesley, New York.

Khan (1987) Theoretical studies in Islamic banking and Finance, Houston institute for research and Islamic studies.

Microfinance Network with GTZ, Technical Guide No. 1 CBK. (2010). Targets Gulf cash with compliance bonds: Business daily news. Nairobi. Kenya.

M.S,(2001).Islamic banking in Brunei Darussalam. University of Brunei Darussalam/National University of Singapore, Singapore, and Tan Kai Joo Kemuda.te for research and Islamic studies.

Mirakhor (1989) Monetary management in an Islamic Economy:International Monetary Fund Washington D.C

Nasim, (1998), ‘Key Concepts of Islamic Finance: Their Relevance to Consumer Finance Products’, Proceedings of the Second Harvard University Forum on Islamic Finance, Cambridge, Massachusetts: Harvard University, 291-93.

Nasim, Pervez (1999), ‘Various Islamic Modes of Home Financing’ in Islamic Banking and Finance, The Concept, the Practice and the Challenge (ed.), Imtiazuddin Ahmad, Indianapolis: Islamic Society of North America, 69-73.

Siddiqi, M.N. 1988 “Islamic Banking: Theory and Practice”. Islamic foundation.

 

102

 

This academic article was published by The International Institute for Science, Technology and Education (IISTE). The IISTE is a pioneer in the Open Access Publishing service based in the U.S. and Europe. The aim of the institute is Accelerating Global Knowledge Sharing.

More information about the publisher can be found in the IISTE’s homepage: http://www.iiste.org

CALL FOR JOURNAL PAPERS

The IISTE is currently hosting more than 30 peer-reviewed academic journals and collaborating with academic institutions around the world. There’s no deadline for submission. Prospective authors of IISTE journals can find the submission

instruction on the following page: http://www.iiste.org/journals/ The IISTE

editorial team promises to the review and publish all the qualified submissions in a fast manner. All the journals articles are available online to the readers all over the world without financial, legal, or technical barriers other than those inseparable from gaining access to the internet itself. Printed version of the journals is also available upon request of readers and authors.

MORE RESOURCES

Book publication information: http://www.iiste.org/book/

Recent conferences: http://www.iiste.org/conference/ 

IISTE Knowledge Sharing Partners

EBSCO, Index Copernicus, Ulrich's Periodicals Directory, JournalTOCS, PKP Open Archives Harvester, Bielefeld Academic Search Engine, Elektronische Zeitschriftenbibliothek EZB, Open J-Gate, OCLC WorldCat, Universe Digtial Library , NewJour, Google Scholar

 

 

Jeffrey Graves

Curriculum Vitae

 

EDUCATION

2014–Present Ph.D., Computer Science, Tennessee Tech University, Cookeville, TN. Adviser William Eberle

2011–2016 M.S., Mathematics, Tennessee Tech University, Cookeville, TN.

Thesis Walsh Functions as Group Characters

Adviser Rafal Ablamowicz

2007–2011 M.S., Computer Science, Tennessee Tech University, Cookeville, TN.

Thesis Source Code Plagiarism Detection Using a Graph-Based Approach

Adviser William Eberle

2003–2007 B.S., Computer Science, Tennessee Tech University, Cookeville, TN.

GPA 3.61/4.0

Honors Cum Laude

AWARDS AND HONORS

2014 Graduate Student Teaching Award, Mathematics Department, Ten¬nessee Tech University.

2013 Challenge Assistantship, Mathematics Department, Tennessee Tech University.

TEACHING EXPERIENCE

Computer Science

o Design and Analysis of Algorithms, Fall 2014, Spring 2015, Fall 2015

Mathematics

o College Algebra, Spring 2012, Fall 2012, Fall 2013

o Finite Math, Spring 2013, Spring 2014

o Pre-Calculus I, Fall 2013, Spring 2014

o Pre-Calculus II, Spring 2014

o Transitional Algebra, Fall 2013

 

RESEARCH EXPERIENCE

Summer 2016 Advanced Data and Workflows Intern, National Center for Compu¬tational Sciences, Oak Ridge Natural Laboratory.

Summer 2015 Computational Data Analytics Intern, Computational Sciences and Engineering Division, Oak Ridge Natural Laboratory.

2009–2010 Department of Energy Grant, Center for Manufacturing Research, Tennessee Tech University.

Development of an electrical capacitance tomography sensing system to detect defects that arise in lost-foam casting processes as well as the development of packet recovery techniques for use wireless sensor networks deployed in foundry environments.

2008–2011 Department of Homeland Security Grant, Computer Science De¬partment, Tennessee Tech University.

Focussed on applying graph-based anomaly detection methods to areas such as insider threat detection, intrusion detection, cargo shipping, and source code similarity metrics.

2007–2008 Vanderbilt eHealth Initiative Subgrant, Computer Science Depart¬ment, Tennessee Tech University.

Development of a framework to integrate electronic patient record systems. Implementation of a distributed patient information system for health care providers to enable the sharing of patient records between separate health care organization environments.

UNIVERSITY SERVICE

2015–2017 Graduate Executive Committee Member, College of Engineering, Tennessee Tech University.

2015–2017 Graduate Student Mentor, Computer Science Department, Tennessee Tech University.

RESEARCH COMMUNITY SERVICE

2015 SIAM Data Mining Conference (SIAM) External Reviewer.

ACADEMIC WORK EXPERIENCE

2014–Present Graduate Teaching Assistant, Computer Science Department, Ten¬nessee Tech University.

 

2013–2014 Adjunct Faculty, Mathematics Department, Tennessee Tech University.

2012–2013 Graduate Teaching Assistant, Mathematics Department, Tennessee Tech University.

2011–2012 Mathematics Tutor, Library Learning Commons, Tennessee Tech University.

NON-ACADEMIC WORK EXPERIENCE

2011–2014 Contract Programmer, Automate My Data, Algood, TN.

Wrote and maintained materials requirement planning software and billing systems. Work involved Java, Python, MySQL, Pervasive.SQL, HTML, CSS, and Java Servlets

TECHNICAL PROFICIENCIES

Proficient In

Languages C, C++, OpenMPI, OpenMP, Java, Python

Technologies BSD, Linux, MacOSX, LATEX

Familiar With

Languages C#, PHP, Lisp, SQL, XML, RDF/SPARQL, HTML, CSS, Java Script

Libraries OpenMPI, OpenMP

Technologies Apache Jena Fuseki, Stardog, MySQL, PostgreSQL, SQLite, Pervasive.SQL

RESEARCH INTERESTS

Computer Algorithms, Theory of Computation, Artificial Intelligence, Graph Based Science Anomaly Detection

Mathematics Graph Theory, Coding Theory, Abstract Algebra, Real Analysis Recreational Fractals, Strage Attractors

PUBLICATIONS

Journal Articles

William Eberle, Lawrence Holder, and Jeffrey Graves. Insider threat detection using a graph-based approach. Journal of Applied Security Research, 6(1):32–81, January 2011

Conference Papers

Sreenivas R. Sukumar, Larry W. Roberts, and Jeffrey A. Graves. A reasoning and hypothesis-generation framework based on scalable graph analytics. Proceedings of the Cray User Group Conference, 2016

 

Michael Okaro, Mohamed Abdelrahman, and Jeff Graves. Monitoring metal fill profile in lost foam casting process using capacitive sensors and metal fill time estimation. In Sensors Applications Symposium (SAS), 2011 IEEE, pages 76–81. IEEE, February 2011

William Eberle, Lawrence Holder, and Jeffrey Graves. Using a graph-based approach for discovering cybercrime. In Florida Artificial Intelligence Research Society Conference, May 2010

William Eberle, Lawrence Holder, and Jeffrey Graves. Detecting employee leaks using badge and network IP traffic. In IEEE Symposium on Visual Analytics Science and Technology, October 2009

Mike Rogers, Sheikh Ghafoor, and Jeff Graves. Braid: Distributed patient information for health care providers. In International Conference on Internet Computing, pages 132–138, July 2009

Workshop Papers

Ferrol Aderholdt, Jeffrey A. Graves, and Manjunath Gorentla Venkata. Parallelizing single source shortest path with OpenSHMEM. In OpenSH-MEM 2017: Fourth workshop on OpenSHMEM and Related Technologies, 2017

Rina Singh, Jeffrey A Graves, Sangkeun Lee, Sreenivas R Sukumar, and Mallikarjun Shankar. Enabling graph appliance for genome assembly. In Proceedings of the 2015 IEEE International Conference on Big Data, BIG DATA '15, pages 2583–2590, Washington, DC, USA, October 2015. IEEE Computer Society

Posters

Rina Singh, Jeffrey A. Graves, and Douglas A. Talbert. Complex patterns in dynamic attributed graphs. In Proceedings of the 25th International Conference Companion on World Wide Web, WWW '16 Companion, pages 105–106, Republic and Canton of Geneva, Switzerland, 2016. International World Wide Web Conferences Steering Committee

Mike Rogers, Sheikh Ghafoor, and Jeff Graves. A framework for inte¬grating electronic patient record systems. American Medical Informatics Association Annual Symposium, 2009

 

Preference-based evolutionary algorithm for airport

runway scheduling and ground movement

optimisation

 

Michal Weiszer

School of Engineering

University of Lincoln

Lincoln, United Kingdom

Email: mweiszer@lincoln.ac.uk

 

Jun Chen

School of Engineering

University of Lincoln

Lincoln, United Kingdom

Email: juchen@lincoln.ac.uk

 

Paul Stewart

Institute for Innovation

in Sustainable Engineering

Derby, United Kingdom

Email: p.stewart1@derby.ac.uk

 

 

Abstract—As airports all over the world are becoming more congested together with stricter environmental regulations put in place, research on optimisation of airport surface operations started to consider both time and fuel related objectives. However, as both time and fuel can have a monetary cost associated with them, this information can be utilised as preference during the optimisation to guide the search process to a region with the most cost efficient solutions. In this paper, we solve the integrated optimisation problem combining runway scheduling and ground movement problem by using a multi-objective evo¬lutionary framework. The proposed evolutionary algorithm is based on modified crowding distance and outranking relation which considers cost of delay and price of fuel. Moreover, the preferences are expressed in a such way, that they define a certain range in prices reflecting uncertainty. The preliminary results of computational experiments with data from a major airport show the efficiency of the proposed approach.

I. INTRODUCTION

With continuous growth of air traffic, which is predicted to carry double passengers in 2030 compared to 2013 [1], it is forecasted that without any action taken, many airports will become congested and air transportation as a whole will have a significant impact on the environment. As a result, research on management of aircraft movements on and in the proximity of airports attracted a lot of attention in general (e.g. [2], [3]) and airport surface operations in particular [4]. Specifically, the most research focused on optimisation of individual airport surface operations such as ground movement, runway schedul¬ing or gate assignment to maximise the utilisation of available resources.

The objective of runway scheduling problem is of¬ten expressed as a minimisation of delay, the number of changes compared to First-come-first-served (FCFS) sequence, makespan or their combination. A wide range of exact and heuristic methods has been employed to solve this problem including dynamic programming [5], hybrid tabu search [6], and genetic algorithms [7]. A detailed review of recent research on runway scheduling problem can be found in [8].

The objective of ground movement has been considered to be mainly the minimisation of the total taxi time or other time related objectives [4]. Approaches to solve the ground movement problem include integer programming [9], [10] or 

 

a graph-based approach utilised in [11], [12]. In addition to the total taxi time, a few researchers started to consider also fuel consumption during ground movement as an objective of the multi-objective optimisation problem [13]. Since runway scheduling and ground movement are interconnected problems, recently both problems have been integrated together, min¬imising time related objective [14]–[16] or both time and fuel consumption [17] in a multi-objective manner.

As some of the above-mentioned approaches use multi-objective optimisation to minimise distinct objectives such as time and fuel at the same time, the result of such approach is a set of non-dominated solutions, from which the decision maker (DM) has to implement one solution. In practice, often is the case that there exist some preferences beforehand, such as the price of fuel or cost of a single minute of delay. However, as argued in [18], utilising this information to convert the multi-objective problem into a single-objective one is con-traproductive. By finding only a single solution, DM is unable to investigate the properties of optimal/near-optimal solutions respecting higher-level preference information. Instead, this preference can be utilised during the optimisation to guide the search to an preferred region of interest (RoI) on the Pareto front. Research on incorporating preferences into multi-objective optimisation, whether in a-posteriori manner when the preferences are known before the search or in an interactive approach when preference information is iteratively obtained from the DM during the search, has been very active in the last years. Preferences are often expressed as reference points, goal vectors or aspiration levels corresponding to desired levels of objective values. The dominance relation is modified according to the distance to the reference point in [19] or aspiration level satisfaction in [20]. An achievement scalarizing function taking into account reference point is used to prefer some solutions closer to RoI in [21]–[24]. The original optimisation problem is modified by a weight distribution function in [25]. A binary preference in terms of linguistic variables is used in [26]; the crowding distance in Nondominated Sorting Genetic Algorithm-II (NSGA-II) [27] is changed in order to incorporate reference direction (weights) [28] or reference point [18], [23].

However, the desired values of objectives expressed as a reference point or aspiration levels are often unknown to the DM before the optimisation. As the result, an initial run of the optimisation algorithm is necessary to discover the whole

 

Pareto front from which the DM can subsequently choose a RoI. On the other hand, if the preferences are modelled as weights, the extent of RoI is usually defined by some non-intuitive parameter during the search as in [28].

In this paper, we solve the integrated optimisation problem combining runway scheduling and ground movement problem, similar as in [17], by using a multi-objective evolutionary opti¬misation (EMO) framework based on genetic algorithm (GA) and NSGA-II. The proposed algorithm employs a modified crowding distance presented in [23] and outranking relation of the Light Beam Search (LBS) method [29], which considers preference information in terms of cost of delay and price of fuel, guiding the search towards the most cost optimal RoI. Moreover, the preferences are expressed in a such way, that they give the DM a possibility to define a certain range of prices reflecting uncertainty.

The rest of the paper is organised as follows. Section II provides details about individual components of the integrated model, including runway scheduling and ground movement problem. The proposed EMO framework taking into account preferences is described in Section III. A set of computational experiments is carried out using data instances from Doha International Airport in Section IV. Finally, conclusions are drawn in Section V.

II. PROBLEM DESCRIPTION

This section provides a description of the integrated opti¬misation problem of airport ground operations combining the runway scheduling and ground movement problem similar to one introduced in [17].

A. Runway scheduling problem

The aim of runway scheduling in this paper is to find the optimal landing/take-off time of arriving/departing aircraft at the given runway minimising the delay and fuel burnt during waiting while respecting given safety constraints. In this paper, only take-off times are optimised whereas landing times are considered to be fixed, as from the practical point of view, it is easier to control taking-off aircraft still on the ground rather than airborne arriving aircraft.

The main constraint that limits the throughput of the runway is the minimum time interval between landing/taking-off aircraft. These minimum separations are due to wake vortices created by moving aircraft and in-flight separation constraints. In this paper, only separation due wake vortices is taken into account. The strength of wake vortices and thus separation depends on the aircraft type and is approximately proportional to its weight.

Let M = (A U D) be the set of total |M| = m arriving aircraft A and departing aircraft D. Let V (vi, vj) be the function to calculate the wake vortex separations from weight categories vi and vj of leading aircraft i and trailing aircraft j. The wake vortex separations used in this paper satisfy the triangle inequality V (vi, vj) + V (vj, ve) > V (vi, ve) for aircraft taking off in the order i, j, e and are given in Table I.

Let ri be the actual landing time for aircraft i E A and take¬off time for aircraft i E D. For arriving aircraft, ri is given. For departing aircraft, let di denote the time the departing 

 

TABLE I: Separations in seconds between departing (D) and arriving (A) flights for weight classes: Heavy (H), Large (L), Small (S).

Trailing

A-H A-L A-S D-H D-L D-S

A-H 96 157 207 60 60 60

Leading A-L A-S 60

60 69

69 123

82 60

60 60

60 60

60

D-H 60 60 60 96 120 120

D-L 60 60 60 60 60 60

D-S 60 60 60 60 60 60

aircraft i E D arrived at the runway holding point. Aircraft i E D can take-off immediately, i.e. di = ri if there is enough time elapsed from landing/take-off time ri1 of the previous aircraft i  1 to comply with separation given by V (vi, vi1), otherwise, the departing aircraft i has to wait at the runway holding point until it is safe to take-off:

ri =

Then, the waiting time wi of the departing aircraft i E D is equal to wi = di  ri.

The objective of the runway scheduling is to minimise the total runway delay trwy and the total runway fuel fr wy burned by aircraft while waiting to take-off which depends on the delay wi and idle fuel flow 0vi specified for the weight category vi:

wi, (1)

wi • 0vi. (2)

The idle fuel flow 0vi corresponds to fuel flow from the Inter¬national Civil Aviation Organization (ICAO) engine database for 5 % of full power thrust of the representative aircraft, as explained in Section II-B.

B. Ground movement problem

The aim of the ground movement problem is to find routes and schedules for aircraft taxiing from runway to gate/stand and vice versa in a time and fuel efficient manner, respecting routes of other aircraft while preventing conflicts between them.

In this paper, due to the simple layout of the airport, the problem of finding routes is reduced to the shortest path problem. The shortest paths between each gate/stand and runway exit points are pre-calculated beforehand. Then, for each path optimised speed profiles are found by a specialized heuristic described in [30]. The speed profiles are optimised in a multi-objective manner, minimising taxi time and fuel consumption for the given route. The resulting non-dominated speed profiles are stored in a look-up table, with a separate table for each weight category vi and then retrieved during the on-line optimisation in order to save computational time.

 

Given the route of the aircraft and time needed to travel from origin to destination depends on the chosen speed profile, some delay may be added in order to prevent conflicts between taxiing aircraft. The conflicts between aircraft are solved by adding a small buffer time such that aircraft always maintain a safe time distance δ = 12 s between them (which corresponds to approximately 62 m at taxiing speed 10 knots).

Let yi be an integer representing the speed profile of aircraft i from the Pareto front of efficient speed profiles retrieved from the look-up table for the shortest route qi from the runway to the stand si for arriving aircraft i E A, or vice versa for departing aircraft i E D. We define a function T(qi, yi) which returns travel time of aircraft i taxiing on route qi for the given speed profile yi, including a delay to prevent taxiing conflicts.

In order to retrieve the fuel consumption of aircraft i of weight category vi, for route qi, using speed profile yi, a function F(qi, yi, vi) is defined. The function F(qi, yi, vi) is based on the method introduced in [30], which employs physics-based equations taking into account the acceleration force and rolling resistance to calculate thrust. The calculated thrust and aircraft engines are then mapped into corresponding fuel flows according to the ICAO emissions database.

Then, the objective of the ground movement problem is to minimise the total taxi time ttaxi and the total fuel ftaxi burned during ground movement:

M

ttaxi = T(yi), (3)

i=1

M

ftaxi = F(yi). (4)

i=1

C. Integrated optimisation problem

The runway scheduling and ground movement problem are combined into the integrated bi-objective optimisation problem with the following objective functions:

min g1 = ttaxi + trwy, (5)

min g2 = ftaxi + frwy, (6)

where g1 corresponds to the total time and g2 is the fuel consumption to be minimised. The decision variables for this optimisation problem are the pushback time for departing aircraft xi E 1-300, ..., 300} and the speed profile yi E 11, ... ,12} for all aircraft. The pushback time xi represents an integer number of seconds before/after the baseline time given as input flight schedule when the aircraft starts taxiing. The value of speed profile yi determines which solution from the pre-computed look-up table of non-dominated speed profiles is going to be selected, ranging from 1 representing the most time-efficient (fastest) speed profile to 12 standing for the most fuel efficient one. The number of speed profiles depends on the heuristic used for their generation, for details see [30].

Given the decision variables, the objective function values g1, g2 are determined as follows. Firstly, aircraft are considered sequentially according to their initial sequence specified by the input flight schedule. For each arriving aircraft i E A a 

 

shortest route qi is retrieved from the look-up table between the runway and designated gate/stand si or in the opposite direction for departing aircraft i E D. The route qi, weight category vi and speed profile value yi determine which pre-computed speed profile is retrieved from the look-up table and used to schedule aircraft i along the route qi after all taxiing conflicts have been resolved. Then, the total taxi time ttaxi and the total fuel ftaxi are computed as stated in Section II-B. Based on taxi time T(yi), the runway holding point arrival time di is determined for each departing aircraft i E D. The delay wi is calculated as defined in section II-A and summed for all departing aircraft i E D to get the total runway delay trwy and subsequently the total fuel frwy. Finally, objectives g1, g2 are calculated according to 5 and 6.

D. Preferences based on costs

The bi-objective optimisation problem formulated in Sec¬tion II-C minimises two objective functions g1, g2 simulta¬neously. In practice, the total time g1 and fuel consumption g2 are associated with economic costs incurred by operations on the airport surface. For example, the fuel consumption g2 is connected to fuel cost defined by fuel price. Similarly, total time g1 incurs costs due to the fact, that each second of delay (i.e. unproductive time spent taxiing or waiting at runway) is an extra cost in terms of crew, maintenance, deprecation of aircraft, etc. Therefore, once costs related to g1, g2 are known, they can be used as weights during the optimisation, directing the search to a single cost-optimal solution. However, determining these cost, specifically cost of delay, is often difficult and only approximate as for example, crew wages or maintenance fees vary from airline to airline. Therefore, it might be more useful to guide the search to a region on the Pareto front containing solutions optimal for a certain cost range. We define a price vector c = [c1, ... , cn] which, in general, specifies cost for each one of n objectives. Furthermore, to include the uncertainty in costs, l boundary price vectors cB = [cB 1, ... , cB

n ] are defined. The number of boundary vectors depends on the DM. For example, DM can specify only upper bound for cost, both upper and lower bounds or different combinations in case of many objectives.

Then, each solution zj found during optimisation can be evaluated by function Ctotal(zj, c) in terms of total monetary cost, using price vector c:

Ctotal(zj, c) = En ck ' gk(zj). (7)

k =1

III. EMO FRAMEWORK WITH PREFERENCES

In order to optimise the objective functions g1, g2 of the integrated optimisation problem stated in Section II taking into account preferences given by costs, the following EMO framework is proposed in this section. Firstly, the DM is asked to provide a price vector c and l boundary price vectors cB as defined in Section II-D. Then, the search is divided into two parts:

1) A single objective optimisation search with aggre-gated objective function is performed by GA to quickly approach to the vicinity of the global Pareto front.

 

2) The best solutions from the single objective optimi 

sation part are seeded into a preference-based EMO algorithm to reach RoI of the global Pareto front.

The single-objective optimisation is carried out by the GA minimising total costs Ctotal as defined in 7. The single-objective search is further accelerated by taking advantage of known preferences during the search process. For each aircraft i, speed profile is fixed to a value yi  {1,12} with minimum Ctotal. As a result, the search space is restricted to decision variable xi promising faster convergence, however sacrificing some of the optimality.

A. Preference-based NSGA-II

For the second phase, the EMO algorithm proposed in this paper, denoted as Preference-based NSGA-II (P-NSGA-II), is based on Light Beam Search EMO algorithm [23] which is a derivative of NSGA-II [27]. The initial population of random solutions is seeded with a specified number of best solutions (set to 10 in this paper) found by GA during the first phase. In the second phase, P-NSGA-II considers all decision variables xi and yi thus finding potentially better solutions. The P-NSGA-II performs the usual EMO operations: selection, crossover, mutation, replacement for the specified number of generations similar to original NSGA-II. In order to incorporate preferences into EMO algorithm, the replacement procedure which selects the solutions surviving to the next generation is modified. The original replacement procedure in NSGA-II favours non-dominated solutions with large crowding distance. In P-NSGA-II, the replacement is carried out with modified crowding distance to prefer solutions closer to RoI.

During replacement procedure, non-domination ranking is carried out for all solutions in the population. Then, the middle point and characteristic neighbours are determined as depicted on Fig. 1. The middle point ZC is a solution Zj with minimum cost defined by price vector C:

ZC = arg min (Ctotal(Zj, C)). (8)

zj

The characteristic neighbour ZB, is a solution Zj for which the cost using the boundary price vector CB, is minimum:

ZB, = arg min (Ctotal(Zj, CB,)). (9)

zj

Subsequently, for each solution Zj, belonging to the same non-dominated front, crowding distance cdj is calculated:

 if Zj is middle point ZC,

M + dj else if Zj outranks ZC,

1/Ctotal(Zj, C) otherwise.

If the solution Zj is the middle point, it is assigned an infinite crowding distance, so it is always included in the next generation. Other solutions are categorized according to their outranking relation to ZC. A solution Zj is said to outrank middle point ZC (denoted as ZjSZC) if mv(ZC, ZX) = 0, where mv is defined as follows:

mv(ZC, ZX) = card{k : gk(ZX)gk(ZC)  vk, k = 1,.. ., n} (10) Solutions outranking the middle point ZC are considered to be as good as ZC and form the outranking neighbourhood. As

 

Fig. 1: Preferred region on Pareto front.

all solutions in outranking neighbourhood belong to the same non-dominated front, if solution Zj is better than ZC in some objectives, then it must be worse in at least one other objective. Solution ZjSZC if this deterioration in objective k is not larger than veto threshold vk. The veto threshold vk for objective k is determined by characteristic neighbours:

vk = max{gk(ZB,), l = 1, ... , n}. (11)

Solutions Zj in the outranking neighbourhood are assigned M + dj crowding distance, where M is a big positive integer, set to M = 106 in this work and dj is the original crowding distance as defined by Deb [27]:

. (12)

gmax

k  gmin

k

Where gmax

k is the maximum and gmin

k the minimum value of the objective function gk found so far. The big integer M ensures that crowding distance cdj is large enough to guarantee inclusion into the next generation whereas dj causes the preference of solutions in less crowded areas, resulting in a uniformly distributed solutions across the characteristic neighbourhood.

For remaining solutions, the crowding distance cdj is inversely related to Ctotal(Zj, C). As a result, solutions Zj with smaller Ctotal(Zj, C) and thus closer to the RoI are assigned larger crowding distance giving them larger chance to be included in the next generation.

IV. COMPUTATIONAL RESULTS AND DISCUSSION

A. Experimental setup

The proposed EMO framework was tested on a dataset of real arrival and departure flights on Doha International Airport (DOH) which was the largest airport in Qatar and a hub airport for Qatar Airways until the new Hamad International airport was completed in late April 2014. DOH airport has one runway and 55 stands. The data were recorded on 16th March 2014 and divided into two instances doh1 representing medium traffic conditions and doh2 for high traffic conditions. The instance doh1 includes 96 flights between 17:00 and 21:00 UTC from which 50 are arrivals and 46 departures. The instance doh2

 

TABLE II: Specifications of the representative aircraft.

Learjet 35A Airbus A320 Airbus A333

Take-off weight 8300 kg 78000 kg 230000 kg

Engines TFE731-2-2B CMF56-5-A1 CF6-80E1A2

Number of engines 2 2 2

Rated output R 2x15.6 kN 2x111.2 kN 2x287 kN

Rolling resistance 1221 N 11.48 kN 33.84 kN

Fuel flow at 7% R 0.024 kg-s1 0.101 kg-s1 0.228 kg-s1

Fuel flow at 30% R 0.067 kg-s1 0.291 kg-s1 0.724 kg-s1

TABLE III: Average IE indicator for 30 runs.

NSGA-II P-NSGA-II P-NSGA-II with seed

doh1 1.061 1.053 1.016

doh2 1.219 1.207 1.040

 

Fig. 2: Pareto fronts for doh1.

 

consists of 84 flights between 21:00 and 23:00 UTC from which there are 27 arrivals and 57 departures. The data provided specified landing/pushback times and gates/runway exits for each flight.

The aircraft have been divided into 3 groups according their wake vortex separation requirements as defined in [31]. For each category, a representative aircraft is designated and its specifications are used during the fuel consumption calculation. The specifications are summarized in Table II.

The computational experiments were performed on a com¬puter with an Intel i3-2120 processor and 3.16 GB of RAM, running Linux. The EMO framework is implemented using the Inspyred package for Python [32]. The termination criteria for EMO framework was set to 900 seconds, of which 600 seconds is allocated to GA and 300 seconds to P-NSGA-II. Based on initial experiments, the number of individuals in population for GA was set to 200 and 50 for P-NSGA-II. The price vector c is equal to [0.469, 0.71], where the first element corresponds to costs of delay as defined in [17] and the second element is the jet fuel price (as of 14.1.2014), both in Euro. The boundary vector cB1 = [1.2  0.469, 0.8  0.71] and cB2[0.8  0.469,1.2  0.71].

B. Computational results

The performance of the proposed two-phase EMO frame¬work (P-NSGA-II with seed) was compared to original NSGA-II and P-NSGA-II without the seed from single-objective GA. Fig. 2 shows the best Pareto fronts from 30 runs of the algorithms for doh1 instance. As can be seen, single-objective GA was able to find better solutions than NSGA-II and comparable with P-NSGA-II alone. Then, P-NSGA-II could further improve the solutions from the initial seed, focusing the search on RoI.

The multiplicative unary epsilon indicator IE is used as the performance index, expressing the distance of the resulting Pareto front to the global Pareto front, which was constructed by considering all solutions found during the experiments and leaving only non-dominated ones. Average IE indicator values for 30 runs of the algorithms are given in Table III.

 



Fig. 3: Averaged convergence for 30 runs for doh1 instance.

The results of IE show that P-NSGA-II alone could only marginally improve the performance compared to NSGA-II. However, adding seed to P-NSGA-II significantly helped the algorithm to find better solutions. The improving effect is even more evident for doh2 instance with high traffic conditions. We can hypothesize that beneficial impact of initial single-objective GA on the overall performance of the EMO framework will increase in case of many (i.e. more than 3) objectives as indicated in [22], due to the fact that many objectives resulting in high number of non-dominated solutions deteriorate the performance of EMO algorithms.

The convergence during the evolution for the doh1 is documented in Fig. 3. Single-objective GA in the first phase of the search quickly finds good solutions, helping the P-NSGA-II in the second phase obtain good values of I, Furthermore, fixing speed profiles yi for GA results in better values of IE for the initial population, effectively accelerating the search.

V. CONCLUSION

In this paper, a multi-objective evolutionary framework tak¬ing into account preferences in an a-priori manner is proposed to solve integrated optimisation problem of runway scheduling and ground movement on airport surface. The preference was incorporated in the first phase to transform the original multi-objective problem to a single-objective one to quickly approach

 

the region of interest. Then, in the second phase a modified crowding distance in NSGA-II was applied to finally arrive to Pareto front of cost-efficient solutions. The preliminary computational experiments conducted on real-world data from a major Asian airport showed promising results and indicate that the proposed approach could be suitable as an optimisation framework for decision support at the airport. The evolutionary framework utilising the preferences given as costs associated with objectives successfully accelerated the search. Moreover, the input values of costs enabled to control the extent of the resulting region and include some uncertainty about the prices. The incorporation of preferences helps the decision maker to focus his attention on only cost-effective solutions in a shorter computational time. For the future research, the approach needs to be validated on a more diverse set of problem instances. Also, the idea of including uncertainty into preference information deserves more attention. As an example, the uncertainty could be expressed in terms of fuzzy values.

REFERENCES

[1] ICAO. (2014) Annual Report of the ICAO Council: 2013 The World of Air Transport. [Online]. Available: http://www.icao.int/ annual-report-2013/Pages/the-world-of-air-transport-in-2013.aspx

[2] L. Bianco, P. Dell’Olmo, and S. Giordani, “Scheduling models for air traffic control in terminal areas,” Journal of Scheduling, vol. 9, no. 3, pp. 223–253, 2006. [Online]. Available: http: //dx.doi.org/10.1007/s10951-006-6779-7

[3] M. Sam`a, A. D’Ariano, and D. Pacciarelli, “Rolling horizon approach for aircraft scheduling in the terminal control area of busy airports,” Transportation Research Part E: Logistics and Transportation Review, vol. 60, no. 0, pp. 140–155, 2013. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S136655451300118X

[4] J. A. Atkin, E. K. Burke, and S. Ravizza, “The airport ground movement

problem: Past and current research and future directions,” Proceedings of the 4th International Conference on Research in Air Transportation (ICRAT), Budapest, Hungary, pp. 131–138, 2010.

[5] H. Balakrishnan and B. Chandran, “Scheduling aircraft landings under constrained position shifting,” in AIAA Guidance, Navigation, and Control Conference and Exhibit, Keystone, CO, 2006.

[6] J. A. D. Atkin, E. K. Burke, J. S. Greenwood, and D. Reeson, “Hybrid Metaheuristics to Aid Runway Scheduling at London Heathrow Airport,” Transportation Science, vol. 41, no. 1, pp. 90–106, 2007.

[7] X.-B. Hu and E. Di Paolo, “Binary-Representation-Based Genetic Algorithm for Aircraft Arrival Sequencing and Scheduling,” Intelligent Transportation Systems, IEEE Transactions on, vol. 9, no. 2, pp. 301– 310, June 2008.

[8] J. Bennell, M. Mesgarpour, and C. Potts, “Airport runway scheduling,” Annals of Operations Research, vol. 204, no. 1, pp. 249–270, 2013. [Online]. Available: http://dx.doi.org/10.1007/s10479-012-1268-1

[9] A. Mar´ın and E. Codina, “Network design: taxi planning,” Annals of Operations Research, vol. 157, no. 1, pp. 135–151, 2008.

[10] P. C. Roling and H. G. Visser, “Optimal Airport Surface Traffic Planning Using Mixed-Integer Linear Programming,” International Journal of Aerospace Engineering, vol. vol. 2008, p. 11, 2008.

[11] S. Ravizza, J. A. Atkin, and E. K. Burke, “A more realistic approach for airport ground movement optimisation with stand holding,” Journal of Scheduling, pp. 1–14, 2013.

[12] C. Lesire, “An Iterative A* Algorithm for Planning of Airport Ground Movements,” 19th European Conference on Artificial Intelligence (ECAI)/6th Conference on Prestigious Applications of Intelligent Sys¬tems (PAIS), Lisbon, Portugal, August 16-20, 2010.

[13] S. Ravizza, J. Chen, J. A. Atkin, E. K. Burke, and P. Stewart, “The trade-off between taxi time and fuel consumption in airport ground movement,” Public Transport, vol. 5, no. 1-2, pp. 25–40, 2013.

 

[14] R. Deau, J. Gotteland, and N. Durand, “Airport surface management and runways scheduling,” in Proceedings of the 8th USA/Europe Air Traffic Management Research and Development Seminar, Napa, CA, USA, 2009.

[15] G. Clare and A. Richards, “Optimization of Taxiway Routing and Runway Scheduling,” Intelligent Transportation Systems, IEEE Trans¬actions on, vol. 12, no. 4, pp. 1000–1013, Dec 2011.

[16] M. Frankovich and D. Bertsimas, “Air Traffic Flow Management at Airports: A Unified Optimization Approach,” in Tenth USA/EUROPE Air Traffic Management Research & Development Seminar, June 2013.

[17] M. Weiszer, J. Chen, and G. Locatelli, “An integrated optimisation approach to airport ground operations to foster sustainability in the aviation sector,” Applied Energy, 2015, in press. [Online]. Available: http://dx.doi.org/10.1016/j.apenergy.2015.04.039.

[18] K. Deb, J. Sundar, N. Udaya Bhaskara Rao, and S. Chaudhuri, “Refer¬ence point based multi-objective optimization using evolutionary algo¬rithms,” International Journal of Computational Intelligence Research, vol. 2, no. 3, pp. 273–286, 2006.

[19] L. Ben Said, S. Bechikh, and K. Gh´edira, “The r-dominance: a new dominance relation for interactive evolutionary multicriteria decision making,” Evolutionary Computation, IEEE Transactions on, vol. 14, no. 5, pp. 801–818, 2010.

[20] J. Molina, L. V. Santana, A. G. Hern´andez-D´ıaz, C. A. Coello Coello, and R. Caballero, “g-dominance: Reference point based dominance for multiobjective metaheuristics,” European Journal of Operational Research, vol. 197, no. 2, pp. 685–692, 2009.

[21] L. Thiele, K. Miettinen, P. J. Korhonen, and J. Molina, “A preference-based evolutionary algorithm for multi-objective optimization,” Evolu¬tionary Computation, vol. 17, no. 3, pp. 411–436, 2009.

[22] A. L´opez-Jaimes and C. A. C. Coello, “Including preferences into a multiobjective evolutionary algorithm to deal with many-objective engineering optimization problems,” Information Sciences, vol. 277, no. 0, pp. 1–20, 2014. [Online]. Available: http: //www.sciencedirect.com/science/article/pii/S0020025514004708

[23] K. Deb and A. Kumar, “Light beam search based multi-objective opti¬mization using evolutionary algorithms,” in Evolutionary Computation, 2007. CEC 2007. IEEE Congress on. IEEE, 2007, pp. 2125–2132.

[24] S. Huber, M. J. Geiger, M. Sevaux et al., “Interactive approach to the Inventory Routing Problem: computational speedup through focused search,” in Proceedings of the Logistikmanagement 2013 conference, 2013.

[25] T. Friedrich, T. Kroeger, and F. Neumann, “Weighted preferences in evolutionary multi-objective optimization,” in AI 2011: Advances in Artificial Intelligence. Springer, 2011, pp. 291–300.

[26] D. Cvetkovic and I. Parmee, “Preferences and their application in evolutionary multiobjective optimization,” Evolutionary Computation, IEEE Transactions on, vol. 6, no. 1, pp. 42–57, Feb 2002.

[27] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,” Evolutionary Computa¬tion, IEEE Transactions on, vol. 6, no. 2, pp. 182–197, Apr. 2002.

[28] J. Branke and K. Deb, “Integrating user preferences into evolutionary multi-objective optimization,” in Knowledge incorporation in evolution¬ary computation. Springer, 2005, pp. 461–477.

[29] A. Jaszkiewicz and R. Słowi´nski, “The ‘Light Beam Search’approach– an overview of methodology applications,” European Journal of Oper¬ational Research, vol. 113, no. 2, pp. 300–314, 1999.

[30] M. Weiszer, J. Chen, S. Ravizza, J. Atkin, and P. Stewart, “A heuristic approach to greener airport ground movement,” July 2014, pp. 3280– 3286.

[31] ICAO, Air Traffic Management: Procedures for Air Navigation Services, 2007, 15th edition. Doc 4444 ATM/501.

[32] A. Garrett, “Inspyred,” 2012, (Version 1.0) [software]. Inspired Intelli¬gence Initiative. Retrieved from http://inspyred.github.com.

 

Proceedings of the World Congress on Engineering 2009 Vol I

WCE 2009, July 1 - 3, 2009, London, U.K.

Software Fault Proneness Prediction Using

Support Vector Machines

Yogesh Singh, Arvinder Kaur, Ruchika Malhotra

 

Abstract— Empirical validation of software metrics to predict quality using machine learning methods is important to ensure their practical relevance in the software organizations. In this paper, we build a Support Vector Machine (SVM) model to find the relationship between object-oriented metrics given by Chidamber and Kemerer and fault proneness. The proposed model is empirically evaluated using public domain KC1 NASA data set. The performance of the SVM method was evaluated by Receiver Operating Characteristic (ROC) analysis. Based on these results, it is reasonable to claim that such models could help for planning and performing testing by focusing resources on fault-prone parts of the design and code. Thus, the study shows that SVM method may also be used in constructing software quality models.

Keywords: Metrics, Object-oriented, Software Quality, Empirical validation, Fault prediction, Support vector machine, Receiver Operating Characteristics analysis

I. INTRODUCTION

As the complexity and the constraints under which the software is developed are increasing, it is difficult to produce software without faults. One way to deal with this problem is to predict important software quality attributes such as fault-proneness, effort, testability, maintainability, and reliability during early phases of software development. The software metrics [4, 9, 10, 14, 15, 24, 25, 29-32, 41] may be used in predicting these quality attributes.

Manuscript received March 21, 2009.

Prof. Yogesh Singh is with Guru Gobind Singh Indraprastha

University, Delhi, India (email: ys66@rediffmail.com)

Dr. Arvinder Kaur is with Guru Gobind Singh Indraprastha

University, Delhi, India (e-mail:

arvinderkaurtakkar@yahoo.com.)

Ruchika Malhotra (Corresponding Author phone: 91-011-26431421) is with Guru Gobind Singh Indraprastha University, Delhi, India (email: ruchikamalhotra2004@yahoo.com)

 

The behaviour of several quantitative models ranging from using simple linear discriminant analysis to more complex logistic regression, decision tree, and SVM have been proposed. The regression and machine learning approaches are inherently different, raising the question to analyze the performance of these methods.

Several empirical studies have been carried out to predict the fault proneness models such as [1, 2, 5, 7, 11, 12, 14, 16, 20, 21, 23, 27, 34, 35, 40, 44]. There is a need to empirically validate the machine learning methods in predicting software quality attributes. Therefore, more data-based empirical studies that can be used to verify the capability of machine learning methods are needed. The evidence gathered through these empirical studies is considered to be the strongest support for testing a given hypothesis.

Thus, there is a need for both 1) empirically validating the results of machine leaning methods such as SVM and 2) finding the relation between OO metrics given by Chidamber and Kemerer [15] and fault proneness models. Now we briefly describe the work done in this study. In this paper, we investigate the following issue:

• Are the fault proneness models predicted using SVM method feasible and adaptable?

• How accurately and precisely do the OO metrics predict faults?

In order to perform the analysis we validate the performance of the SVM method using public domain KC1 NASA data set [42]. The 145 classes in this data were developed using C++ language. The contributions of this paper are summarized as follows: First, we performed the analysis of public domain NASA data set [42], therefore analyzing valuable data in an important area where empirical studies and data are limited. Second, we applied SVM method to predict the effect of OO metrics on fault proneness. To the best of our knowledge,

there has been no such

 

ISBN: 978-988-17012-5-1 WCE 2009

 

Proceedings of the World Congress on Engineering 2009 Vol I

WCE 2009, July 1 - 3, 2009, London, U.K.

 

previous research using SVM method to predict software fault proneness. The proposed results showed that SVM method predict faulty classes with high accuracy. However, since our analysis is based on only one data set, this study should be replicated on different data sets to generalize our findings.

The paper is organized as follows: Section 2 summarizes the metrics studied, and describes sources from which data is collected. Section 3 presents the overview of the SVM method. The results of the study are given in section 4. The model is evaluated in section 5. Conclusions of the research are presented in section 6.

II. RESEARCH BACKGROUND

In this section, we present the summary of metrics studied in this paper (Section A), and empirical data collection (Section B).

A. Dependent and Independent Variables

The binary dependent variable in our study is fault proneness. The goal of our study is to explore empirically the relationship between OO metrics 

 

and fault proneness. Fault proneness is defined as the probability of fault detection in a class [2]. We use machine learning methods to predict the probability of fault proneness. Our dependent variable will be predicted based on the faults found during software development life cycle. The metrics given by [15] are summarized in Table 1.

B. Empirical Data Collection

This study makes use of the public domain data set KC1 from the NASA Metrics Data Program [42]. The data in KC1 was collected from a storage management system for receiving/processing ground data, which was implemented in the C++ programming language. This system consists of 145 classes that comprise of 2107 methods, with 40K lines of code. KC1 provides both class-level and method-level static metrics. At the method level, 21 software product metrics based on product’s complexity, size and vocabulary are given. Five types of defects such as the number of defects and density of defects are also given. At the class level, values of 10 metrics are computed including seven metrics given by Chidamber and Kemerer [15]. These seven OO metrics are taken in our study (see Table 1) for analyses.

 

TABLE 1

METRICS STUDIED

Metric Definition

Coupling between Objects (CBO) CBO for a class is count of the number of other classes to which it is coupled and vice versa.

Lack of Cohesion (LCOM) For each data field in a class, the percentage of the methods in the class using that data field; the percentages are averaged then subtracted from 100%.

Number of Children (NOC) The NOC is the number of immediate subclasses of a class in a hierarchy.

Depth of Inheritance (DIT) The depth of a class within the inheritance hierarchy is the maximum number of steps from the class node to the root of the tree and is measured by the number of ancestor classes.

Weighted Methods per Class (WMC) A count of methods implemented within a class.

Response for a Class (RFC) A count of methods implemented within a class plus the number of methods accessible to an object class due to inheritance.

Source Lines Of Code

(SLOC) It counts the lines of code.

 

ISBN: 978-988-17012-5-1 WCE 2009

 

Proceedings of the World Congress on Engineering 2009 Vol I

WCE 2009, July 1 - 3, 2009, London, U.K.

 

III. MODEL PREDICTION USING SUPPORT VECTOR MACHINE (SVM) METHOD

SVM are useful tools for performing data classification, and have been successfully used in applications such as face identification, medical diagnosis, text classification [43], pattern recognition [13], chinese character classification [45], and identification of organisms [33]. SVM constructs an N-dimensional hyperplane that optimally separates the data set into two categories. The purpose of SVM modeling is to find the optimal hyperplane that separates clusters of vector in such a way that cases with one category of the dependent variable on one side of the plane and the cases with the other category on the other side of the plane [37]. The support vectors are the vectors near the hyperplane. The SVM modeling finds the hyperpalne that is oriented so that the margin between the support vectors is maximized. When the points are separated by a nonlinear region, SVM handles this by using a kernel function inorder to map the data into a different space when a hyperplane can be used to do the separation. Details on SVM can be found in [17, 18].

The recommended kernel function is the Radial basis Function (RBF) [37]. Thus, we used RBF function in SVM modeling to predict faulty classes in this study. The RBF kernel maps non-linearly data into a higher dimensional space, so it can handle non linear relationships between the dependent and the independent variables. Figure 1 shows the RBF kernel. One category of the

 

Figure 1: Radial basis function

other as triangles. The shaded circles and rectangles are support vectors.

 

dependent variable is shown as rectangles and the

Given a set of (xi, yi), ,(xm, ym) and

yi {1,+1} training samples. αi =(i=1, ,m) is

a lagrangian multipliers. K(xi,yi) is called a kernel function and b is a bias. The discriminant function D of two class SVM is given below [45]:

m

D(x) ==yiαiK(xi, x) + b (1)

i 1

Then an input pattern x is classified as [45]:

+1 if D(x)>0 x =  1 if D(x) < 0  (2)

The performance of the models predicted was evaluated using sensitivity, specificity, precision, completeness, and Area Under the Curve (AUC). Details on these measures can be found in [19]. The ROC curve, which is defined as a plot of sensitivity on the y-coordinate versus its 1-specificity on the x coordinate, is an effective method of evaluating the quality or performance of the predicted models [22].

IV. ANALYSIS RESULTS

This section presents the analysis results, following the procedure described in Section 3. The results of the model predicted and evaluated are presented.

Results of prediction model: Table 2 shows the sensitivity, specificity, precision, completeness, and cutoff point of each metric and the model predicted. SLOC and CBO metrics have the highest values of the sensitivity and completeness. However, in case of NOC metric all the classes were predicted to be non faulty hence the results are not shown, as testing all the classes will be a high waste of the testing resources.

The model was applied to 145 classes and Table 3 presents the results of correctness of the fault proneness model predicted. As shown in Table 2, the cut off point for the model build to predict fault proneness is 0.44. Out of 59 classes, actually fault prone, 45 classes were predicted to be fault prone. The sensitivity of the model is 76.27 percent. Similarly, 70 out of 86 classes were predicted not to be fault prone. Thus, the specificity of the model is 81.39 percent. The completeness of the model is

85.66 percent. Thus,

 

ISBN: 978-988-17012-5-1 WCE 2009

 

Proceedings of the World Congress on Engineering 2009 Vol I

WCE 2009, July 1 - 3, 2009, London, U.K.

 

the accuracy of the model is very high.

V. MODEL EVALUATION USING ROC ANALYSIS

The accuracy of the models predicted is somewhat optimistic since the models are applied on same data set from which they are derived. To predict accuracy of the model it should be applied on different data sets thus we performed 10-cross validation of the SVM model. Details on cross validation can be found in [39]. Table 4 summarizes the results of 10-cross validation of models using the SVM method. The AUC of the model predicted is 0.89.

In line with other predictive models, likewise findings of this study need to be externally validated. Although regression analysis is widely 

 

used method in literature, our results show that performance of the SVM model is good. In a previous study, we validated SVM using open source data set [38]. In [38], results also showed good performance of the SVM method. Therefore, it appears that the model predicted using machine-learning methods might lead to development of optimum software quality models for predicting fault prone classes.

Planning and resource allocating for inspection and testing is difficult and it is usually done on empirical basis. The model predicted in the above section could be of great help for planning and executing testing activities. To illustrate how the prediction model can be applied in practice, consider Figure 2. The values for the predicted fault proneness were taken from the results of validation of the models. On X-axis, we plot

 

TABLE 2

SENSITIVITY, SPECIFICITY, PRECISION, AND COMPLETENESS

Metric Sensitivity Specificity Precision Completeness Cutoff

CBO 76.27 68.6 71.7 82.24 0.40

WMC 55.93 65.1 61.37 71.18 0.41

RFC 52.54 66.27 60.68 67.13 0.37

SLOC 62.7 75.58 70.34 76.32 0.32

LCOM 59.32 63.95 62.75 65.88 0.33

NOC - - - - -

DIT 71.18 29 46.2 78.34 0.38

Model IV 76.27 81.39 78.62 85.66 0.44

TABLE 3

PREDICTED CORRECTNESS OF MODEL

Predicted

Po<=0.43 Po>0.43

Observed Not faulty Faulty

Not faulty 70 16

Faulty 14(92) 45(550)

TABLE 4

. RESULTS OF 10-CROSS VALIDATION OF MODELS

SUPPORT VECTOR MACHINE

Model

Cutoff point 0.45

Sensitivity 69.49

Specificity 82.55

Precision 77.24

Completeness 79.59

AUC 0.89

ISBN: 978-988-17012-5-1 WCE 2009

 

Proceedings of the World Congress on Engineering 2009 Vol I WCE 2009, July 1 - 3, 2009, London, U.K.

 

the probability of classes sorted in increasing order of their cutoff points. On Y-axis each column presents two lines. The first and second line shows the sensitivity and completeness of the predicted model, respectively. For example, in Figure 2 at X=0, the sensitivity of the model is 100% and the completeness is 100%.

VI. CONCLUSIONS

The goal of our research is to empirically analyze the performance of the SVM method. We find the individual and combined effect of each metric on fault proneness of classes. Based on public domain NASA data set KC1, we empirically analyzed the performance of the SVM method using ROC analysis.

We analyzed OO design metrics given by Chidamber and Kemerer. Our main results are as follows:

• The CBO, RFC, and SLOC metrics

were found to be related to fault proneness. NOC and DIT metric was not found to be significantly related to fault proneness. Thus, the results of machine learning method (SVM) show that usefulness of NOC and DIT metric is poor.

• The SVM method yielded good AUC

using ROC analysis. This study confirms that construction of the SVM models is feasible, adaptable to OO systems, and useful in predicting fault prone classes. While research continues, practitioners and researchers may apply machine learning methods for constructing the model to predict faulty classes.

As in all empirical studies, the relationship we established is valid only for certain population of systems. In this case, the authors can roughly characterize this population as “object-oriented, large-sized systems.”

We plan to replicate our study to predict the models based on machine learning algorithms such as genetic algorithms. We will also focus on cost benefit analysis of models that will help to determine whether a given fault proneness model would be economically viable.

 

Figure 2: Sensitivity and Completeness of the

Model

REFERENCES

[1] Aggarwal K.K., Singh Y., Kaur A., Malhotra R. (2007). Application of Artificial Neural Network for Predicting Fault Proneness Models, International Conference on Information Systems, Technology and. Management (ICISTM 2007), March 12-13, New Delhi, India.

[2] Aggarwal K.K., Singh Y., Kaur A., Malhotra R. (2008). Empirical Analysis for Investigating the Effect of Object-Oriented Metrics on Fault Proneness: A Replicated Case Study, Forthcoming in Software Process Improvement and Practice, Wiley.

[3] Aggarwal K.K., Singh Y., Kaur A., Malhotra R. (2006). Empirical study of object-oriented metrics. Journal of Object Technology, 5(8), 149-173.

[4] Aggarwal K.K., Singh Y., Kaur A., Malhotra R. (2005). Software reuse metrics for object-oriented systems. In Proceedings of the Third ACIS Int’l Conference on Software Engineering Research, Management and Applications (SERA ’05), 48-55.

[5] Aggarwal K.K., Singh Y., Kaur A., Malhotra R. (2006). Investigating the Effect of Coupling Metrics on Fault Proneness in Object-Oriented Systems. Software Quality Professional, 8(4), 4-16.

[6] Barnett V., Price T. (1995). Outliers in Statistical Data. John Wiley & Sons.

[7] Basili,V., Briand, L., and Melo, W. (1996). A validation of object-oriented design metrics as quality indicators. IEEE Transactions on Software Engineering, 22(10), 751¬761.

[8] Binkley, A., and Schach, S. (1998). Validation of the coupling dependency metric as a risk predictor. In Proceedings of the International Conference on Software Engineering, 452-455.

[9] Briand, L., Daly, W., and Wust J. (1998). Unified framework for cohesion measurement in object-oriented systems. Empirical Software Engineering, 3(1), 65-117.

[10] Briand, L., Daly, W., and Wust J. (1999). A unified framework for coupling measurement in object-oriented systems. IEEE Transactions on Software Engineering , 25(1), 91-121.

[11] Briand, L., Daly, W., and Wust J. (2000). Exploring the relationships between design measures and software quality. Journal of Systems and Software, 51(3), 245-273.

[12] Briand L., Wüst J., Lounis H. (2001). Replicated Case Studies for Investigating Quality Factors in Object-Oriented Designs, Empirical Software Engineering: An International Journal , 6(1), 11-58.

[13] Burges, C. (1998). A tutorial on support vector machines for pattern recognition. Data Mining and Knowledge Discovery 2, 121–167.

 

ISBN: 978-988-17012-5-1 WCE 2009

 

Proceedings of the World Congress on Engineering 2009 Vol I WCE 2009, July 1 - 3, 2009, London, U.K.

 

[14] Cartwright, M., and Shepperd, M. (1999). An empirical investigation of an object-oriented software system. IEEE Transactions of Software Engineering, 26(8), 786-796.

[15] Chidamber, S., and Kamerer, C. (1994). A metrics suite for object-oriented design. IEEE Transactions on Software Engineering , 20(6), 476-493.

[16] Chidamber, S., Darcy, D., and Kemerer, C. (1998). Managerial use of metrics for object-oriented software: An exploratory analysis. IEEE Transactions on Software Engineering, 24(8), 629-639.

[17] Cortes, C., Vapnik, V. (1995). Support-vector networks. Machine Learning 20, 273–297.

[18] Cristianini, N., Shawe-Taylor, J. (2000). An Introduction to Support Vector Machines and Other Kernel-based Learning Methods. Cambridge University Press, Cambridge, UK.

[19] El Emam, K., Benlarbi, S., Goel, N. And Rai, S. 1999. A Validation of Object-Oriented Metrics, Technical Report ERB-1063, NRC.

[20] El Emam, K., Benlarbi, S., Goel, N., and Rai, S. (2001). The Confounding Effect of Class Size on The Validity of Object-Oriented Metrics. IEEE Transactions on Software Engineering, 27(7), 630-650.

[21] Gyimothy, T., Ferenc, R., Siket, I. (2005). Empirical validation of object-oriented metrics on open source software for fault prediction, IEEE Trans. Software Engineering, 31 (10), 897 – 910.

[22] Hanley, J., McNeil, BJ. (1982). The meaning and use of the area under a Receiver Operating Characteristic ROC curve. Radiology, 143: 29-36.

[23] Harrison, R., Counsell, S. J., and Nithi, R.V. (1998). An evaluation of MOOD set of object-oriented software metrics. IEEE Transactions on Software Engineering, 24(6), 491-496.

[24] Henderson-Sellers, B. (1996). Object-oriented metrics, measures of complexity. Englewood Cliffs, N.J.: Prentice Hall.

[25] Hitz, M., and Montazeri, B. (1995). Measuring coupling and cohesion in object-oriented systems. In Proceedings of the International Symposium on Applied Corporate Computing, Monterrey, Mexico.

[26] Hosmer, D., and Lemeshow, S. (1989). Applied logistic regression. New York: John Wiley and Sons.

[27] Khoshgaftaar, T.M., Allen, E.D., Hudepohl, J.P, Aud, S.J. (1997). Application of neural networks to software quality modeling of a very large telecommunications system," IEEE Transactions on Neural Networks, 8(4), 902-909.

[28] Kothari, C. R. (2004). Research Methodology. Methods and Techniques. New Delhi: New Age International Limited.

[29] Lake, A., and Cook, C. (1994). Use of factor analysis to develop OOP software complexity metrics. In Proceedings of the 6th Annual Oregon Workshop on Software Metrics, Silver Falls, Oregon.

 

[30] Lee, Y., Liang, B., Wu, S., and Wang, F. (1995). Measuring the coupling and cohesion of an object-oriented program based on information flow. In Proceedings of the International Conference on Software Quality, Maribor, Slovenia.

[31] Li, W., and Henry, S. (1993). Object-oriented metrics that predict maintainability. Journal of Systems and Software, 23(2), 111-122.

[32] Lorenz, M., and Kidd, J. (1994). Object-oriented software metrics. Englewood Cliffs, N.J.: Prentice-Hall.

[33] Morris, C., Autret, A., Boddy, L. (2001). Support vector machines for identifying organisms-a comparison with strongly partitioned radial basis function networks. Ecological Modeling 146, 57–67.

[34] Olague, H., Etzkorn, L., Gholston, S., and Quattlebaum, S. (2007). Empirical Validation of Three Software Metrics Suites to Predict Fault-Proneness of Object-Oriented Classes Developed Using Highly Iterative or Agile Software Development Processes. IEEE Transactions on software Engineering, 33(8), 402-419.

[35] Pai, G. (2007). Empirical analysis of Software Fault Content and Fault Proneness Using Bayesian Methods, IEEE Transactions on software Engineering, 33(10), 675-686.

[36] Porter, A., Selly, R. (1990). Empirically guided Software Development using Metric-Based Classification Trees, IEEE Software, 7(2), 46-54.

[37] Sherrod, P. (2003) DTreg Predictive Modeling Software.

[38] Singh, Y., Arvinder, K., Malhotra, R. (2009). Application of Support Vector Machine to Predict Fault Prone Classes, ACM SIGSOFT Software Engineering Notes, 34( 9), 1-6.

[39] Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. J. Royal Stat. Soc., 36, 111-147.

[40] Tang, M.H, Kao,, M.H., and Chen, M.H. (1999). An Empirical Study on Object-Oriented Metrics, In Proceedings of Metrics, 242-249.

[41] Tegarden, D., Sheetz, S., and Monarchi, D. (1995). A software complexity model of object-oriented systems. Decision Support Systems 13 (3-4), 241-262.

[42] www.mdp.ivv.nasa.gov, NASA Metrics data Repository.

[43] Wang, X., Bi, D., and Wang, S. (2007). Fault recognition with Labeled multi-category’, Third conference on Natural Computation, Haikou, China.

[44] Yuming, Z. and Hareton, L. (2006). Empirical analysis of Object-Oriented Design Metrics for predicting high severity faults. IEEE Transactions on Software Engineering, 32(10), 771-784.

[45] Zhao, L., Takagi, N. (2007). An application of Support vector machines to Chinese character classification problem. IEEE International Conference on systems, Man and Cybernetics, Montreal.

 

ISBN: 978-988-17012-5-1 WCE 2009

 

  Communications of the

A I S

ssociation for nformation ystems

Panel Report DOI: 10.17705/1CAIS.043XX ISSN: 1529-3181

How do Machine Learning, Robotic Process

Automation, and Blockchains Affect the Human Factor

in Business Process Management?

Wirtschaftsuniversität Wien, Vienna

Austria

jan.mendling@wu.ac.at

Signavio IBM Research

Germany USA

Vrije Universiteit Amsterdam Data61, CSIRO

The Netherlands Australia

This paper summarizes a panel discussion at the 15th International Conference on Business Process Management. The panel discussed to what extent the emergence of recent technologies including machine learning, robotic process automation, and blockchain will reduce the human factor in business process management. The panel discussion took place on 14 September, 2017, at the Universitat Politècnica de Catalunya in Barcelona, Spain. Jan Mendling served as a chair; Gero Decker, Richard Hull, Hajo Reijers, and Ingo Weber participated as panelists. The discussions emphasized the impact of emerging technologies at the task level and the coordination level. The major challenges that the panel identified relate to employment, technology acceptance, ethics, customer experience, job design, social integration, and regulation.

Business Process Management, Process Automation, Artificial Intelligence, Machine Learning, Robotic Process Automation, Blockchain.

[Department statements, if appropriate, will be added by the editors. Teaching cases and panel reports will have a statement, which is also added by the editors.]

[Note: this page has no footnotes.]

This manuscript underwent peer review. It was received 12/09/2017 and was with the authors for 1 month for 1 revision. Christoph Peters served as Associate Editor.

Volume 43 Paper XXX pp. 1 – 23 June 2018

 

2 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

1 Introduction

The business process management (BPM) discipline investigates methods and techniques to organize business processes in an efficient and effective manner (Dumas, La Rosa, Mendling, & Reijers, 2013). A key idea of BPM involves improving business processes by redesigning information systems to best support the people who are working in the process. Indeed, many early office automation systems (Hirschheim, 1985), workflow systems (van der Aalst & van Hee, 2004), and various more recent process-aware information systems (Dumas, van der Aalst, & ter Hofstede, 2005)—which researchers often subsume under the term BPM systems (dumas et al., 2013)—all focus on this idea. Such systems hold and provide information to workers, schedule and coordinate specific pieces of work, and support decisions on how to best proceed.

Recent advancements in the area of artificial intelligence, machine learning, cryptography, and distributed systems have provided the foundations for new technologies, including robotic process automation (Aguirre & Rodriguez, 2017), chatbots (Shawar & Atwell, 2007), self-driving cars (Daily, Medasani, Behringer, & Trivedi, 2017), smart objects (Beverungen, Müller, Matzner, Mendling, & vom Brocke, 2017), blockchains (Nakamoto, 2008), and the Internet of things (Atzori, Iera, Morabito, 2010). Several recent papers discuss the implications of the emergence of these technologies for BPM (e.g., Beverungen et al., 2017; Mendling et al., 2017; Oberländer, Röglinger, Rosemann, & Kees, 2017). These technologies will likely affect how organizations design and execute business processes in the future. However, it is not clear in which specific way they will change BPM.

This paper summarizes the research background and the major arguments of a panel discussion at the 15th International Conference on Business Process Management. The panel discussed to what extent the emergence of recent technologies including machine learning, robotic process automation, and blockchain will reduce the human factor in business process management. As Shazia Sadiq highlighted, these technologies have a broad potential to affect BPM; however, it is not clear whether this impact will yield a peaceful decentralization (Star Trek scenario) or of darkness and extinction (Terminator scenario). Thus, this paper also contributes to our understanding of what impact these emerging technologies will have on the way processes are designed.

The paper proceeds as follows. In Section 2, we overview BPM and summarize research that discusses the impact of technology on business processes. In Section 3, we sketch some of the emerging technologies and investigate their impact at the task level and the coordination level of business processes. In Section 4, we discuss challenges and opportunities for research. We provide an edited transcript of the panel discussed in Appendix A.

2 Business Process Management and Technological Impact

In this section, we overview BPM with the BPM lifecycle’s assistance. New technologies allow one to design processes in novel ways. With reference to the redesign phase of this lifecycle, we discuss how technology affects the way how one can improve processes.

2.1 BPM Lifecycle

The BPM lifecycle model describes how the different management activities associated with BPM relate to one another. At the single process level, the lifecycle has five different phases: process discovery, process analysis, process redesign, process implementation, and process monitoring (see Figure 1) (Dumas et al., 2013). At its heart, the model illustrates how one can organize a BPM project or a BPM initiative such that it arrives at an improved process.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 3

Discovery

Implementation

The BPM lifecycle starts with the process discovery phase. It focuses on one specific process. This phase focuses on producing detailed descriptions of a business process as it currently exists (i.e., the “so-called” or “as-is” process model). During process analysis, one applies analytical tools and techniques in order to determine a business process’s current weaknesses. Process redesign addresses the most important weaknesses and yields a reworked design of the process (i.e., a “to-be” process model). One subsequently uses this model as the basis for process implementation. Process implementation refers to the various steps to put the to-be process into operation, such as implementing information systems and measures to facilitate organizational change. In the process-monitoring phase after one has implemented the redesigned process, one continuously collects and analyzes execution data for their compliance with performance and conformance objectives. Failing to meet objectives or changes of the goals and the business environment can trigger new iterations of the BPM lifecycle.

Subjecting a business process to the management activities of the BPM lifecycle can lead to improvements at the task and coordination levels. An organization achieves improvements at the task level when it improves the duration, the costs, the quality, or the flexibility of a singular task. An organization achieves improvements at the coordination level when the overall organization of handoffs between the tasks leads to faster processing, lower costs, better quality, or more flexibility. Some indications suggest that striving for improvements at the coordination level might have a relatively stronger impact on process performance than improving singular tasks. Blackburn (1992) investigated the flow-time efficiency of business processes in various industries and found that the cycle time of most business processes contains more than 95 percent of waiting time. At least for speeding up a business process, this finding means that reducing the waiting time between the tasks (coordination level) is more likely to improve flow-time efficiency than reducing the processing time of individual tasks (task level). One needs to keep this finding in mind when we discuss the impact of specific technologies on a business process: the technology might have a dominant impact at the task or the coordination level.

2.2 Technological Impact on Business Processes

New technologies affect how organizations execute and coordinate tasks in a process. Thus, one can see a new technology’s impact most visibly in the redesign phase of the BPM lifecycle and, in particular, in specific redesign heuristics. Reijers and Mansar (2005) present an extensive list of such heuristics. Many of these heuristics explicitly refer to information technology as a means to achieve process improvements. For instance, the task automation heuristic suggests that one should take an existing task and subject it to automation. This heuristic relates to the task level. This heuristic ideally produces a faster, cheaper, and more accurate execution of the task. The interfacing heuristic represents another example. It incorporates the idea that organizations can use standardized interfaces to integrate their operations with partners’ and customers’ information systems in order to make processes faster and more reliable. This heuristic impacts the coordination level more strongly than the task level. These heuristics describe two examples of information technology that affect business processes.

The 1990s saw a strong wave of business process reengineering (Hammer & Champy 1993) together with major investments in information technology newly introduced to the market back then. At the same time, researchers, including Brynjolfsson (1993), observed a productivity paradox of information technology.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

4 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

Apparently, investments in information technology did not always lead to productivity gains. Some of the works that have tried to resolve this paradox demonstrate that productivity gains from information technology investments require organizations to change their business processes in order to reap the potential benefits (Mukhopadhyay, Rajiv, & Srinivasan, 1997; Grover, Teng, Segars, & Fiedler 1998). From the perspective of new technology, Mooney, Gurbaxani, and Kraemer (1996) distinguish automational effects, informational effects, and transformational effects. Automational effects emerge when an organization uses a new technology to automate tasks that it previously did manually or with partial system support. Informational effects materialize from better tracking, monitoring, and analytical insights. Transformational effects relate to the changes in the mechanisms of coordination, which include disintermediation, outsourcing, or offshoring.

3 Emerging Technologies

In this section, we focus on three specific technologies that might affect business processes’ potential to automate tasks and facilitate new ways of coordination: machine learning, robotic process automation, and blockchains. We briefly sketch their central characteristics and point to more detailed references.

3.1 Machine Learning

Machine learning is a branch of the artificial intelligence research area. One prominent category of machine-learning applications is classification (Bishop, 2006). One can find classification tasks in various domains that require expert judgment, such as in healthcare (e.g., determining if someone has a tumor), law (e.g., determining whether to sentence someone for a crime), or construction (e.g., determining whether a certain construction would be stable). With the availability of big data in certain application domains, the potential of applying machine learning for classification has also increased. For instance, Sim (2016) emphasizes the importance of technologies such as IBM Watson for various diagnostic tasks in a medical context. One also needs big data to train machine-learning techniques such that they can provide accurate classification results. In this way, machine learning has the potential to partially automate a broader spectrum of tasks that experts have conducted in the past. It might also help to coordinate different tasks in a business process. In the context of BPM, these observations raise the questions for which specific application scenarios machine learning can be effectively devised and which type of training data is required to make it useful in a practical setting.

3.2 Robotic Process Automation

Robotic process automation (RPA) is an industrial response to the huge amount of manual work that individuals perform on a daily, weekly, or monthly basis to support a broad array of high-volume business processing (Aguirre & Rodriguez, 2017, Lacity & Willcocks, 2016). RPA is mostly associated with the task level. The application areas include finance and accounting, IT infrastructure maintenance, and front-office processing. The so-called robots are software programs that interact with systems such as enterprise resource planning and customer relationship management systems. The robots can gather data from systems and update them by imitating manual screen-based manipulations. From a business perspective, RPA solutions are appealing because they automate repetitive tasks while being minimally invasive into the overall processing that they support. An increasing number of organizations have begun to adopt RPA solutions recently; however, this growth might diminish in the future when the next generation of enterprise resource planning systems and IT infrastructure directly incorporates services for accessing data and making updates. RPA raises interesting academic research questions such as how to design and program robots and to integrate them with BPM systems, how to leverage RPA as a vehicle to support AI-enhanced processes, and how to use artificial intelligence techniques to program RPA solutions based on goals.

3.3 Blockchain

Blockchain, the technology underlying crypto-currencies such as Bitcoin, is a distributed ledger technology that enables organizations to engage in transactions without the need for a commonly trusted authority. It is a promising technology at the coordination level and a potential infrastructure for facilitating inter-organizational business processes. Its key strength is that it supports transactions between parties that do not trust each other over a computer network in which trust emerges from a combination of peer-to-peer technologies, consensus making, cryptography, and market mechanisms. Smart contracts are user-definable programs that the network of computer nodes in a blockchain executes. With the addition of these smart contracts, one can design the control logic between transactions in order to meet a diverse set of use

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 5

cases that span the financial industry, logistics and supply chains, healthcare, sharing economy, and many more. Proposals from the BPM research community include using smart contracts to express processes, particularly inter-organizational ones, in an imperative form such as BPMN (Weber et al., 2016), in an artifact-centric form (Hull et al., 2016), or in a rule form (Mery & Selman, 2017). In this way, large parts of the business logic of inter-organizational business processes can be compiled from process models into smart contracts to ensure that the joint process is correctly executed. Executing inter-organizational business processes using smart contracts on a blockchain can remove several barriers (Weber et al., 2016, Mendling et al., 2018). First, the blockchain can serve as an immutable public ledger, so that participants can review a trustworthy history of messages. Second, smart contracts can offer independent process monitoring from a global viewpoint. Third, encryption can ensure that data relevant for making decisions is visible while the remaining data is only visible to the process participants that require it. Blockchain technology raises interesting research questions such as how to devise novel execution and monitoring systems for inter-organizational business processes, how to define appropriate mechanisms for process evolution and adaptation, and how to identify patterns of redesigning processes using blockchain technology.

4 Impact of Emerging Technologies on Business Process Management

In this section, we discuss the impact of the various emerging technologies on BPM. We first discuss the impact at the task level and then at the coordination level.

4.1 Impact at the Task Level

To determine the impact of the various technologies, we follow Autor (2015) and distinguish three different types of tasks:

1. Routine tasks are explicit and codifiable. They include the calculations involved in bookkeeping; the retrieving, sorting, and storing of structured information in association with clerical work; and the precise execution of repetitive physical operations in a stable environment.

2. Abstract tasks require problem-solving capabilities, intuition, creativity, and persuasion. Tasks of this kind are typically associated with professional, technical, and managerial occupations. They require employees with a high degree of education and analytical capabilities. They emphasize inductive reasoning, communication, and professional expertise in open and underspecified contexts.

3. Manual tasks require situational adaptability, visual and language skills, and personal interactions. Manual tasks typically characterize food preparation and service jobs, cleaning and janitorial work, grounds cleaning and maintenance, health assistance, and jobs in security and protection services.

Emerging technologies will likely strongly impact routine tasks since they often provide the potential to benefit from what Mooney et al. (1996) call automational effects. As such, these technologies could displace workers in these routine tasks because they follow precise, well-understood procedures that can be either codified or mimicked. Machine learning and RPA will likely contribute to this trend. The panel discussion highlighted data entry and data validation as examples of routine tasks increasingly replaced by automatic solutions that companies such as Parlamind provide. As a consequence, we might observe a substantial decline in employment in clerical and administrative support.

Emerging technologies could also have a strong impact on abstract tasks. For these tasks, we might see informational effects. The panel discussion emphasized that systems already yield much better results for tasks such as diagnosing skin cancer. A serious challenge for these tasks involves the trust in the correctness and accuracy of the solutions. The panel described one example: the process of assessing whether to grant an asylum application or not. German authorities trialed a prototype system that could have sped up the application processing drastically; however, the country did not end up using it because decision makers lacked trust in its accuracy.

Emerging technologies can also have a transformational effect on manual tasks. Applications related to the Internet of things, Industry 4.0, and the industrial Internet contribute to these developments. For example, in a classical picking process in a warehouse, workers pick the products from their respective positions. Companies such as Amazon have introduced picking robots that connect to the order information systems.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

6 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

In this way, the company has transformed the picking process from manual work to work that machines perform.

One generally finds jobs intensive in either abstract or manual tasks at opposite ends of the occupational skill spectrum: professional, managerial, and technical jobs are on the one end and service and laborer jobs on the other. The computerization of routine tasks likely leads to the simultaneous growth of high-education, high-wage jobs at one end and low-education, low-wage jobs at the other end. Both developments will take place at the expense of middle-wage, middle-education jobs—a phenomenon that Goos and Manning (2003) call “job polarization”. Various economic studies at different levels of abstraction have also confirmed this phenomenon (Frey & Osbourne, 2017). However, the panel emphasized that jobs with routine tasks will continue to exist because the emerging technologies still have too many limitations. Currently, they are not profitable for tasks not highly standardized.

4.2 Impact at the Coordination Level and Work Organization

While emerging technologies may have a substantial impact on separate tasks, one needs to remember that any job involves more than one task. Many middle-wage, middle-education jobs include routine tasks but not exclusively so. The automation of routine tasks generally enhance the more complex tasks that such jobs comprise and that automation cannot replace (Autor, 2015). One can see as much in particular at a level where one needs to coordinate the tasks of many parties. Most business processes draw from a variety of inputs: labor, capital, intellect, creativity, technical skills, intuition, rules, and so on. Typically, each of these inputs plays an essential role. Thus, improving one task does not make another superfluous. In other words, productivity improvements in one set of tasks will likely increase the economic value of the remaining tasks either in a single job or a process as a whole.

One can find an iconic representation of this idea in the O-ring production function that Kremer (1993) has studied. In the O-ring model, failure of any one step in the production chain leads the entire production process to fail. Conversely, improvements in the reliability of any given link increase the value of improvements in all of the others. Intuitively, if n  1 links in the chain are reasonably likely to fail, the fact that link n is somewhat unreliable has little consequence. If the other n  1 links are made reliable, then the value of making link n more reliable rises as well. Analogously, when automation or computerization makes some steps in a work process more reliable, cheaper, or faster, the value of the remaining human links in the production chain also increases. Benefits in this dimension might result from easier coordination of inter-organizational business processes using blockchains. The panel discussed the case of AgriDigital that achieves such improvements in the agricultural sector.

Kremer (1993) discusses the application of the O-ring model for the case of automatic teller machines (ATMs). ATMs appeared in the 1970s and their number in the U.S. economy quadrupled from approximately 100,000 to 400,000 between 1995 and 2010. One might expect that such machines would have wiped out the job of bank tellers in that period. Yet, U.S. bank teller employment actually rose, albeit modestly, from 500,000 to approximately 550,000 over the 30-year period from 1980 to 2010 (although, given the growth in the labor force in this time interval, these numbers do imply that bank tellers declined as a share of overall U.S. employment). Bessen (2015) explains this somewhat paradoxical development in observing that two forces worked in opposite directions. First, by reducing the cost of operating a bank branch, ATMs indirectly increased the demand for tellers: the number of tellers per branch fell by more than a third between 1988 and 2004, but the number of urban bank branches rose by more than 40 percent. Second, as the routine cash-handling tasks of bank tellers receded, information technology also enabled a broader range of bank personnel to become involved in customer service. Increasingly, banks recognized the value of tellers supported by information technology as salespersons who forge relationships with customers and introduce them to additional bank services such as credit cards, loans, and investment products.

5 Impact beyond Singular Business Processes

Clearly, the emerging technologies we mention in this paper impact more than singular processes. The panel discussed the things they might impact and the challenges that this impact might bring for society. We summarize this discussion in seven points.

1. : the panelists expect that a good share of today’s job profiles will change or disappear in the next decade. Frey and Osbourne’s (2017) model, for instance, sees a high probability of computerization for jobs such as dishwashers, court clerks, and telemarketers, which could imply that people have to become more flexible and change jobs more often than in

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 7

the past. At least a share of the workforce will find having to become more flexible challenging. However, it does not mean that our society will run out of work. The past two centuries of automation and technological progress have not made human labor obsolete. The employment-to-population ratio rose during the 20th century, and, although the unemployment rate fluctuates cyclically, we have not seen any apparent long-run increase in unemployment according to Autor (2015). Two effects compete here: technology’s destructive effect of labor substitution and a capitalization effect of rising employment in sectors that achieve productivity gains (Frey & Osbourne, 2017). It is difficult to foresee how these effects will balance out.

2. : the panelists observed that the emerging technologies mentioned often have low acceptance. As for why, one reason may concern the level of perceived behavioral control (Venkatesh, Morris, Davis, & Davis, 2003). Indeed, technologies such as machine learning, RPA, and blockchain are complex and difficult to understand, which might explain low perceived behavioral control. Paradoxically, trust in human experts is high even though they often do not agree or come to consistent diagnoses (Schön 1983). In particular, solutions based on artificial intelligence need techniques that explain automatic decisions. Otherwise, people and decision makers may not adopt fast enough even though the technology has high factual accuracy.

: new technologies have effects that can be judged as good and bad from an ethical perspective. On the downside, artificial intelligence-based solutions might simply adopt the biases and prejudices that the training data includes. Such biases are concerns of ethical standards in systems engineering (Spiekermann, 2015). On the bright side, technologies have the potential to speed up processes that people find stressful due to their long duration. The panel featured the example of a partially automated asylum application-handling process. Beyond that, technology has also the potential to make business processes fairer and less susceptible to corruption.

4. : the panelists observed that organizations often use emerging technologies mentioned to improve customer experience. New process designs increasingly use insights from design thinking (Norman, 2013). Technologies such as chatbots offer scalable solutions for customer communication and interaction, which were formerly too expensive with human workers. However, in this context, one faces challenges in balancing automatic interaction and human interaction. Customers might or might not realize that chatbots serve them. In case they realize, one needs to question how they will act and perceive the interaction.

5. : one can make similar observations about the design of the workplace and the support of office workers. Research has established that job design has an impact not only on performance but also on creativity and employee wellbeing (Oldham & Fried 2016). Using emerging technologies can contribute to building an attractive workplace. In specific scenarios, such technologies might also have the potential to protect workers from risks (e.g., sending a remotely navigated drone instead of humans to a contaminated area). An important question concerns how one can best integrate automated tasks and human work.

6. : the panelists observed that novel information technologies have the potential to make people happier and more satisfied with their life. For instance, Ibarra et al. (2016) describe how tools can help older people to make online contributions. The panel also mentioned the case of elderly people using online tools to make appointments for knitting together and the case of education management systems adapting to the pace of the learner. On the downside, various actors have increasingly begun to use social media to manipulate elections and to disintegrate society. Currently, we have no clear account on the balance of benefits and drawbacks of these technologies from a social perspective.

7. : the panel highlighted that regulations are often discussed as a means to handle the impact of emerging technologies such as blockchains and cryptocurrencies. Blind (2016) high¬lights that regulations have an ambivalent impact: empirical evidence shows the dampening effect of compliance cost and stimulating effects of regulatory incentives. The panelists mentioned the healthcare sector as an example where regulations hinder the adoption of new technologies. On the other hand, anecdotal evidence suggests that entrepreneurs in the blockchain space value regulatory clarity because it gives them certainty regarding the legality and taxation of their ventures. In many cases, national legislators and regulators and supranational organizations have or will become active in setting the rules regulating the usage of specific new technologies.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

8 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

All these seven aspects require the research efforts of interdisciplinary teams. Insights from computer science, psychology, business administration, economics, engineering, political sciences, law, and other studies have to be integrated to investigate them in an adequate way. Also, curricula will have to evolve beyond the narrow boundaries of specialized fields in order to develop a broader perspective on these developments. Business processes will continue to be relevant research subjects in understanding the impact of new information technology on the profitability of existing business models and the emergence of new ones. We call for the BPM research community to reach out to these neighboring disciplines to study the impact of emerging technologies such as RPA and blockchains and directions for further improving them.

Acknowledgments

We thank the organizers of the 15th International Conference on Business Process Management, in particular Josep Carmona, for the opportunity to run this panel. We also thank the Tutorial and Panel Chairs of the conference Joaquin Ezpeleta (UZ, Zaragoza), Dirk Fahland (TU/e, Eindhoven) and Barbara Weber (DTU, Copenhagen) for their encouragement to organize this panel. Finally, we thank the audience for their active participation in the discussion. Unfortunately, we did not know the names of all persons who asked questions. We refer to these individuals as “person in the audience” in the transcript. Finally, we thank Adam LeBrocq for his great help in improving the stylistic quality of this paper.

 

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 9

References

Aguirre, S., & Rodriguez, A. (2017). Automation of a business process using robotic process automation (RPA): A case study. In J. Figueroa-García, E. López-Santana, J. Villa-Ramírez, & R. Ferro-Escobar (Eds.), Applied computer sciences in engineering (pp. 65-71). Berlin: Springer.

Atzori, L., Iera, A., & Morabito, G. (2010). The Internet of things: A survey. Computer Networks, 54(15), 2787-2805.

Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. The Journal of Economic Perspectives, 29(3), 3-30.

Bessen, J. (2015). Toil and technology. Finance and Development, 52(1), 16-19.

Beverungen, D., Müller, O., Matzner, M., Mendling, J., & vom Brocke, J. (2017). Conceptualizing smart service systems. Electronic Markets.

Bishop, C. M. (2006). Pattern recognition and machine learning. Berlin: Springer.

Blackburn, J. D. (1992). Time-based competition: White-collar activities. Business Horizons, 35(4), 96-101.

Blind, K. (2016). The impact of regulation on innovation. In J. Edler, P. Cunningham, A. Gok, & P. Shapira (Eds.), Handbook of innovation policy impact (pp. 450-482) Cheltenham, UK: Edward Elgar Publishing.

Brynjolfsson, E. (1993). The productivity paradox of information technology. Communications of the ACM, 36(12), 66-77.

Daily, M., Medasani, S., Behringer, R., & Trivedi, M. (2017). Self-driving cars. Computer, 50(12), 18-23.

Dumas, M., La Rosa, M., Mendling, J., & Reijers, H. A. (2013). Fundamentals of business process management. Heidelberg: Springer.

Dumas, M., van der Aalst, W. M., & ter Hofstede, A. H. (2005). Process-aware information systems: Bridging people and software through process technology. New York: Wiley.

Frey, C. B., & Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, 114, 254-280.

Goos, M., & Manning, A. (2007). Lousy and lovely jobs: The rising polarization of work in Britain. The Review of Economics and Statistics, 89(1), 118-133.

Grover, V., Teng, J., Segars, A. H., & Fiedler, K. (1998). The influence of information technology diffusion and business process change on perceived productivity: The IS executive's perspective. Information & Management, 34(3), 141-159.

Hammer, M., & Champy, J. (1993). Reengineering the corporation. New York: Harper Collins.

Heilprin, L. (1964). American Documentation Institute Committee on organization of information: Report for 1962-63. Journal of the American Society for Information Science, 15(4), 274–288.

Hirschheim, R. A. (1985). Office automation: A social and organizational perspective. Chichester, UK: Wiley.

Hull, R., Batra, V. S., Chen, Y.-M., Deutsch, A., Heath, F. F. T., & Vianu, V. (2016). Towards a shared ledger business collaboration language based on data-aware processes. In Q. Z. Sheng, E. Stroulia, S. Tata, & S. Bhiri (Eds.), Proceedings of the 14th International Conference on Service-oriented computing (LNCS vol. 9936, pp. 18-36). Berlin: Springer.

Ibarra, F., Korovina, O., Baez, M., Casati, F., Marchese, M., Cernuzzi, L., & Barysheva, G. A. (2016). Tools enabling online contributions by older adults. IEEE Internet Computing, 20(5), 58-65.

Kremer, M. (1993). The O-ring theory of economic development. The Quarterly Journal of Economics, 108(3), 551-575.

Lacity, M. C., & Willcocks, L. P. (2016). Robotic process automation at Telefonica O2. MIS Quarterly Executive, 15(1), 21-35.

Mendling, J. Weber, I., van der Aalst, W., Brocke, J. V., Cabanillas, C., Daniel, F., Debois, S., Di Ciccio, C., Dumas, M., Dustdar, S., Gal, A;, García-Bañuelos, L., Governatori, G., Hull, R., La Rosa, M., Leopold,

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

10 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

H., Leymann, F., Recker, J., Reichert, M., Reijers, H.A., Rinderle-Ma, S., Solti, A., Rosemann, M., Schulte, S., Singh, M.P., Slaats, T., Staples, M., Weber, B., Weidlich, M., Weske, M., Xu, X., & Zhu, L. (2018). Blockchains for business process management—challenges and opportunities. ACM Transactions on Management Information Systems, 9(1), 1-16.

Mery, S., & Selman, D. (2017). Make your blockchain smart contracts smarter with business rules. IBM developerWorks. Retrieved from https://www.ibm.com/developerworks/library/mw-1708-mery-blockchain/1708-mery.htm l

Mooney, J. G., Gurbaxani, V., & Kraemer, K. L. (1996). A process oriented framework for assessing the business value of information technology. ACM SIGMIS Database, 27(2), 68-81.

Mukhopadhyay, T., Rajiv, S., & Srinivasan, K. (1997). Information technology impact on process output and quality. Management Science, 43(12), 1645-1659.

Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system. Retrieved from https://bitcoin.org/bitcoin.pdf

Norman, D. (2013). The design of everyday things: Revised and expanded edition. New York, NY: Basic Books.

Oberländer, A. M., Röglinger, M., Rosemann, M., & Kees, A. (2017). Conceptualizing business-to-thing interactions–A sociomaterial perspective on the Internet of Things. European Journal of Information Systems.

Oldham, G. R., & Fried, Y. (2016). Job design research and theory: Past, present and future. Organizational Behavior and Human Decision Processes, 136, 20-35.

Reijers, H. A., & Mansar, S. L. (2005). Best practices in business process redesign: An overview and qualitative evaluation of successful redesign heuristics. Omega, 33(4), 283-306.

Schön, D. (1983). The reflective practitioner: How practitioners think in action. London: Temple Smith. Shawar, B. A., & Atwell, E. (2007). Chatbots: Are they really useful? LDV Forum, 22(1), 29-49.

Sim, I. (2016). Two ways of knowing: Big data and evidence-based medicine. Annals of Internal Medicine, 164(8), 562-563.

Spiekermann, S. (2015). Ethical IT Innovation: A value-based system design approach. Boca Raton, FL: CRC Press.

van der Aalst, W., & van Hee, K. M. (2004). Workflow management: Models, methods, and systems. Cambridge, MA: MIT Press.

Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425-478.

Weber, I., Xu, X., Riveret, R., Governatori, G., Ponomarev, A., & Mendling, J. (2016). Untrusted business process monitoring and execution using blockchain. In La Rosa, M., Loos, P., & Pastor, O. (Eds.), Proceedings of the 14th International Conference on Business Process Management (LNCS vol. 9850, pp. 329-347). Berlin: Springer.

 

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 11

Appendix A: Transcript of the Panel Discussion

15th International Conference on Business Process Management

Thursday 14th September 2017, 10:40h - 12:10h

Auditori, Vertex, Universitat Politècnica de Catalunya, Barcelona, Spain

Part 1: https://www.youtube.com/watch?v=6kpJDWE3sxc Part 2: https://www.youtube.com/watch?v=p2z1uRKGdoo Part 3: https://www.youtube.com/watch?v=EYB2ZHg7Ygo

Jan Mendling (Chair)

Ingo Weber, Data61, CSIRO, Australia

Gero Decker, Signavio, Germany

Hajo Reijers, Vrije Universiteit Amsterdam, NL

Richard Hull, IBM, USA

I am very happy to welcome you here in the audience. We are heading into a panel discussion. There are some exciting developments technology-wise that make us talk today about the human factor in BPM and how far emerging technologies are challenging this human factor in BPM.

I am very happy that we have four experts here as panelists and I want to briefly introduce you to these people. Next to myself, there is Hajo Reijers. He received a PhD degree from Eindhoven University of Technology and is now a full professor at the Vrije Universiteit Amsterdam. Welcome Hajo. Next to Hajo, we have Gero Decker. He received a PhD degree from HPI in Potsdam, took his work into practice and founded Signavio with others. Very happy to have you here Gero. Next to Gero, we have Ingo Weber. He received a PhD degree from Karlsruhe and after working with SAP, he moved over to Australia to join the University of New South Wales and then Data61, CSIRO, which some of you might know by its old name NICTA. And we have Rick Hull with us. He received a PhD from University of California in Berkeley. He was a professor at the University of Southern California for about a decade, and then switched to industry, first at Bell Labs, and now with IBM Research.

Let me set the scene of what we are going to talk about. One thing that is important when we talk about processes is that processes are inherently complicated matters, because there is division of labor in bigger organizations. That means much of the work that needs to be done is split up in smaller pieces: there are different persons, different parties and different companies involved with separate tasks, and we need to coordinate them in order to integrate the fragmented results of these pieces of work. And many of the developments that we see recently have an impact, either on the way how certain tasks are performed in an enterprise setting or on the way how these tasks are coordinated. To mention just two examples, there are technologies that build on artificial intelligence and machine learning, robotic process automation is one of the terms in this context. All these refer to training techniques associated with tasks that can be automated, such that humans are potentially replaced with IT based robots to do their work. On the other hand, there are developments in terms of coordination – blockchain is one of the technologies that facilitates coordination between different parties, and this also raises the question, if we will have infrastructures like blockchain to take care of coordination.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

12 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

One of the questions that is raised in this context is how does that generally impact work, on a smaller process level but also on a greater scale. How does this affect society? And one of the challenging question is: do we run out of work?

I want to briefly read out a small piece that I just recently found, and found quite interesting to reflect upon, and I quote verbatim. It is a piece from the American documentation institute and they write the following:

There is a sizeable fraction of the workers which is unable to adapt to a new or different industry. Recognition that this type of unemployment (personnel who cannot adapt) is a chronic effect of scientific and technical advance, not of population growth, may lead not merely to retraining programs in rapidly evolving fields, but a new attitude toward education. (Heilprin, 1964)

This sounds nice when you think of ourselves as being those in research and teaching promoting education. One of the interesting things is that this quote is from the mid-1960s, so it is not very recent. I would like to invite the panelists to make their entry statements, such as they can share with you their general observations around these different topics and then I would invite you to bring in your comments, ideas and questions such that we can discuss this topic in a broader audience.

So Hajo, do we run out of work?

Well, it wouldn’t be so bad from some perspectives, but I understand that your question is of course a question that is on the minds of many people, economists and other people. I think there is also a genuine fear about the effects that automation has on the work that people do and indeed there have been incredible changes over the recent past.

My position, however, is that fear for running out of jobs is heavily exaggerated and I think one of the key elements to understand how this impact works is, that there should be a distinction actually between jobs that people do and the tasks that these jobs are composed of. Any job is composed of multiple tasks and what we see over the past decades is actually that automation, all kinds of algorithmic approaches - what they do is that they target one particular type of task, which is often part of the jobs that people have. These task are what you can refer to as routine tasks or to tasks, for which there is a certain procedure behind it, there is a sort of repetition, sort of a structure that can be unveiled, can be discovered and that can be translated into an algorithmic approach. These tasks clearly can be aimed at, can be targeted with automatic approaches. By the way these kind of routine tasks are not always the cheapest tasks in the jobs that people do. There may be other things manually that require manual skills, but these routine tasks are at an enormous rate being automated in all kinds of areas. And if it economically pays off this will continue. Now what I find interesting is that when you look in different areas in different studies and when you look at the automation effects of the routine component in something that is bigger, it is not so much that the overall system or the overall process is being reduced with specific parts of it. More than that, the overall value of the system or the job that this task is part of is actually enhanced. So let me give you an example here. I read an interesting study about the use of the introduction and the use of ATMs in the United States. ATMs were introduced in the 1970s. In a study that looked at the replacement effect of ATMs on tellers - people who would be working at banks and would actually hand over money to people who visited these banks, you could clearly see that this element of the job of these tellers, the human tellers, was replaced and was automated with these ATMs. These ATMs, the use, the introduction and the presence of these systems in the 1990s approximately quadrupled. But what happened is that these people who previously did this part, handing over money to clients, that this kind of job did not diminish, it actually increased over in the same time period. And there were two factors behind it. One of them was that because the overall job became more economical to hand over work in an automated way. There were also opportunities for banks to open more branches; so there was an actually need for more personnel, and also personnel who were previously working as tellers. But more importantly, these tellers had other tasks that they would do. They, the human tellers, would explain to clients what the other Bank products were, for example. They would introduce their clients to these bank products and would also do, I would say, minor sales elements. And these parts of their job were actually enhanced through the automation of the more routine part of their work. So instead of a decline, you would see an increase, though it was a small increase, and to be honest the number of tellers as part of the labor population has decreased over time but in absolute numbers this went up.

And this is actually a part of a sort of a wider phenomenon, which is sometimes referred to as the O-ring theory. If you have any type of system that is composed of different parts and the overall system would stop working when one of these parts fails, by improving one part that actually the value of the overall system enhances and doesn’t diminish. And you can see this also in business processes where, if you enhance a certain task in a process, the overall process becomes more valuable and it becomes also more important

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 13

that the other tasks in this business process are executed well or with a high quality. So, having said all this, this explains my statement that it’s not so much that jobs will go away. Jobs will be affected, heavily affected by automation and keep being affected, especially jobs which of course contain to a large extent routine tasks will be heavily affected, but I think the fear that these jobs will completely disappear, those fears are exaggerated.

Thank you very much Hajo. So, we are talking about these repetitive tasks. They seem to be the key to understand what is going to happen with these new technologies. I want to hand over to Gero, because he is not only the co-founder of Signavio, he has also expertise from being involved with a company that is called Parlamind. And there are some scenarios more concretely visible that illustrate what we have talked about. So Gero, maybe you can share some of your experiences in this context.

Sure. I’m an engineer by heart, so I love technology and I love experimenting what’s possible and push the boundaries You know overall I’m excited about all of the things that are available now and how it could affect things. But on the other hand, I am an entrepreneur. Having brought things to market and having to convince people that they actually have to pay money for using that type of stuff, gives you a reality check of what’s actually needed and what is viable in the marketplace. So talking about AI in particular, conceptually, theoretically it can do so many things and it is an exciting thought exercise. It helps you build self-optimizing, self-learning machines that can do all kinds of things that humans did before. The question is when you want to bring that to market and make that work – what are the economics behind that and when does it actually work and when doesn’t it work. Parlamind is an interesting example. It’s a company that I’ve invested in very, very early, and what they do is, they automate customer service request emails or customer service requests. They specialize on e-commerce only at the moment. So you order something at Amazon; or you order something at Zalando. You have questions like “where is my package, next Monday is the birthday of my girlfriend, I am waiting desperately, what can I do to speed this up”, or “I ordered the wrong thing, I want to cancel it, or the thing that arrived is the wrong thing, I want to exchange it, how does it work?” This use case is interesting because it seems to be very repeatable, a very replicable scenario so that it is very easy to build an AI, a self-optimizing machine, that respond to these kinds of requests. Because it’s one domain, you have a high volume of requests that you can learn from and it’s actually something that is replicable over the companies, so it is not specific to one company but it is replicable to let’s say, ten thousands of online shops out there. And with that type of scenario you can build a machine that understands the sentiment of the email. So, for example, this person is angry or this person is happy, or there is some urgency in there, or here it’s about conciliating someone. For example, “next Monday is the birthday – I understand that it is super important to you. We will make sure we do everything to make your girlfriend happy.” And this will be a robot saying that. I is not a human but a robot saying that. And saying, well we can offer you, 10 Euro more for overnight shipping, to make that work and the likelihood we have seen in the past is that, in the area that you work in – that you should be fine. In other areas, however, I have seen a lot of attempts of applying AI and it has been super complicated and super hard, and did not return the investment that you have to make. Why? Because it is super specific to the particular use case, the training data that you need is specific to one company or to one’s special task at hand, and you simply don’t get economies of scale to make that work at a reasonable price. To give you one example: a friend of mine has a company that specializes in dynamic pricing. This is what amazon introduced, every person at every moment in time gets a personalized price, and you optimize on certain criteria like for example getting rid of your stock until a certain date and reaching the maximum price until then. And what they found was that AI technology and machine learning technology works great, if you have more than a billion in revenue with your company. If you have less than a billion in revenue for your retailer, the machine learning is simply not powerful enough, and it’s much smarter to have people sitting there, building all kinds of rules and decision models for pricing and campaigns and tailoring the price to the specific person, because you simply don’t reach the scale for machine learning technology to be economically viable. Long story short, great technology – but I think we are still early in the game to find out where is the viability of those things and where can we make it work, not only conceptually or theoretically, but also economically to return the investment that you want to have in business. If that investment is not returned, people will still do things for a long time to come.

Thank you very much Gero. I want to turn to Rick with this observation you made that many things are much more complicated. Data is versatile, and Rick has strong expertise in data management, which he accumulated over several decades. Rick, I would like to ask you to comment on how the scene has changed in terms of how we work with data in order to make all these scenarios work in the last couple of years. How do you look at that?

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

14 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

Thank you. I want to mention three areas where data is really central to how the BPM world is changing, both because of AI and because of Blockchain. And the first thing I want to hit on, and it’s kind of related to what we were just listening to is, namely, data entry. So much of business process comes down to getting data into the system, maybe at the beginning to launch a process or maybe in the middle because more data is needed. And it’s really kind of a curse of our existence I would say – and I think it’s going to be a curse of our existence for quite a while. Just one example in a back office processing for human resources for hiring, terminating, promotions turns out that about 60% of the labor today is focused on data validations. Data comes in and if it’s wrong, it has to be fixed – that can be a major pain. Now, how is AI helping in that? It’s through the conversational front ends, as an alternative to forcing people to work with one kind of menu based situations or putting in through forms that are handwritten. Conversational front ends, that are given to us by natural language understanding, natural language generation and kind of machine learning to figure out how conversations should be going, allow the data entry to be more natural, more intuitive, easier, and faster to go in. But at the same time as just mentioned, the target for those conversational front ends is still a very, can be a heterogeneous environment. Now, even though the conversational front end is kind of simplifying, not only do I want to put data in, but I want to make sure that it’s consistent with the other data, that may be in my environment. That can be a challenge because that data may be spread across multiple silos. So, there is still this challenge of data integration.

This brings us to the second thing I want to get at, which is blockchain. What’s interesting is that in this room, of course, everybody’s heard of blockchain, but probably half of you have kind of been reading up on blockchain and understand where it’s going. I think for others that I have talked to it is still kind of an unknown, you are thinking, “oh yes, maybe I should start looking at it”. The thing about blockchain is it does represent a disruptive moment for business process or at least for some large portion of business processes. This is because blockchain enables seamless data sharing between multiple organizations that are trying to do business processes. To have a single shared repository of data, a single ledger, if you will. And it’s got the encryption and consensus and this and that, so that even though it’s a single shared repository, people are able to trust that the different organizations are able to trust it and it has privacy guarantees built in. For example, maybe I am working with Ingo on something, but then Hajo won’t be able to see it, if I don’t want him to see it. So, it provides this basis because of the underlying technology, it provides a new way of enabling a shared repository. So this new approach is very different than today’s approach to having multiple companies work together where each company has its own silo of information. This totally streamlines or gets rid of the friction of data inconsistencies in many ways. Instead of having my copy of data and Ingo’s copy of data, we have one copy of the data. And as the data goes in, we can be checking whether it’s consistent with other data that’s already in there. Now, what are the implications of this shared repository? First off, it is going to change the way a lot of business processes deal with crossing silos within an organization or crossing boundaries between companies. That friction is going away, it is going to lead to the development of shared data models so the companies have to get together about, what’s the data schema of what they are representing. So suddenly, there’s economic motivation for industries to arrive at as standardized data models. In essence this is going to bring back the whole attempt of the Semantic Web community. Right? The Semantic Web Community was truly hoping that we would come up with standardized ontologies for healthcare, for logistics, for finance and accounting, for HR and etc. Because of blockchain people will want to start sharing information against a common data model and so these common data models will arise.

Now moving to the third thing that I want to mention. Even though blockchain right now is focused on going between organizations, the same principles will start to pervade within organizations so that if you have different silos within an organization even those will start saying “oh I want the benefits of a shared data model, oh I can use blockchain as that basis for the shared data model.” As the data becomes more uniform and standardized, this actually will be an enabler for a shift in business process management from a process centric or a data centric perspective towards a goal centric perspective. This will be an opportunity for people instead of saying: “oh I need to put this data in and I want this task to happen etc.” They will be able to say: “this is my goal and the goal can be expressed in the vocabulary of the shared data model.” And so now this will enable the emergence of applying AI planning technologies against processes as we know it.

So, just to summarize, I am trying to get at three mechanisms where blockchain and AI are going to be transforming how we do business processes. It will create a lot of employment for us, the research community, because there’s a lot of details to be worked out. And as Hajo was saying, it will be transitioning the kinds of jobs that people do, but it will be allowing more people to think about kind of goals and the business values and the objectives that they are trying to reach. Thank you!

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 15

Thank you very much. You mentioned blockchain and I am looking at Ingo now because he is kind of the mastermind behind the blockchain activities in CSIRO. Ingo, you guys in Australia are working on various scenarios and, as Rick said, the concept is quite abstract for many people. Maybe you can illustrate some of the applications that you have been working on and your perceptions of what this technology is going to bring.

Thank you Jan. Maybe that was a little bit too much honor. So, I want to do exactly that but before I do, I just want to briefly take a step back and look at the premise of the question for the panel which is: will machines eat the human factor in BPM? Now there are two premises in there, one is that basically all human touch is good and, two, that people will mind if their tasks are automated. And as Hajo was referring to these routine tasks, I think there are a lot of tasks that some of us have to do, where we wouldn’t really mind those tasks being automated and that will lead to this transformation. Also in terms of all human touch is good – in these routine tasks, quite often we find ourselves in a situation where the best performance a human can achieve is not stuffing up. And so it is not necessarily a bad thing if this work is transformed. Also of course the ethics of humans making decisions or the way they perform tasks are not always perfect. You have things like corruption or bias. I think of things like corruption that we can probably reduce by automating tasks. For example if we take customs processes in developing countries that in some countries there is a lot of corruption, you have to bribe the right people to get your goods through customs on time. Reducing that is probably a pretty good thing, at least for the general society, probably not so much for the customs official. But when it comes to the ethics, we also have to consider the ethics of the AI, because if you have a bias in the data from which you learn, you can learn to make unethical decisions and can automate them such that they always be made in this unethical way. A different question also is from this broader perspective is if people lose their jobs, then we have to also reconsider that wealth distribution and purpose in life to a large degree for many people are associated with work. And that I think is a challenge to society as a whole, which we probably won’t be solving in the BPM community or on this panel today.

But now coming back to your actual questions or blockchain opportunities and scenarios. Blockchain, I think, has two primary features that are great in this regard: one is that they enable inter-organizational processes, collaborative processes in a different way, they enable potentially more complex supply chains for one example. They might make it easier for Fijian producers to export their goods to Australia, which can be very positive. Topics like these we are working on. There is a startup company in Australia, AgriDigital, with whom we interact quite a bit, and they want to reduce the cash flow issues in grain supply chains for Australian farmers, which is a very important topic for these farmers. In Australia the suicide rate among farmers is about twice as high as for the average population, and to some degree this is attributed to cash flow problems, with the farmers not knowing how they are supposed to pay the salaries and for the goods in the next month and by using blockchain to make this process smoother and make the payments faster, that can be fantastic. I think back office automation can happen also to some degrees in the banking sector with settlements potentially happening much quicker and needing less human routine operations to be performed. The other thing that you can also get out of blockchain is that you have reliable data, data that’s more reliable than previously. There is a start up from some friends of mine in Sydney, who want to bring supply-chain financing and invoice financing to a broader share of the market. Invoice financing, meaning that you can take an invoice, go to the bank and get a loan against this invoice or sell the invoice to somebody. And for a lot of small companies this is currently not possible because the volume of the invoice is not big enough and the checks that are needed by people are too costly for the banks to be interested in doing that. So, I think when we talk about automation in this sense we also have to realize that, yes some tasks that are currently performed by people may be automated away, but also, potentially, we can create new business models and processes which increase the overall market dramatically and there are many opportunities in that regard.

Thank you very much. I would like to open up the discussion for the broader audience and I invite you to comment and challenge the different observations that our colleagues shared here.

Thank you. I have to say I work in two very huge projects. One is on work 4.01 asking what the impact of digital transformation on the employees is. What I have learned there is that we as computer scientists are not able to speak about the human factor, so we need people from other disciplines. In this project, there are ten professors and only two computer scientists, all the others are from psychology, sociology, from economics, from law and so on. And I think we need this discussion, this opening of our

1 See http://www.bmas.de/SharedDocs/Downloads/EN/PDF-Publikationen/a883-white-paper.pdf?__blob=publicationFile&v=3

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

16 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

community to these areas. This morning, we saw in the talk2 by Alan this pyramid of values. We are very good in this functional values, but in all upper layers we have no idea how to realize systems, so that people become happy, satisfied and so on. So, what I would like to ask you, why do you think that you are qualified to speak about human factors, you are computer scientists, maybe business informatics people. You have no idea about humans.

How we are qualified to sit here? Because we got invited. So, we didn’t self-select ourselves. Maybe I can speak to that a little bit. We at Signavio deal with a lot of companies out there applying process management. What is interesting to me is, we always look at what are people trying to achieve, what are the goals, what are their challenges. Five – ten years ago a lot of people were focusing exclusively on the lower part of the pyramid – how can I save cost? How can I streamline things? How can I make things faster? So, a lot about operational efficiency and making operations run smoother for the company. What we have seen as a trend in the last two, three, four years is that a lot more companies are using process management now. What we have seen is actually the upper right corner of Miguel’s chart of digital transformation,3 or both upper parts, so for customer experience and for employee experience, which is very interesting. We have a lot of customers who optimize processes with the main goal of building in a more attractive workplace, because it is so competitive to find people that they have to make sure that the environment that people work in is attractive, is appealing, that it brings out the best of the people. So, this is interesting. I haven’t seen this particular piece and research how to increase employee happiness through designing process in one way or the other. But it’s happening out there in practice. Customer experience –

I have seen a couple of things here at the conference as well on how do you actually get back to the roots of process management, if we remember where process comes from: it basically puts the customer at the center and tries to work in a way that you achieve the customer’s goals or fulfil his need. That seemed to have gotten lost a little bit in the last 10, 15, 20 years, and the focus was a lot more on the internal things happening within one company. But in the latest survey that we did with our customers, actually 35 percent of all initiatives are mostly driven by customer experience and improving that. I’ll give you an example: imagine you are a car insurance company. What are your interaction points with a customer today? You have exactly one interaction point with a customer every year and that is when you send the invoice and tell them they just renewed their contract. Right? So, these nice letters are the only touch point that you have with your customer. But what you actually want to provide as an experience is that you want to be there for your customer every day when they enter their car. You want to make sure that they understand that when they enter the car and they touch the steering wheel, that you are the safety net for them. Right? Whatever happens to you on this particular day, I am there with you, I am working with you to keep you mobile, to keep you safe, to lead your life in a way that you want to lead it. This has a heavy impact on how you design your product, your service and this has a huge impact on processes and the things that you have to do to deliver on that customer experience. So, I see good signs in there, walking up the pyramid and in practice I see a lot of optimization towards those things already.

Let me add to that. I think you are right as computer scientists we’re probably not so qualified, but there’s some good news in some industrial settings and it started with Apple Computer and their approach to designing products. Apple pioneered a new way of thinking about making products, putting the user first, the consumer in their case. And so there has evolved a method that we call design thinking. And now at least at IBM, when we endeavor to create a transformation or some technology for a client, one of the first things we are doing is a so called “design thinking workshop”. This is a systematic method of steps that we go through, starting with who are the stakeholders, who are going to be using whatever this technology is. And we think in terms that there might be multiple stakeholders and for each one systematically: what are they trying to do? What are they thinking? What are their pain points and what are the feeling? So, there’s a strong discipline now of thinking in terms of the user as the starting point before getting down to what are the kinds of solutions we’re going to provide and then finally underneath that, what are the kinds pf processes that we will be setting up to support those solutions.

I think that what is happening when you get more computational power, is that you start to do more unnecessary things. It was the case before when you tried to break the Enigma: People calculates by hand and then suddenly you have a massive computer and you said: oh this computer will solve every calculus that is needed of mankind. And with more and more, we have more graphics, and now

2 Alan W. Brown: A Leaders Guide to Understanding New Business Models in the Digital Economy. Slides and video recording of the keynote are available at https://bpm2017.cs.upc.edu/keynotes/keynote-brown-2/

3 Miguel Valdés: Intelligent continuous improvement, when BPM meets AI. Slides and video recording of the keynote are available at https://bpm2017.cs.upc.edu/keynotes/keynote-valdes-2/

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 17

we have AI. So, what I actually want to say is that we’re starting to do the unnecessary things, which will become more necessary, because of the experience. So, I totally agree with what you say. You start to do more things and you have more possibilities, so that is what will happen. And also maybe you can start to have new revenue streams from that.

Thank you very much for the questions before and your comment on the inter¬disciplinary nature of how BPM has to collaborate with our disciplines, because I see a lot in the industries right now that BPM alone is not really surviving. We have to join with some other disciplines so that we could give the business value. Because right now the companies are facing not only how to improve the processes; but also how to find good people, how to really find a good profile, a fitting profile for the qualifications, that they are looking for. And that is coming to my first question about the vision within academia: how would you like to shape all the syllabus that you are producing right now to fit to the industry needs, because frankly I have to speak, we are working on digital transformations, and we hardly find people who can really do things. Most of the things people work, being specialist on some issues, but for some certain point of view in a company and doing digital transformation, you need good affinities on the technical side, you need an understanding of the organizational structures and people, and you need also to understand processes. And not every studying program provides this all, and you have to be a good project manager. So, the requirements of the industry is higher than before and they expect also that once that the students are getting out of the university, they have to work and they have to function. So that is my first question: How the academia has visions to change or to adjust the syllabus or to fit into the industry needs.

My second question is about the human factor in the BPM. I don’t think it is just only the questions of how much automation and how much digitalization that we would need. Sometimes we are driven by some factors, especially the demographical structures change. I went to Japan last year and I visited a hotel, there is not a single person when I go to the receptions, there are robots. I got all my processes done via my mobile phone and because Japan is one of the greatest country that has an eldering or aging society and they don’t have much workforce to really respond to the needs of the communities. That is one of the issues that probably we have to see how the technology could help and to what extent the technology could really deliver. What do you think in your opinion, up to what certain level of service delivery that could be really positive for the society? And to what extent of these kind of technologies, we are going to not be working anymore, if we see the issues from the theme “I, Robot” [movie from 2004], for example.

So let me indeed try to pick up on the role of the universities may have, I think on the one hand we must be modest. I’ve been working in industry, I’ve been in touch with industry over the past 20 years, I always get request, why don’t you train people more to do this because we need this more and we need that more. And I think what universities can accomplish in an educational program of students is in the sense modest. You cannot prepare people for all the challenges which are multifaceted as you are pointing out, a good project manager, somebody who knows from a technical perspective what’s going on, the human side to it. You cannot address all these elements, but I do think that there is a fundamental change that needs to be made and I think you can also observe this in different settings where for example in educational programs, that I am aware of which were traditionally much more focused on a particular type of technology, in even the state-of-the-art of that technology, that there is a shift towards helping students assess the impact of these technologies, for example, in business process settings, where the technology itself is of course important and is current, but where there’s a complete understanding that once the students will graduate that there will be new technologies. So, for example, in the curriculum that I am aware of in Amsterdam, courses relate to digital innovation where students take a particular technology and look at existing processes or existing business models to think through, how these things could be affected by incorporating new types of technologies. And of course the students have to understand that technology, but it’s not so much the education on that technology, which is the essence of that part. And I myself am in favor of having people think, train them to rethink existing structures or existing processes with knowledge, profound knowledge, of new technologies. And I think that this is something that is very helpful to prepare them for, I would say, a situation where we are fully aware that the technology and streams of new technologies will be evolving all the time. So, if you have this mindset that you are trained in changing existing structures, because new things happen, I think that is a very important strength for our future students.

I first would like to also address the first of the two questions. So if we want to really solve problems, quite often we have to work in an interdisciplinary way. I probably don’t need people from other sciences to help me write a BPM paper. But if we want to, for example, apply blockchain in the food provenance sector, that is different. Wil [van der Aalst] raised a point I think two days ago: how do you know

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

18 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

that the data that’s on the blockchain, that is being fed into the blockchain, is actually correct. If you look at food provenance, let me give you an example: a box of prime beef from Queensland in Northern Australia makes its way to China, and somewhere along the way the prime beef in the box is being sold separately and replaced, and so it “multiplies”. There was another example of a vitamin company, who produces I think one million bottles per year globally, and in China three million bottles of their products are being sold every year. So, with these issues, food provenance, medicine, etc. you have to understand that the product is genuine. And so, the data that is being fed into a blockchain solution has to be trustworthy as well. To come back to the beef example: we at CSIRO, we have different divisions across different sciences and sectors, so for example we have an agricultural business unit and they are looking at things like analyzing the beef and being able to tell from which pasture it came, on which piece of land the cow lived and fed. And if we combine that with blockchain, then of course we can get a very valuable solution in total. But I agree [with Gregor], we by ourselves might not be able to solve the real problems for the customer, so for the industry.

As for the second question about ethics: what I was referring to when I started talking about it, I think there was this example of one company from the Silicon Valley adding facial recognition software to one of their photo products, so it could recognize faces. And it turns out they trained this product only with the pictures from their employees, which were primarily Caucasian and Asian. And that led to a case, where the picture of somebody with an African background was then matched to a monkey. When we look at things like machine learning, we have to consider these issues that we don’t have these kinds of racial biases in the training data and in the algorithms that we produce.

I like the title [of the panel] a lot, and AI, machine learning, blockchain, I see them as technologies. So the aim for which they are used might vary. I was thinking in certain cases, we like when computer-aided systems can prove to be more effective and more efficient, because they allow us to reduce the efforts and ultimately time, so we save costs, make more money, etc. But how about instead having another viewpoint, which is not necessarily using these technologies to make things faster, more efficient, quicker, whatever, but just to try to reduce the human risk. Like for instance, I remember during the Fukushima disaster, we sent human beings to check the situation, this is high risk. Maybe in that case the automation of this activity—I know I’m dreaming—would have been much more effective. So, instead of thinking about these techniques as eating the human factor, we can use these techniques to protect the human. [This is] question number one. The second question is more regarding the self-adaptation of humans. There are certain jobs that for sure are going to decrease in appeal and eventually disappear. But this requires people that were working in that area to re-adapt to do something different. However, we all here are into computer science or related fields, and we are quite used to change the topics of our research continuously, but some of us and our friends and relatives are not used to it. They can do one thing very well and then –this is like business process reengineering tasks–it could be hard to really re-adapt again. What I had in mind is that looking far ahead, couldn’t it rather be the risk that having these technologies we can even increase the gap between who can actually have this at hand and who cannot. Because who cannot and does not adapt is left behind. And isn’t this risky, in this case, for the human factor? Not directly, but rather for the evolution and self-evolution. Like big fish eats small fish.

So let me address the saving humans question. I fully agree. Everything you said is totally correct. The one constraint is, what is actually accepted by people and where do they have reservation of applying that technology. I give you two examples that we were involved in and where we simply couldn’t get through, although to us it was all obvious that technology would have been the better solution. One is recommendations for medical treatments. It is proven that if you take the knowledge that is out there and you derive certain rules from it, you let the machine learn certain things, then machines in many cases make better decisions than the doctors at hand. Yet doctors don’t accept that and they refuse that type of technology. So, adoption is, at least in the cases we have been involved in, very minimal, because just of the reservations of the different people involved and having fear of letting go of certain things. Another example in Germany, a while ago we had a huge wave of immigrants, refugees coming to the country. And all of the asylum applications where you basically decide whether to let that person in to grant asylum or to reject them was completely overwhelming the authorities. More than a million people came, and the agencies and the processes were designed to handle 50,000 cases a year. So, we brought in decision management technology and all of the good stuff that you can do with technology, and we said: look we can go from nine months cycle time of this process and we can nail it down to two or three days. From days of time spent on each case we can reduce that to one or two hours. And we can make better decisions with the technology and we can show that as a proof of concept that the technology actually delivers faster and better results, unbiased, much better than the people involved. And they would have been able to minimize

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 19

the backlog like immediately and solve a huge problem for those people not knowing what their future is, whether they can stay or not and so on and so on.

But in this example, many people involved simply said they don’t trust the machines. Let’s rather hire 7,000 additional people to do the job. Did they find 7,000 people? Of course not. Did the backlogs go away? Not at all. Is it still a problem? Yes, just reservation and people not wanting to adopt technology. So, I think this is the biggest barrier. Maybe I am too optimistic of what technology can do and maybe I only see the good parts of it, but I think the bigger question or the bigger problem we have right now is that people don’t adopt it fast enough.

I just want to respond to your second question which was the adaptation of the workforce towards new kinds of jobs, and I want to make an argument for optimism. I think that there’s a new research area that AI people could be looking at and it would be to combine two existing capabilities into a third new partial solution to the problem of retraining the workforce. The one capability is this notion of the AI-driven interaction with humans. We see chatbots and we see other kinds of conversational interfaces, we see virtual reality systems. Imagine if those things were aimed towards helping workers get into a new kind of work. So, here I have got a worker, he is doing one thing, now he is put into a new kind of position. Can we be coaching him, almost as if you had a human coaching him in the background, but now it’s more an AI-driven system, so we can do it at scale. The second kind of technology that may be applicable here is this so-called education management systems and the idea of personalized education pathways. So we’re starting to see this kind of technology of a digitized style of education, where for each student it takes into account what are their interests, what are their skills, what do they already know, what are their learning styles. It is an AI driven approach to cater to each individual in the best way. So, maybe in a few years we can take these two things, AI-enabled interfaces and personalized education pathways train workers in a much more efficient way.

I have a comment just for fun. So there’s two futures sometimes, particularly around AI. There is the terminator future human of annihilation and then there is the Star Trek future of human empowerment. We as a community push the boundaries of human knowledge. Mathias [Weske in his keynote4] very nicely explained it in a sense that puts us in a position of responsibility to be careful with the narrative that we use in terms of technological advancement. So, will machines eat the human factor, why didn’t we say machines amplify the human factor. So, I’d like to get your positions on it and it would be nice if you disagree, just for fun.

First of all, I almost never dared to disagree with you Shazia. So, it’s quite a challenge. So, if I understand you correctly, you’d like me to disagree with you turning this perspective, right? You’d like me to support the Terminator, the terminator future. I simply don’t believe in that it is true, but I agree with you that framing and the way we talk about technology says a lot about, also may instill fear in people and how we talk about this. In the discussion that we just had, how easily we of course talk about technology and use of it as Rick did, which is a fantastic technology: AI, to train people for example, attain new jobs. The pure mentioning of AI, I would say for many people is already something which it instills fear, which is alien to elderly people, who already have a problem with working with their smartphone. How are you going to convince them that if they would use an AI training system to get on the path again? That’s going to be a huge challenge. So, as in many things the way we position it, the way we discussed and talked about these things has an impact of course on how people will perceive this and how successful it may be. Even if we are on the path, if we all aim towards this, I would say more this Star Trek scenario. That’s the best I can do.

So, this was exactly what I was getting at, when I said I want to look at the premise of the question, in my opening statement. To take a different stand, I am almost 100% convinced that there will be a percentage of people who won’t be able to cope. They will be left behind. And that is what I meant with challenges for society at large in terms of wealth distribution and in terms of finding purpose in life. If you work, if you derive purpose from your work and of course you get a salary, that is individually of course a challenge if you don’t have that anymore. But also for the state it is a challenge, because if some people don’t receive a salary, then how do we raise taxes? Can we tax the work of robots or the robotic process automation? Probably not. So, I think there are many challenges and if you want to be pessimistic, then there is ample opportunity to follow the Terminator scenario.

4 Mathias Weske: BPM: Reflections on a Broad Discipline. Slides and video recording of the keynote are available at https://bpm2017.cs.upc.edu/keynotes/keynote-weske-2/.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

20 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

Let me jump in on that. The question is, what the scope of what you are looking at is. If you look at the world overall I believe in the Star Trek vision and technology amplifying what we do but that’s often not the scope or horizon that people look at. So, for example, another scope for people, the more relevant scope might be, what happens to one particular company. And there is simply no way around getting the truth and what’s necessarily going to happen, that there are organizations and certain industry verticals that are going through massive transformation. Where in five years from now, you will only see a fourth of the people working there. And there is no hiding from that. And people try to fight an argumentation: yeah, but we create all the jobs in other parts of the organization. No! 80% of the workforce will be gone in five years. That’s the truth, because technology does the job better, more efficiently, with higher quality than what the people are doing today. And that is of course a difficult discussion with people. And people reject that thought, but these are then the kinds of people who perceive things as the Terminator scenario. So, then the question is how do you make it work for everybody? Because, if you compare the world now versus the world 50 years ago where people could stay with one company for the whole life, this is simply not the case anymore. And you need to educate people about that and show them the opportunities outside of the scope that they are currently in. But to those people technology feels like terminator.

I’ll try to be short but I couldn’t resist the opportunity to be controversial. I think there are two forces driving how AI is going to get used. The one force is economic, corporations, business. What makes businesses run more profitably and so how are they going to want to use AI. That is one side. The other side is going to be basically the public. Public opinion – how academics are talking, how government is talking, etc. And that’s going to have two different impacts. One is for the workers – businesses generally want their workers to be productive, so they will be paying attention that AI is used in ways that help those workers be more productive and they will also be making the human factor easier for those workers, enabling those workers. Now what about AI in terms of society at large. And this gets to all these ethical questions and my feeling is that corporations generally are not so worried about the impacts of AI at large, they are not so worried about, say, the influencers that AI driven recommendation engines or AI driven information sources, kind of lead to and etc. I think this is where there is the fear of kind of a Terminator future, but where the force of public opinion and governments and academic institutions will have to play an important account or balance.

Short, two nice news articles, I read recently: one feeds the terminator scenario the other feeds the Star Trek scenario. The terminator scenario is that in 2020 it is expected that in Germany one out of five of the elderly people above 55, will live in poverty. That’s in 2020 and its one of the richest countries of the world, which is showing that people cannot keep up. That is the Terminator perspective. Let’s say the Star Trek part is about a new initiative that I read about, it brings together people of 80 years and older, who like to knit. And what they do is that they bring these people together which is fantastic for them to sit to be together and all be knitting and selling this stuff against design prices all over the world, because they are genuine granny wool shawls and what have you, and these people love it. There is almost no production cost and all the money they earn is fed back to these people, which is a fantastic initiative.

So I found these questions not interesting, because the interesting question is how machines will eat the human factor and this is a very broad question and there’s a legal aspect for sure, there’s a political aspect, there’s a cultural aspect and now let me ignore all of these and come back to the technical questions. Actually, I want to come to the earlier question about adoption. Rather than complaining about politicians not adopting, let’s put ourselves in their shoes and see what’s going on out there. A month ago I sat in a room with hospital administrator on one side, researchers on the other side. The researchers were presenting fantastic results about thyroid cancer diagnostics, very nice results. But the hospital administrators did not have time to actually comprehend this, because they have to make real decisions to actually replace that particular procedure in the actual medical diagnostic process, and this is a big effort to do. Also, a few months ago, I read about Stanford Hospital, they have discovered a very effective procedure to detect skin cancers. And they actually got 21 licensed practitioners, the doctors, in the same place and given them same pictures asked them to detect skin cancers and that is compared with the human, I mean the machine and the result is comparable. So now in Stanford’s case, because in the [United] States, I know that they had to go through FDA approval and FDA had go through, it’s a long process. Now I am thinking in these cases the hospital treating in Shanghai or FDA approval panel and what evidence can we presented to show them that this is really good. So now I want to get to the technical side of this, how can we assess the quality of these machines work in replacing humans work? So, do we need another profession, called maybe “machine humans certifier and software tester” to do this. Now getting more detail or closer to this community. I’m inviting the entire panel to speculate what are the technical questions, research questions that we can help to address this equality issue.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 21

So, I’m not sure if I can answer that overall question, but I know a little bit about the state of the art in skin cancer diagnosis. We are working together in the Netherlands with surgeons in this area and as you are saying, there’s a huge increase in the accuracy of automated techniques to diagnose skin cancer indications. There have been all kinds of tests on these accuracy in labs and the results are stunning. The question is of course how and I think that is the way you position the question, how we can assess whether the use of these technologies will actually be effective. So what is happening now, what I’m involved in with a group of skin cancer surgeons in the Netherlands and a health insurance company, is that we are actually going to do a double process, so it’s a methodological approach, that in one area in Eindhoven, where there is a group of family doctors, who are very open to innovation, that we are going to do both paths. So, it is incredibly expensive but we are going to follow the patients through the automated diagnosis, and the same patients going through the traditional process to determine the differences between, if there are any differences in their diagnosis and also the follow-up steps that are being suggested here. So, this is a really a methodological approach, it’s not a technological solution but simply testing in real life, outside of the lab, whether and how these things can be applied. So, in a sense it’s perhaps not the cancer, that you are looking for from a technological perspective but I think there’s a lot of methodological side to it.

So for the blockchain side, this is something that we actually addressed in our report published in June, which was commissioned by the Australian Treasury Department.5 One of the things that we said is: the legislator has to define what the rules are. Like, how can a company provide evidence that this is sufficient for the legislation. If you look at the Australian legal system, a blockchain transaction doesn't have any legal standing as yet, but an email, which is way easier to fake, has the same legal power as a letter with a handwritten signature.

And so the legislators, yes, they need to progress on that. I think yes, I agree with your statement, we basically need a profession of people who work for certification authorities who understand the machine learning and blockchain technologies etc. well enough to be able to make assessments.

Maybe just the entrepreneur’s perspective on that topic. My observation is bringing products to market in the medical space is incredibly hard due to all the regulation, clinical trials that they might have to go through and so on and so on. That's why my observation is that there is a lot less innovation happening, because there's a lot less incentive for people to do so. There's a lot less money around it seems, getting investment for these types of things is incredibly hard. I don't know if the system, like in terms of certification is too strict, I am not an expert for that, but my observation is just that, we don't see the advances we could see due to the nature of that and the smartest people staying out of that area.

So very interesting question, how people trust a machine learning algorithm or an AI output more generally. I think there's going to be an evolution. Part of it is there's a lot of trust for human experts today, even though human experts will often disagree with each other. There's not a uniformity of opinion, when you get a bunch of human experts together. Take this panel as an example. I think part of helping people have more trust in machine learning answers will be making it more clear that human answers are not necessarily uniformly consistent. So, there's kind of a consciousness-raising that will go on.

I think often when people are doing medical procedures, they do try to get a second opinion. That's an illustration of people already saying: “oh, I know that humans don't have the absolute truth.” At the same time I think there's going to be increasing trust in machine learning and AI, as it pervades our culture more and more as it improves. So as automated conversation systems get better, people will start to say: “oh, I see the computer is getting smarter” – this is the Star Trek version of the computer. And also I think as we see that in the physical world like the self-driving cars, which so far have a very low accident rate.

I see an evolution. The final point I'd make is this emphasis now in explanation of AI results. I know DARPA has put a big emphasis on that. I also start to see it around in the research community, so that putting the more human face onto something that was just a black box. This will be another contributor in the positive direction.

So we had various very interesting perspectives. I would now like to invite the panelists to maybe think for a second what could be one takeaway that you want to share with the audience in terms of what we have to do as the BPM community about this; and maybe I take the first turn. What I observe is that, at stages, we have been talking about the perspective of design. And design is something that is not only just technical, but it has a very strong connection with various organizational and psychological

5 http://www.data61.csiro.au/en/Our-Work/Safety-and-security/Secure-Systems-and-Platforms/Blockchain

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

22 How do Machine Learning, Robotic Process Automation and Blockchains Affect the Human Factor in BPM?

questions. In that regard, I'm very happy that we have a BPM management track in the future that helps us to reach out to other communities and bring in these perspectives. What does work in terms of processes, in terms of what do people accept, what do people adopt? I think this is important to reflect, not only what can be done, but also what really works. We as a community should address these questions. Maybe Hajo you want to start: What are your conclusions from these discussions.

So, my conclusion follows up on the first question we got from Gregor. I see the future is bright if experts work together. I believe in, if we have been talking about a topic which of course has IT very central in it, I think computer scientists play a very important role in thinking through what these things will happen but they should work with other experts and I think that is something that the BPM community is especially good at. I think this is also something that Mathias told in his keynote that we always had this broader perspective that we welcome the expertise from other areas to look at the things we are studying. I also know from the people I work with, for example Barbara Weber likes to work with psychologists to address all kinds of interesting issues. So, my takeaway is experts should work together and we as BPM discipline are very well positioned to do so.

I mean technology is exciting and but to me AI, machine learning, blockchain, they are all just part of a toolbox that you should keep in the back of your minds, don't get distracted too much by that. I think the big leverage of BPM is actually the human factor, so that's why I am super excited about this Track 3 to increase the importance of the people side of things and the management side of things in BPM. And to me, being a practitioner out there, this seems to be the biggest barrier to adopting all of the great things that we invent here. It's not so much that we tweak the algorithm from here to there but the bigger challenge is that people out there have to really make it work in the organizations. And I think the BPM community could help a lot more. People won't go away anytime soon, let's focus on them.

Probably allow me to challenge us as a community. Lately, whenever somebody in Australia talks about innovation and process in the same sentence, in excess of 90% percent of the cases they refer to robotic process automation. And I have not seen any papers, any works on this side in this year's conference. So, I think this is a bit of a black spot that we should invest in. When it comes to blockchain, of course, there are so many research opportunities, new business models that can be enabled by them. And I am very excited to be a part of that journey.

Similar to Hajo, I want to come back to Gregor’s first question, which was what can academia be doing to help train the workforce for the future. And I am reminded of something that IBM was pushing three or four years ago, maybe not as successfully as they wanted, but it was so-called Services Science and they said, that we need to think about employees of the future as having a fairly broad knowledge of different aspects of how companies run from maybe economics and management to different engineering disciplines and financial considerations and sales considerations. So kind of a broad knowledge and then also a specific area of depth. So maybe it would be depth in business process management, maybe data management, maybe economic factors, maybe finance factors.

I think that in universities, they do attempt to give that kind of shape. Somebody has a major, maybe it is computer science, but they are required to take a few courses outside their discipline. What's happened, though, is that the schools of engineering, the breadth is limited, and it's often things to do with science, with math, and other kind of technical considerations. Then the schools of Arts and Sciences, the humanities side, they have a notion of breadth and that breadth is history and sociology and languages. I think what could be advantageous is that if engineering schools would rethink what constitutes an appropriate breadth factor for their graduates. They could bring in ethical issues, they could bring in design factor, they could bring in psychology and human factors along with concentrations in technical areas

Thank you very much, and with this, I want to conclude. Please reach out to your colleagues at your universities and to your practitioner partners who can bring in new and complementary perspectives into these discussions. It will be great to see some of these points being raised and analyzed in next year's BPM papers. Thank you very much for being here.

Volume 43 10.17705/1CAIS.043XX Paper XXX

 

Communications of the Association for Information Systems 23

About the Authors

is a Full Professor with the Institute for Information Business at Wirtschaftsuniversität Wien, Austria. His research interests include business process management and information systems. He has published more than 300 research papers and articles, among others in the Journal of the Association of Information Systems, ACM Transactions on Software Engineering and Methodology, IEEE Transaction on Software Engineering, Information Systems, European Journal of Information Systems, and Decision Support Systems. He is a board member of the Austrian Society for Process Management, one of the founders of the Berliner BPM-Offensive, and member of the IEEE Task Force on Process Mining. He is a co-author of the textbooks Fundamentals of Business Process Management and Wirtschaftsinformatik.

is co-founder and CEO of Signavio, a Business Process Management software company headquartered in Berlin, Germany. He was named “Innovator of the Year” by MIT Technology Review and received numerous awards for Signavio as one of the fastest-growing companies in Europe. Prior to Signavio, Gero has worked for SAP and McKinsey. He holds a PhD in Business Process Management from Hasso-Plattner-Institute and is a co-author of “The Process”.

is a Senior Research Scientist at the IBM T.J. Watson Research Laboratory in Yorktown Heights. He received his Ph.D. in Mathematics from the University of California, Berkeley, in 1979 and served as a professor of Computer Science at the University of Southern California for more than a decade. He then spent over a decade at Bell Labs (first as part of Lucent Technologies, and then Alcatel-Lucent). While there his research impacted two products and he became a Bell Labs Fellow and an ACM Fellow. He joined IBM Research in 2008, working initially on data-centric business process and Business Artifacts; these became foundational elements of the IBM Case Manager product and the OMG Case Management Modeling and Notation (CMMN) standard. His current work is focused on infusing AI into BPM and applications of Blockchain. Dr. Hull is co-author of the book “Foundations of Databases” (1996), holds 12 US patents, and has published over 150 articles in refereed journals and conferences.

is a Full Professor of Business Informatics at the Vrije Universiteit Amsterdam, the Netherlands. He also holds a position as part-time, full professor at Eindhoven University of Technology. Previously, he worked as a management consultant in the BPM field for Deloitte Consulting and led an R&D team within Lexmark. On his topics of interest, such as process innovation and conceptual modeling, he published over 200 scientific articles, chapters in books, and professional viewpoints. He is one of the founders of the Business Process Management Forum, a Dutch platform for the exchange of knowledge between industry and academia.

is a Principal Research Scientist & Team Leader of the Architecture & Analytics Platforms (AAP) team at Data61, CSIRO in Sydney. In addition, he is a Conjoint Associate Professor at the University of New South Wales (UNSW) and an Adjunct Associate Professor at Swinburne University. He has published over 80 refereed papers and two books. Prior to Data61, CSIRO, Ingo worked for UNSW in Sydney, Australia, and at SAP Research in Germany. While at SAP, he completed his PhD with the University of Karlsruhe (TH). He also holds an MSc from the University of Massachusetts, Amherst, USA.

Copyright © 2018 by the Association for Information Systems. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and full citation on the first page. Copyright for components of this work owned by others than the Association for Information Systems must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists requires prior specific permission and/or fee. Request permission to publish from: AIS Administrative Office, P.O. Box 2712 Atlanta, GA, 30301-2712 Attn: Reprints or via e-mail from publications@aisnet.org.

Volume 43 10.17705/1CAIS.043XX Paper XXX