Electrical and Computer Engineering
Permanent URI for this community
Browse
Browsing Electrical and Computer Engineering by Author "Aazhang, Behnaam"
Now showing 1 - 20 of 214
Results Per Page
Sort Options
Item A Data and Platform-Aware Framework For Large-Scale Machine Learning(2015-04-24) Mirhoseini, Azalia; Koushanfar, Farinaz; Aazhang, Behnaam; Baraniuk, Richard; Jermaine, ChristopherThis thesis introduces a novel framework for execution of a broad class of iterative machine learning algorithms on massive and dense (non-sparse) datasets. Several classes of critical and fast-growing data, including image and video content, contain dense dependencies. Current pursuits are overwhelmed by the excessive computation, memory access, and inter-processor communication overhead incurred by processing dense data. On the one hand, solutions that employ data-aware processing techniques produce transformations that are oblivious to the overhead created on the underlying computing platform. On the other hand, solutions that leverage platform-aware approaches do not exploit the non-apparent data geometry. My work is the first to develop a comprehensive data- and platform-aware solution that provably optimizes the cost (in terms of runtime, energy, power, and memory usage) of iterative learning analysis on dense data. My solution is founded on a novel tunable data transformation methodology that can be customized with respect to the underlying computing resources and constraints. My key contributions include: (i) introducing a scalable and parametric data transformation methodology that leverages coarse-grained parallelism in the data to create versatile and tunable data representations, (ii) developing automated methods for quantifying platform-specific computing costs in distributed settings, (iii) devising optimally-bounded partitioning and distributed flow scheduling techniques for running iterative updates on dense correlation matrices, (iv) devising methods that enable transforming and learning on streaming dense data, and (v) providing user-friendly open-source APIs that facilitate adoption of my solution on multiple platforms including (multi-core and many-core) CPUs and FPGAs. Several learning algorithms such as regularized regression, cone optimization, and power iteration can be readily solved using my APIs. My solutions are evaluated on a number of learning applications including image classification, super-resolution, and denoising. I perform experiments on various real-world datasets with up to 5 billion non-zeros on a range of computing platforms including Intel i7 CPUs, Amazon EC2, IBM iDataPlex, and Xilinx Virtex-6 FPGAs. I demonstrate that my framework can achieve up to 2 orders of magnitude performance improvement in comparison with current state-of-the-art solutions.Item A globally convergent algorithm for training multilayer perceptrons for data classification and interpolation(1991) Madyastha, Raghavendra K.; Aazhang, BehnaamThis thesis addresses the issue of applying a "globally" convergent optimization scheme to the training of multi-layer perceptrons, a class of Artificial Neural Networks, for the detection and classification of signals in single- and multi-user communication systems. The research is motivated by the fact that a multi-layer perceptron is theoretically capable of approximating any nonlinear function to within any specified accuracy. The object function to which we apply the optimization algorithm is the error function of the multilayer perceptron, i.e., the average of the sum of the squares of the differences between the actual and the desired outputs to specified inputs. Until recently, the most widely used training algorithm has been the Backward Error Propagation algorithm, which is based on the algorithm for "steepest descent" and hence, is at best linearly convergent. The algorithm discussed here combines the merits of two well known "global" algorithms--the Conjugate Gradients and the Trust Region algorithms. A further technique known as preconditioning is used to speed up the convergence by clustering the eigenvalues of the "effective Hessian". The Preconditioned Conjugate Gradients--Trust Regions algorithm is found to be superlinearly convergent and hence, outperforms the standard backpropagation routine.Item A hybrid relaying protocol for the parallel-relay network(2010) Summerson, Samantha Rose; Aazhang, BehnaamCooperation among radios in wireless networks has been shown to improve communication in several aspects. We analyze a wireless network which employs multiple parallel relay transceivers to assist in communication between a single source-destination pair, demonstrating that gains are achieved when a random subset of relays is selected. We derive threshold values for the received signal-to-noise ratios (SNRs) at the relays based on outage probabilities; these thresholds essentially determine the active subset of relays in each time frame for our parallel relay network; due the random nature of wireless channels, this active subset is a random. Two established forwarding protocols for the relays, Amplify-and-Forward and Decode-and-Forward, are combined to create a hybrid relaying protocol which is analyzed in conjunction with both regenerative coding and distributed space-time coding at the relays. Finally, the allocation of power resources to minimize the end-to-end probability of outage is considered.Item A Matter of Perspective: Reliable Communication and Coping with Interference with Only Local Views(2012-09-05) Kao, David; Sabharwal, Ashutosh; Aazhang, Behnaam; Knightly, Edward W.; Tapia, Richard A.; Chiang, MungThis dissertation studies interference in wireless networks. Interference results from multiple simultaneous attempts to communicate, often between unassociated sources and receivers, preventing extensive coordination. Moreover, in practical wireless networks, learning network state is inherently expensive, and nodes often have incomplete and mismatched views of the network. The fundamental communication limits of a network with such views is unknown. To address this, we present a local view model which captures asymmetries in node knowledge. Our local view model does not rely on accurate knowledge of an underlying probability distribution governing network state. Therefore, we can make robust statements about the fundamental limits of communication when the channel is quasi-static or the actual distribution of state is unknown: commonly faced scenarios in modern commercial networks. For each local view, channel state parameters are either perfectly known or completely unknown. While we propose no mechanism for network learning, a local view represents the result of some such mechanism. We apply the local view model to study the two-user Gaussian interference channel: the smallest building block of any interference network. All seven possible local views are studied, and we find that for five of the seven, there exists no policy or protocol that universally outperforms time-division multiplexing (TDM), justifying the orthogonalized approach of many deployed systems. For two of the seven views, TDM-beating performance is possible with use of opportunistic schemes where opportunities are revealed by the local view. We then study how message cooperation --- either at transmitters or receivers --- increases capacity in the local view two-user Gaussian interference channel. The cooperative setup is particularly appropriate for modeling next-generation cellular networks, where costs to share message data among base stations is low relative to costs to learn channel coefficients. For the cooperative setting, we find: (1) opportunistic approaches are still needed to outperform TDM, but (2) opportunities are more abundant and revealed by more local views. For all cases studied, we characterize the capacity region to within some known gap, enabling computation of the generalized degrees of freedom region, a visualization of spatial channel resource usage efficiency.Item A Resource-Aware Streaming-based Framework for Big Data Analysis(2015-12-02) Darvish Rouhani, Bita; Koushanfar, Farinaz; Aazhang, Behnaam; Baraniuk, RichardThe ever growing body of digital data is challenging conventional analytical techniques in machine learning, computer vision, and signal processing. Traditional analytical methods have been mainly developed based on the assumption that designers can work with data within the confines of their own computing environment. The growth of big data, however, is changing that paradigm especially in scenarios where severe memory and computational resource constraints exist. This thesis aims at addressing major challenges in big data learning problem by devising a new customizable computing framework that holistically takes into account the data structure and underlying platform constraints. It targets a widely used class of analytical algorithms that model the data dependencies by iteratively updating a set of matrix parameters, including but not limited to most regression methods, expectation maximization, and stochastic optimizations, as well as the emerging deep learning techniques. The key to our approach is a customizable, streaming-based data projection methodology that adaptively transforms data into a new lower-dimensional embedding by simultaneously considering both data and hardware characteristics. It enables scalable data analysis and rapid prototyping of an arbitrary matrix-based learning task using a sparse-approximation of the collection that is constantly updated inline with the data arrival. Our work is supported by a set of user-friendly Application Programming Interfaces (APIs) that ensure automated adaptation of the proposed framework to various datasets and System on Chip (SoC) platforms including CPUs, GPUs, and FPGAs. Proof of concept evaluations using a variety of large contemporary datasets corroborate the practicability and scalability of our approach in resource-limited settings. For instance, our results demonstrate 50-fold improvement over the best known prior-art in terms of memory, energy, power, and runtime for training and execution of deep learning models in deployment of different sensing applications including indoor localization and speech recognition on constrained embedded platforms used in today's IoT enabled devices such as autonomous vehicles, robots, and smartphone.Item A sample realization approach for optimization of code division multiple access systems(1994) Mandayam, Narayan B.T.; Aazhang, BehnaamEfforts in performance analysis of Code Division Multiple Access (CDMA) systems have concentrated on obtaining asymptotic approximations and bounds for system error probabilities. As such, these cannot capture the sensitivities of the system performance to any class of parameters, and the optimization of such systems (with respect to any class of parameters) presents itself to no analytical solutions. A discrete event dynamic systems (DEDS) formulation is developed for CDMA systems whereby the sensitivity of the average probability of error can be evaluated with respect to a wide class of system parameters via sample path based gradient estimation techniques like infinitesimal perturbation analysis (IPA) and the likelihood ratio (LR) method. Appropriate choice of the sample path and the corresponding sample performance function leads to analyzing the sensitivity of the average probability of error to near-far effects, power control, and code parameters. Further, these sensitivity analysis methods are incorporated in gradient algorithms for optimizing system performance in terms of the minimum probability of detection error. Specifically, for direct-sequence CDMA systems, IPA based stochastic gradient algorithms are used to develop a class of adaptive linear detectors that are optimum in that they minimize the average probability of bit-error. These detectors outperform both the matched filter and MMSE detectors, and also alleviate the disadvantage of multiuser detection schemes that require implicit information on the multiple access interference. For CDMA systems in the optical domain, IPA based stochastic algorithms are used to develop a class of adaptive threshold detectors that minimize the average probability of bit-error. These detectors outperform the correlation detector and also preclude the need for assumptions on the interference statistics required by existing optimum one-shot detectors. All adaptive detection schemes developed here are easily implementable owing to the simple recursive structures that arise out of our sample realization based approach. The sequential versions of the adaptive detectors developed in here require no preamble, which makes them a viable choice for CDMA channels subject to temporal variations due to dispersion effects and variable number of users in the channel.Item Achievable Rates for Arbitrary Network Topologies with â Cheapâ Nodes(2003-05-01) Khojastepour, Mohammad; Sabharwal, Ashutosh; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)In this paper, we derive achievable rates for arbitrary network topologies consisting of â cheapâ nodes. A node is labeled â cheapâ if its radio can only operate in TDD mode when transmitting and receiving in the same frequency band. Two main results are shown. The first result provides an achievable rate for the channel with either continuous alphabet or discrete channel. The second result provides the achievable rate for the Gaussian channel with average power constraint. The two results are applied to the case of Gaussian relay channel and concatenated channel with cheap nodesItem Addressing indirect frequency coupling via partial generalized coherence(Springer Nature, 2021) Young, Joseph; Homma, Ryota; Aazhang, BehnaamDistinguishing between direct and indirect frequency coupling is an important aspect of functional connectivity analyses because this distinction can determine if two brain regions are directly connected. Although partial coherence quantifies partial frequency coupling in the linear Gaussian case, we introduce a general framework that can address even the nonlinear and non-Gaussian case. Our technique, partial generalized coherence (PGC), expands prior work by allowing pairwise frequency coupling analyses to be conditioned on other processes, enabling model-free partial frequency coupling results. By taking advantage of recent advances in conditional mutual information estimation, we are able to implement our technique in a way that scales well with dimensionality, making it possible to condition on many processes and produce a partial frequency coupling graph. We analyzed both linear Gaussian and nonlinear simulated networks. We then performed PGC analysis of calcium recordings from mouse olfactory bulb glomeruli under anesthesia and quantified the dominant influence of breathing-related activity on the pairwise relationships between glomeruli for breathing-related frequencies. Overall, we introduce a technique capable of eliminating indirect frequency coupling in a model-free way, empowering future research to correct for potentially misleading frequency interactions in functional connectivity analyses.Item Advanced techniques for next-generation wireless systems(1999) Sendonaris, Andrew; Aazhang, BehnaamIn order to meet the demands of next-generation wireless systems, which will be required to support multirate multimedia at high data rates, it is necessary to employ advanced algorithms and techniques that enable the system to guarantee the quality of service desired by the various media classes. In this work, we present a few novel methods for improving wireless system performance and achieving next-generation goals. Our proposed methods include finding signal sets that are designed for fading channels and support multirate, exploiting knowledge of the fading statistics during the data detection process, exploiting the existence of Doppler in the received signal, and allowing mobile users to cooperate in order to send their information to the base station. We evaluate the performance of our proposed ideas and show that they provide gains with respect to conventional systems. The benefits include multirate support, higher data rates, and more stable data rates. It should be mentioned that, while we focus mainly on a CDMA framework for analyzing our ideas, many of these ideas may also be applied to other wireless system environments.Item All-optical CDMA with bipolar codes(1995-03-20) Nguyen, Lim; Aazhang, Behnaam; Young, James F.; Center for Multimedia Communications (http://cmc.rice.edu/)A method for the transmission and detection of bipolar sequences in a unipolar system is presented. It allows all-optical implementation, in noncoherent optical CDMA systems, of the bipolar codes that have been developed for the radio domain. A practical design is described that encodes the spectrum of a broadband optical source to support a large number of subscribers.Item Analysis of Decision-Feedback Based Broadband OFDM Systems(2005-11-01) de Baynast, Alexandre; Sabharwal, Ashutosh; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)In wireless communications, about 25% of the bandwidth is dedicated to training symbols for channel estimation. By using a semi-blind approach, the training sequence length can be reduced while improving performance. The principle is as follows: the detected symbols (hard decision) are fed back to the channel estimator in order to re-estimate the channel more accurately. However, semi-blind approach can significantly deteriorate the performance if the bit error rate is high. In this paper, we propose to determine analytically the minimum Signal to Noise Ratio (SNR) from which a semi-blind method starts to outperform a training sequence based only system.Item Antenna arrays for wireless CDMA communication systems(1997) Madyastha, Raghavendra K.; Aazhang, BehnaamThe estimation of code delays along with amplitudes and phases of different users constitutes the first stage in the demodulation process in a CDMA communication system. The delay estimation stage is termed the acquisition stage and forms the bottleneck for the detection of users' bitstreams; accurate detection necessitates accurate acquisition. Most existing schemes incorporate a single sensor at the receiver, which leads to an inherent limit in the acquisition based capacity, which is the number of users that can be simultaneously acquired. In this thesis we combine the benefits of spatial processing in the form of an antenna array at the receiver along with code diversity to gain an increase in the capacity of the system. An additional parameter to be estimated now is the direction of arrival (DOA) of each user. We demonstrate the gains in parameter estimation with the incorporation of spatial diversity. We propose two classes of delay-DOA estimation algorithms--a maximum likelihood algorithm and a subspace based algorithm (MUSIC). With reasonable assumptions on the system we are able to derive computationally efficient estimation algorithms and demonstrate the gains achieved in exploiting multiple sensors at the receiver. In addition, we also investigate the benefits of spatial diversity in linear multiuser detection. We consider two linear multiuser detectors, the decorrelating detector and the linear MMSE detector (chosen for their near-far properties) and characterize the performance increase in the multisensor case. We observe that in many cases, the gain can be directly captured in terms of the number of sensors in the array.Item Antenna Packing in Low Power Systems: Communication Limits and Array Design(2008) Muharemovic, Tarik; Sabharwal, Ashutosh; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)In this paper, we study design of transceiver antenna arrays and its impact on spectral efficiency of low-power systems. Our primary motivation is construction of practical and portable multi-antenna configurations with a very small and a-priori fixed volume for placing antennas. Using spectral efficiency as a target metric for array optimization, we show that any array configuration, transmit or receive, can be characterized via a parameter that we interpret as "effective degrees of freedom." For any array configuration, effective degrees of freedom describes an equivalent uncorrelated array, which results in the same low-power behavior of spectral efficiency. Joint optimization of transmit and receive antenna configurations decouples into maximizing effective degrees of freedom for transmitter and receiver separately. To achieve this goal, we introduce and study a theoretical benchmark of "limiting degrees of freedom," which is the least upper bound on effective degrees of freedom, evaluated over all configurations with finite number of antennas. Limiting degrees of freedom therefore describes the best possible performance for any transceiver array which confines its elements inside a given space. We compute a closed-form expression for limiting degrees of freedom of a circular geometry. Finally, we present numerical procedure and examples for designing linear and square arrays with non-uniform spacing, which typically exhibit significant spectral efficiency gains over uniform arrays.Item An Approach to Capacity Analysis of Coarsely Coordinated Low Power Multiple Access Systems(2004-07-01) Muharemovic, Tarik; Sabharwal, Ashutosh; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)We consider multiaccess problem in low power systems, where we allow each user to select its own data rate and transmit power locally and independently from other users. Here, every user has a set of low power codebooks, labeled a policy, which accomodates a range of small spectral efficiencies, while treating instantaneous data rates of other users as an unknown compound parameter. Even with such coarse user coordination, multiuser detection enables a system which is superior to any classic orthogonal division system. First we fully characterise the set of achievable policies, after which we demonstrate that in multiantenna systems, policies are be viewed as awarding protected receiver spatial dimensions to each user.Item Bayesian Blind PARAFAC Recievers forDS-CDMA Systems(2003-10-01) de Baynast, Alexandre; Declercq, David; De Lathauwer, Lieven; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)In this paper an original Bayesian approach for blind detec-tion for Code Division Multiple Access (CDMA) Systems in presence of spatial diversity at the receiver is developed. In the noiseless context, the blind detection/identification problem relies on the canonical decomposition (also re-ferred as Parallel Factor analysis [Sidiropoulos, IEEE SP 00], PARAFAC. The author in [Bro,INCINC 96] pro-poses a suboptimal solution in least-squares sense. How-ever, poor performance are obtained in presence of high noise level. The recently emerged Markov chain Monte Carlo (MCMC) signal processing method provide a novel paradigm for tackling this problem. Simulation results are presented to demonstrate the effectiveness of this method.Item Beamformer Design with Feedback Rate Constraints : Criteria and Constructions(2003-07-20) Mukkavilli, Krishna Kiran; Sabharwal, Ashutosh; Erkip, Elza; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)In this work, we provide a geometrical framework for the analysis and design of beamformer codebooks with finite number of beamformer vectors. We present a design criterion for good beamfomer codebooks and show the equivalence of the beamformer design problem to two other known problems. First, the beamformer design problem can be directly posed as a problem of packing 2 dimensional subspaces in a 2t dimensional Grassmannian manifold, t being the number of transmit antennas. And second, under certain conditions, the beamformer design problem is equivalent to the construction of unitary space time constellations.Item BER of optical communication system using fiber source(1995-11-20) Nguyen, Lim; Aazhang, Behnaam; Young, James F.; Center for Multimedia Communications (http://cmc.rice.edu/)We analyze the bit-error rate (BER) of an optical communication system using the superfluorescent fiber source (SFS). The counting statistics of thermal light give improved performance relative to the Gaussian statistics that predict a BER floor. We consider a spectrum-sliced wavelength-division multiple access (WDMA) system that employs the SFS.Item Beyond Interference Avoidance: Distributed Sun-network Scheduling in Wireless Networks with Local Views(2013-09-16) Santacruz, Pedro; Sabharwal, Ashutosh; Aazhang, Behnaam; Knightly, Edward W.; Hicks, Illya V.In most wireless networks, nodes have only limited local information about the state of the network, which includes connectivity and channel state information. With limited local information about the network, each node’s knowledge is mismatched; therefore, they must make distributed decisions. In this thesis, we pose the following question - if every node has network state information only about a small neighborhood, how and when should nodes choose to transmit? While link scheduling answers the above question for point-to-point physical layers which are designed for an interference-avoidance paradigm, we look for answers in cases when interference can be embraced by advanced code design, as suggested by results in network information theory. To make progress on this challenging problem, we propose two constructive distributed algorithms, one conservative and one aggressive, which achieve rates higher than link scheduling based on interference avoidance, especially if each node knows more than one hop of network state information. Both algorithms schedule sub-networks such that each sub-network can employ advanced interference-embracing coding schemes to achieve higher rates. Our innovation is in the identification, selection and scheduling of sub-networks, especially when sub-networks are larger than a single link. Using normalized sum-rate as the metric of network performance, we prove that the proposed conservative sub-network scheduling algorithm is guaranteed to have performance greater than or equal to pure coloring-based link scheduling. In addition, the proposed aggressive sub-network scheduling algorithm is shown, through simulations, to achieve better normalized sum-rate than the conservative algorithm for several network classes. Our results highlight the advantages of extending the design space of possible scheduling strategies to include those that leverage local network information.Item Blind PARAFAC Receivers for Multiple Access-Multiple Antenna Systems(2003-10-01) de Baynast, Alexandre; De Lathauwer, Lieven; Aazhang, Behnaam; Center for Multimedia Communications (http://cmc.rice.edu/)In this paper, we present a new blind receiver for multiple access channel with multiple transmit antennas per user and multiple receive antennas (MIMO channel). After being multiplied by a spreading sequence, each user s data is split into Nt streams that are simultaneously transmitted using Nt transmit antennas. The received signal at each receive antenna is a linear superposition of the Nt transmitted signals of the Nu users perturbed by noise. We propose a new blind detection/identification algorithm under the assumption that the fading is slow and frequency non-selective. This algorithm relies on a generalization of parallel factor analysis (PARAFAC analysis, [Kruskal, Lin. Alg. Appl. 77, Sidiropoulos, Tr. on Sig. Proc. 00]): we show that a generalized canonical decomposition (CANDECOMP) of the 3D data tensor is unique under mild assumptions without noise. Neither algebraic orthogonality nor independence between sources is needed for uniqueness of the decomposition. By performing this decomposi-tion, in rank-(Nt,Nt,1) terms, we are able to retrieve the three sets of parameters: the symbols, the channel fading coefficients (including the antenna gains) and the spreading sequences. In a noisy context, we propose a simple algorithm of the alternating least squares (ALS) type, which yields a performance close to the linear minimum mean square error (LMMSE) receiver which requires knowledge of the channel and spreading sequences.Item Bounds on Achievable Rates for General Multi-terminal Networks with Practical Constraints(2003-04-20) Khojastepour, Mohammad; Aazhang, Behnaam; Sabharwal, Ashutosh; Center for Multimedia Communications (http://cmc.rice.edu/)We consider the problem of communication in a general multi-terminal network where each node of the network is a potential sender or receiver (or both) but it cannot do both functions together. The motivation for this assumption comes from the fact that current radios in sensor nodes operate in TDD mode when the transmitting and receiving frequencies are the same. We label such a radio as a cheap radio and the corresponding node of the network as a cheap node. We derive bounds on the achievable rates in a general multi-terminal network with finite number of states. The derived bounds coincide with the known cut-set bound of network information theory if the network has just one state. Also, the bounds trivially hold in the network with cheap nodes because such a network operates in a finite number of states when the number of nodes is finite. As an example, application of these bounds in the multi-hop network and the relay channel with cheap nodes is presented. In both of these cases, the bounds are tight enough to provide converses for the coding theorems, and thus their respective capacities are derived.