Current Research Topics/Projects

The line of research pursued by my research group focuses on developing various alternative formulations (preferably convex) for some of the fundamental system design and real time processing problems in the area of communication and signal processing, and designing efficient optimization algorithms to solve or approximate them. The research outlined here is expected to advance significantly the state of the art of digital signal processing and data communication, both of which are experiencing tremendous growth worldwide. The major problems under investigation by my research group are:

1.      The application of Interior Point Least Square (IPLS) algorithm to adaptive filtering

A novel approach to adaptive filtering based on interior point optimization techniques has been proposed recently by my research group. The algorithm's asymptotic convergence was shown and its fast transient performance was demonstrated (both experimentally and theoretically) in the context of a system identification application. It turns out that for both RLS (Recursive Least Square) and IPLS, the transient behavior is dominated by the mis-adjustment due to initialization of the regularization term. While RLS eliminates the mis-adjustment at rate O(1/n), with n being the number of samples, IPLS does so at an exponential rate. To demonstrate the full potential offered by IPLS, it will be necessary to further reduce the per-sample computational complexity so that it is comparable to RLS (currently IPLS has a slightly higher theoretical per-sample complexity than RLS). Also, applications of IPLS to equalization (particularly when short training sequences are used), to adaptive space-time processing for multi-antenna systems and to multi-target tracking are promising avenues of further research.

2.      The application of Semidefinite Programming (SDP) to robust filtering and blind equalization

My research group proposes to apply robust optimization techniques to the blind beamforming and equalization problems in digital communication. In practical communication environment, the exact knowledge of channel state information and signal model is difficult, if not impossible, to obtain. The success of these robust methods can help reduce the dependence of the standard beamforming and equalization algorithms on the knowledge of the channel and signal model. For example, in designing a CDMA receiver, one often only knows the undistorted signature waveform of the desired user. In the situation where unknown multi-access interference is present and the channel experiences multi-path fading, simply using the desired user’s signature waveform to linearly de-correlate the received signal offers a poor performance. In fact, even using the minimum output energy criterion of Verdu to design the linear receiver is inadequate because of the channel distortion. Fortunately robust optimization techniques can be used to design a robust CDMA receiver that minimizes the worst-case performance degradation. Other problems currently under investigation include the use of Semidefinite Programming (SDP) to design recursive filters that are robust against modeling errors, and convex reformulation of the CMA (Constant Modulus) blind equalizer. The latter is a powerful and popular blind equalization approach but suffers from local minimums and non-convergence. A convex reformulation of CMA condition can mitigate this major deficiency.

3.      The application of SDP to transceiver design for a multi-user communication system

Multicarrier modulation includes transmission schemes whose main feature is the decomposition of any wide-band channel into a set of independent narrowband parallel channels. Within this family, the most commonly used schemes are Discrete MultiTone (DMT) and Orthogonal Frequency Division Multiplexing (OFDM). These schemes employ the Fast Discrete Fourier Transform (DFT) at the transmitter and the receiver, resulting in a low cost, high-performance implementation. Many of the xDSL over wired media use DMT and Digital Audio Broadcast (DAB) or Digital Video Broadcast (DVB) for wireless communications use OFDM. To achieve the capacity of a spectrally shaped Gaussian channel in a single user case, it is well known one must use appropriate bit and power allocation among the subcarriers for the single user in a way that corresponds to the classic water-filling distribution. However, in the multi-user case, achieving multiuser capacity requires more sophisticated resource allocation. Simply using TDMA or FDMA non-overlapping resource allocation schemes in an arbitrary way will result in multiuser rate far below capacity. The goal of this research is to devise optimal resource allocation schemes in the MMSE (Minimum Mean Squared Error) sense for multi-user communication systems.  

4.      Efficient optimal/sub-optimal solution methods for space-time processing in wireless communication

Recently the use of multiple antennas at both the transmitter and the receiver has been shown to provide several-fold increase in capacity. In a rich scattering environment such as indoor wireless scenario, the channel gains between any pair of transmitter antenna and receiver antenna can be assumed independent. In this case, the capacity increase is shown to be proportional to the number of transmit antennas. To reap the benefits brought by the MIMO (Multi-Input-Multi-Output) channel, one must develop efficient decoding algorithm used in such a system, especially for cases where channel matrix is ill conditioned. My research group will explore robust optimal designs for the space-time linear precoders and decoders for a MIMO communication system. The existing V-Blast type of receiver algorithm developed in Bell Labs suffers from significant performance loss when the channel matrix is ill conditioned. My group proposes to investigate new transceiver/receiver structures and algorithms that are less sensitive to the ill conditioning and the estimation errors of the channel matrix and yet still computationally efficient. The state of the art robust optimization techniques (such robust linear programming) will be used. My research group also intends to explore various robust space-time transmitter/receiver schemes using the minimum symbol mean square error and the (approximate) maximization of the minimum distance between symbol hypotheses. Furthermore, the scalability issue with respect to the number of antennas, size of the coding block and transmit average/peak power will be investigated. The goal is to turn these robust design formulations into convex optimization problems and devise highly efficient interior point algorithms to solve them. Some of the recent successes along this line include the work from Cioffi's group at Stanford University. The design criteria to be used include the (weighted) Minimum Mean Square Error (MMSE) as well as Bit Error Rate (BER). This work has received initial funding from CITO (Communications Information Technology Ontario, the provincial research center of excellence).

5.      Data fusion via convex optimization

The problem of data/information fusion arises in many applications including, for example, the multi-sensor target tracking whereby sensors generate reports on the target identities that are subsequently combined in a fusion center. Since the sensor reports are typically fuzzy, ``incomplete'' and inconsistent, the fusion of such sensor reports becomes a major challenge. My group is currently developing new identity information fusion methods based on several criteria: the minimization of inconsistencies between the sensor reports by using convex Quadratic Programming (QP) and linear programming (LP), the minimization of the inconsistency of reports augmented with a logarithmic barrier for the nonnegativity constraint, and the minimization of a regularized Kullback-Leibler distance. In contrast to the Dempster-Shafer's evidential reasoning approach that suffers from exponentially growing complexity, these approaches are highly efficient (polynomial time solvable). Moreover, these new approaches are capable of fusing ``Ratio type'' sensor reports. Thus they are more general than the evidential reasoning theory. When the sensor reports are consistent, the solution generated by the new fusion methods can be shown to converge to the true probability distribution. Various theoretical properties of these new fusion approaches as well as the classical Baysian and Dempster-Shafer approaches are being analyzed and compared. This work has been funded by CITO and by DREV (Defense Research Establishment at Valcartier, a research arm of the Department of Defense of Canada).

 

6.      Efficient optimization methods for the intelligent provisioning and management of optical networks

In an optical network, one needs to accurately assess resource availability before accepting a new service order. By including wavelengths in the resource pool, the provider can make transport an essential part of each service. An optical network provisioning system can check, for example, the availability of unused wavelengths and IP address blocks before turning on a broadband Internet service. With this capability, innovative network operators can use advanced asset management to create a variety of new, revenue-enhancing services and to identify and resolve problems before they impede service delivery. My group plans to develop a mathematical model for the resource allocation problem and propose efficient algorithms to perform real-time asset management of optical networks. A main goal of the proposed automated real-time asset management system is to help service providers get the most out of their transport networks. Even though DWDM (Dense Wave Division Multiplexing) adds tremendous capacity to the wide-area infrastructure, networks will still be congested. By maintaining up-to-the-millisecond records of bandwidth utilization and dynamically reconfiguring the networks when necessary, real-time asset management will ensure that each wavelength is used as efficiently as possible. My group will examine various issues relating to traffic management. For example, it plans to investigate efficient ways of dynamically reconfiguring a network in order to reduce network congestion. As a start, some simple network structures (such as a ring network) will be considered first. Various combinatorial optimization and integer programming techniques (such as linear programming relaxation, Lagrangian relaxation) will be explored in solving some of the NP-hard formulations of the optical network management problems. Software prototypes will also be developed in collaboration with Nortel Networks.

There are a number of major theoretical issues arising from the investigation of above problems, three of which are listed below.

·         Stochastic convergence and rate of convergence of IPLS (Interior Point Least Square) or its variants in a time-varying environment

The objective is to provide a rigorous and comprehensive study of advanced interior point methods in the context of real time signal processing and communication applications where data are collected dynamically and are corrupted by noise. Previously interior point methods have only been studied in the noise-free deterministic setting. To demonstrate the effectiveness of IPLS algorithm in a time-varying environment, it is proposed that the tracking behavior of IPLS be theoretically analyzed for some specific noise and system model. For example, one can assume the noise is Gaussian and the system undergoes periodic changes with a given frequency. For such environment, it should be possible to derive the tracking error estimates for RLS (Recursive Least Square) and IPLS in terms of the sample size and the rate at which the system varies. It is expected that IPLS will generate a smaller tracking error than RLS in such time-varying environment.

·         Convex reformulation of transceiver design problem

Substantial progress has been made by my research group in solving the MMSE transceiver design problem for a multi-access network. Indeed, even though the straightforward formulation of the MMSE (Minimum Mean Square Error) joint transmitter-receiver design problem for a multi-access communication system is nonconvex, various alternative (and equivalent) formulations can be constructed using techniques of Linear Matrix Inequalities (LMIs), Second Order Cone programming and geometric programming. These formulations can be solved using the highly efficient interior point methods. However, for the downlink portion of a multi-user communication network, the current MMSE transceiver design results no longer hold due to a different channel model. Also, the issue of how to deal with dynamic changes in the network and how to model the QoS (Quality of Service) are all important questions that are beginning to be addressed by my research group.

·          Linear Matrix Inequality (LMI) formulation of certain convex cones

An important technique in reformulating a nonconvex optimization problem as an SDP is the use of the classical S-procedure by Yakubovich in system’s theory. In this procedure, one can convert the problem of minimizing a nonconvex quadratic function over a region described by another quadratic polynomial inequality into an SDP. The S-procedure can also be seen as a consequence of Nesterov’s recent result that gives a sufficient condition on when a convex cone can be described by LMIs. In many communications applications, one needs to minimize a nonconvex quadratic polynomial over the intersection of two ellipsoids. It is not clear when and if this problem can also be converted to an SDP and thus solved in polynomial time. The answer to this question is related to how to describe the convex cone of nonnegative quadratic polynomials that are nonnegative over the intersection of two ellipsoids by LMIs.

The positive resolution to this question will have major consequences in several areas including control, communication system design, and the study of trust region methods in optimization.