The Digital Signal Processing Lab @ UCSD


Title:Multimodal Dynamic Imaging of Human Brain Activity (PI: Scott Makeig, Co-PI’s: Profs. K. Kreutz-Delgado and Bhaskar Rao)

Source: National Science Foundation

Period: 10/01/06 – 09/30/09

Specific Aim:  To apply innovative computational methods based on statistical learning theory for analyzing the dynamics of activity in large-scale neuronal networks collected non-invasively from the human brain via simultaneous, high-density magnetoencephalographic (MEG) and electroencephalographic (EEG) recordings during performance of complex tasks.

Abstract:  This project will develop an ongoing collaboration among researchers, postdoctoral fellows and graduate students in the Swartz Center for Computational Neuroscience and the department of Electrical and Computer Engineering, UCSD, to develop, test and apply innovative computational methods based on statistical learning theory for analyzing the dynamics of activity in large-scale neuronal networks collected non-invasively from the human brain via simultaneous, high-density magnetoencephalographic (MEG) and electroencephalographic (EEG) recordings during performance of complex tasks. The investigators will combine their expertise in human brain dynamics, inverse source modeling, and information-based statistics to develop and test methods for jointly modeling relationships between recorded electromagnetic signals, brain source dynamics, cognitive performance, and preceding event context.

      The proposed approach builds on applications of independent component analysis to EEG data pioneered by the principal investigator and colleagues. The goal will be to approach a solution to the brain dynamics inverse problem, to identify and model the macroscopic brain dynamics accompanying motivated human behavior and responses to events in tasks requiring active search guided by performance feedback,. To this end, the project will collaboratively develop algorithms for fitting more general time-domain and frequency-domain models of spatially stationary and non-stationary patterns of electromagnetic brain activity, and will test the ability of these algorithms to model the spatiotemporal dynamics of the electromagnetic source processes they identify.

The project will also further develop mathematically, simulate computationally, and apply critically to experimental data an innovative approach to electromagnetic spatial source estimation based on high-density MEG and/or EEG recordings with structural magnetic resonance (MR) head images. The goal will be to obtain more adequate and detailed solutions to the physical brain inverse problem of reconstructing the locations and spatial extents of areas of locally synchronized current density that generate recorded MEG and EEG data. This new approach applies adaptive sparse Bayesian learning to an overcomplete dictionary of electromagnetic source elements, modeling the joint MEG and EEG data as mixtures of electromagnetic fields generated by synchronous current density within a sparse set of overlapping spatial domains..

Title: Space-Time Processing for Tactical Mobile and Ad-Hoc Networks (A Multi-University team, PI: Prof. J. Zeidler)

Source: Army Research Office (ARO/MURI)

Period: 06/01/04 – 04/30/09

Specific Aim: The proposed research will develop cross-layer, energy-efficient multiple-input multiple-output (MIMO) signal processing algorithms for mobile, multi-user ad-hoc networks employing directional antenna arrays for tactical applications. Specifically, at the physical layer we plan to investigate signaling issues, the beam-forming (BF) versus space-time coding (STC) trade-off, channel state estimation in an interference-limited environment, as well as distributed wireless (virtual) antennas. Following a cross-layer design approach, at the medium access control (MAC) layer we will take advantage of transmission-rate adaptiveness, BF antennas, location information, and scheduling of transmissions within the context of STC. Finally, at the networking layer we will address energy-efficiency issues and develop traffic routing protocols that take advantage of MAC-layer scheduling, STC and BF, in an effort to dramatically reduce the signaling overhead.

Abstract:  The development of communication theory has focused on the problem of point-to-point communications, which is concerned with the transport of information from a single source to a single destination. In a networking context, each communication link is often effectively provisioned in isolation, without any attempt to exploit information from the physical layer to improve network performance. These issues limit network capacity and reliability and will be addressed in this research. Tactical mobile ad-hoc networks (MANET) present additional research challenges due to the presence of intentional jammers, hostile intercept and signal exploitation, and the fact that existing centralized control architectures are not easily adapted to meet the communication requirements of mobile, rapidly deployable ground forces.

Title:Single and Multi- Microphone Based Source Separation and Speech Enhancement (PI: B. Rao)

Sponsor: UC MICRO/Qualcomm

Period: 08/01/07-07/31/09

Specific Aim: To develop robust speech enhancement techniques

Abstract: This proposal develops single and multiple microphone based speech enhancement and auditory stream segregation based methods. Harmonic modeling has played an important role in speech coding but their application to the above mentioned problems has been limited and will be a topic of research.  In particular, for the single microphone based methods, we will examine subset selection (sparsity promoting) methods for spectral analysis as a front-end for the audio segregation task. We will develop methods that combine harmonic modeling with auditory cues such as common offsets, harmonic concordance etc. The resulting methods are expected to have superior ability to deal with transient interferences that are usually hard to suppress using conventional methods.  Multiple microphone based enhancement provide an opportunity to combine speech structure with novel spatial processing. Robust adaptive beamforming methods as well as methods based on a synergistic combination of ideas from MIMO blind equalization and ICA (Independent Component Analysis) will be developed.

Title: Multiuser MIMO Systems (PI: B. Rao, Co-PI: Profs. T. Javidi and J. Zeidler)

Sponsor: UC Discovery and several CWC industrial sponsors

Period: 2/1/08-1/31/10

Specific Aims: Providing high data rates to multiple users in band-limited radio channels that are inherently limited by multipath and fading is a challenging task. Multiple-Input, Multiple Output (MIMO) wireless communication systems have many attractive and often unique features such as array gain, interference suppression/avoidance capability, diversity gain, and spatial multiplexing that we plan to leverage to meet the aforementioned challenge. To realize the potential of multiuser MIMO wireless systems, this research will address the following research problems a) Scheduling and power control in multiuser MIMO systems with practical considerations such as receiver complexity, limited channel information, among others to bridge the gap between multiuser MIMO information theoretical research and its application, b) Cross-layer design of the physical and medium access control layers for enhanced throughput and energy efficiency in MIMO ad- hoc networks, c) Consider the interaction between layers and develop the optimal PHY layer operation and cooperation with respect to end-end delay performance for delay sensitive and bursty traffic.