Research Directions: Douglas L. Jones's Group



Current Research

Acoustic signal processing with small arrays
Biologically Inspired Sensory Systems
Ultra-Low-Power Computer Systems

Prior Research

Time-frequency analysis
Wavelets and denoising
Adaptive Signal Processing
Biomedical and hearing aid applications
Communications
Low-Power Computer Systems
Algorithms and implementation
Teaching Innovations

Acoustic signal processing with small arrays

Array processing has been a subject of intense research for several decades, but most successful methods depend on relatively large arrays of sensors, narrowband processing of stationary sources, fewer sources than sensors, and simple acoustic environments. The capabilities of human and other animal hearing systems demonstrate that these constraints can be overcome. Our prior work has developed new bio-inspired binaural algorithms for the localization and separation of desired sources from a cluttered acoustic environment, such as in a restaurant or cocktail party, that show remarkable performance improvements over conventional techniques while using only two to four microphones. Current research is developing meta-adaptive algorithms that automatically and blindly learn the characteristics of complex acoustic environments (such as reverberation or many nonstationary interferers) and adapt or auto-calibrate to them to maximize performance. Four-dimensional sound recording, transmission, and playback that preserves the directional characterics of the sound is being developed. These new methods are being applied to a number of applications such as advanced hearing aids, hands-free telephony, advanced multimedia systems, automotive and military applications, noise suppression for speech recognition systems, and highly accurate direction-finding of RF sources with small antennas.

Some real-world audio examples can be found here.

Biologically Inspired Sensory Systems

Animals have many types of sensory systems that provide vital information about their environment. In collaboration with biologists and MEMS engineers who develop new forms of sensors, we are developing novel bio-inspired sensing systems with funadamentally new capabilities. Recent work has developed an artificial lateral line that detects and locates moving objects in the water surrounding the array, an artificial weak electrosense that performs a similar function using a self-generated electric field, and bio-inspired algorithms for high-resolution acoustic source localization and separation using very small arrays. Current work is extending these methods to localization of neurons, new tactile and vibrissal (whisker) sensing systems, and an "electronic bat" 4-D imaging system.

Ultra-Low-Power Electronic Systems

Reducing energy dissipation and size of wireless sensing systems could enable many high-impact applications such as intelligent materials, smart objects that are aware of and adjust to their surrounding conditions, and enriched human sensory and monitoring interfaces as well as military systems. We believe that only dynamically self-optimizing systems that adaptively select the right information, data resolution and rate, signal representation, communication bandwidth, and data latency over a broad dynamic range can provide the needed functionality at an absolute minimal energy cost. As part of the Multiscale Systems Center, we are developing "attentional" signal processing systems that optimally adapt all system components to the current demands of the environment and the application to maximize the expected long-term system utility within a very limited overall energy budget. This includes highly scalable detection and estimation algorithms that trade off performance and energy over orders of magnitude to dynamically match current system needs, as well as system management algorithms that maximize the total system utility.

Time-frequency analysis and nonstationary signal processing

Adaptive time-frequency representations

Developed the concept of adaptive time-frequency representations (TFRs), and several of the leading signal-dependent and adaptive transforms. These include adaptive window short-time Fourier transforms, adaptive wavelet transforms, adaptive optimal kernels, adaptive cone kernels, and fast algorithms for their computation. Our recent "consistent time-frequency representation" achieves unprecedented resolution and cross-term suppression.

Statistical time-frequency analysis

Developed a fundamental and comprehensive theory of statistical time-frequency analysis. We derived optimal kernels for nonstationary spectrum estimation and time-frequency estimation of TFRs of noisy or random signals. A theory of time-frequency detection has determined the class of detection problems for which time-frequency-based detection is globally optimal, the optimal kernels for these classes, and efficient methods of implementing these detectors. These techniques have been applied in a number of applications, including machine-fault detection and diagnosis, microembolus detection, and ECG classification.

Time-frequency-space processing

Developed efficient, near-optimal time-frequency-space detectors and estimators for partially coherent arrays. Highly efficient, nearly optimal quadratic narrowband array detection algorithms have been an important by-product of this research.

Generalized joint signal representations

Contributed several advances in generalized joint signal representations, including a time-frequency-based derivation of the chirplet transform, new orthogonal chirped bases, and unitarily transformed joint signal representations. We showed the equivalence of Cohen's and Baraniuk's methods for constructing general joint signal representations and contributed new insights to this theory. We extended the theory of adaptive and statistically optimal TFRs to generalized joint signal representations. We have developed four-parameter joint quadratic time-frequency-delay-doppler representations for applications such as improved adaptive time-frequency analysis and detection and classification.

Nonstationary blind source separation and interference cancellation

Introduced new adaptive methods for blind source separation of nonstationary signals. These methods are simpler than existing methods and continuously track environmental changes. New methods for extraction of speech signals in cluttered environments (the "cocktail party" environment) have yielded great improvements over existing methods (see expanded discussion below).

Wavelet techniques and applications

Generalized wavelet decompositions and transforms

Developed new orthogonal chirped wavelet bases, unitarily transformed basis decompositions, and efficient algorithms and implementations. Developed the chirplet transform from a time-frequency context.

Denoising

Innovative new methods for denoising multichannel data provide much better performance than single-channel methods and are very efficient; applications to hyperspectral imagery have shown more than 10 dB SNR gain. Derived worst-case bounds for the performance of denoising methods for both orthonormal bases and overdetermined frames. New "Bayesian pursuit" methods offer improve denoising performance for using overdetermined frames and a hierarchical statistical signal model.

MEMS sensors

In a joint project with Profs. Chang Liu and Naresh Shanbhag, we are developing distributed, multi-element touch and flow sensors and embedded architectures and algorithms for extracting sophisticated touch and flow information, such as texture, turbulence, softness, and slippage to create an "artificial skin." In collaboration with Prof. Ron Miles at SUNY Binghamton, we are developing algorithms for high-accuracy direction-finding and signal recovery using colocated arrays of directional MEMS sensors.

Adaptive signal processing

Blind equalization

Developed a vector constant modulus algorithm for blind equalization of shaped channels, the first practical solution to this problem.

Nonlinear adaptive filters

Developed a low-complexity, LMS-like algorithm for general systems with a memoryless nonlinearity. Application to nonlinear echo cancellation demonstrated substantial improvement over conventional linear echo cancellers.

Nonstationary blind source separation

Introduced new adaptive methods for blind source separation of nonstationary signals, including both instantaneous and convolutive (dynamic) mixtures. These methods are simpler than existing methods and continuously track environmental changes. Research continues on faster algorithms.

Nonstationary adaptive beamforming

Developed a frequency-domain minimum-variance distortionless-response beamformer for small arrays with unprecedented performance in the recovery of speech in nonstationary interference.

Algorithms

Developed reduced-complexity and reduced-delay implementations for adaptive filters. Analyzed the transpose-form implementation for FIR adaptive filters and demonstrated its advantages for high-speed pipelined implementation.

Biomedical applications

Binaural hearing aids

A multidisciplinary group based in the Beckman Institute has developed new binaural algorithms for the extraction of a desired source from a cluttered acoustic environment, such as in a restaurant or cocktail party. The new methods show remarkable performance improvements over conventional techniques in such environments, and have been implemented in a real-time DSP-based system. Major research efforts toward commercialization for advanced hearing aids and other acoustic extraction applications such as hands-free telephony, automotive and military applications, and noise suppression for speech recognition systems continue.

Electrocardiogram analysis

Developed improved methods for denoising electrocardiograms using adaptive time-frequency processing.

Microembolus detection

Developed wavelet-based and chirped wavelet detectors from ultrasound reflections from microemboli. Theoretical and experimental studies demonstrated that these methods approach optimal performance.

Ultrasound image formation and analysis

Developed fast frequency-domain three-dimensional reconstruction algorithms for image reconstruction from measurements on a circular aperture. Applications include high-resolution imaging from ultrasound microprobes on the end of a needle and from small ultrasound catheters. Developed methods for detecting edges and tissue boundaries in ultrasound images.

In collaboration with Prof. W.D. O'Brien, developed new methods for aberration correction in ultrasound imaging that perform much better than existing approaches in severe aberration.

fMRI Image Denoising

Are developing (In collaboration with Prof. Farzad Kamalabadi and Dr. Keith Thulborn at UIC) efficient methods for blind removal of noise from functional MRI image sequences. These techniques will allow precise imaging at much faster rates by greatly reducing the necessary averaging time to construct low-noise functional images.

Telecommunications and other applications

Peak Power Reduction for OFDM systems

Developing new methods providing unprecedented peak-to-average power ratio (PAR) reduction for large-constellation OFDM systems. These methods are based on novel constellation-shaping approaches, and obtain these reductions with no loss of data rate or increase in symbol error rate. Waveform-modification methods that offer more modest reductions but are compatible with current standards have also been developed. Our Active Constellation Extension method has been included in the DVB-T2 next-generation digital terrestrial television broadcast standard. Extension of these ideas to peak power reduction in optical, CDMA, and MIMO communication systems continues.

Optimal discrete multi-tone (DMT) power allocation

Developed the first fast, exactly optimal algorithm for power allocation in discrete multi-tone modulation.

Joint source-channel matching

In collaboration with Profs. Shanbhag and Ramchandran, developed joint source-channel coding methods for wireless image and video transmission. These general methods allow near-optimal matching of most source and channel codes, as well as on-line adaptation to time-varying channels. Techniques that minimize the total system power have also been developed. Continuing research is developing optimal methods for multi-level broadcasting, joint source-network coding for video transmission over wireless networks and the internet, and total system optimization for distributed wireless sensor networks.

Wireless communication for binaural hearing aids

Developing new methods for low-power wireless communication for binaural hearing aids and other near-the-body applications.

Nonstationary interference cancellation

Developed new frame-based methods for blind removal of nonstationary interference from direct-sequence spread-spectrum communications signals.

Instantaneous frequency estimation/FM demodulation

Developed an adaptive TFR-based IF estimator for FM demodulation that lowers the SNR threshhold by 3-4 dB over existing methods.

Stochastic sensor networks

Developing a very simple, robust, low-power sensor network approach and protocol for large networks of sensors. Each sensor controls power by operating, independently of the others, on a low duty cycle. Recent results prove that such a network can operate reliably with very high probability and can perform all network functions without any coordination of wake/sleep cycle between the nodes.

Low-Power Computer Systems

Global Resource Allocation through Cooperation (GRACE)

In collaboration with Profs. S. Adve, R. Kravets, and K. Nahrstedt in the Computer Science Department, we are developing a new computing framework allowing joint, cooperative adaptation of the hardware, networking, operating system, and media application software to jointly minimize the total energy consumption in power-limited, mobile, general-purpose computers.

Signal processing algorithms and implementations

FFTs and Hartley transforms

Performed the first accurate analysis of the computational complexity of the Hartley transform, showing conclusively that it is virtually equivalent to the real-valued FFT. Co-authored two heavily-cited papers on the Hartley transform and real-valued FFTs.

Joint hardware/algorithm design

Developed several new algorithms/architectures for FFTs, FIR filters, and adaptive filters offering higher performance or reduced hardware complexity.

Teaching Innovations

Introductory Computer Engineering

With Prof. Steven Lumetta, developing a new first Computer Engineering course for all ECE majors. A new, "21st century" integrated hardware/software/systems approach to teaching computer engineering develops an across-the-layers understanding of computing, with an emphasis on digital logic and hardware but with an introduction to programming so that students can understand the context and primary application of that hardware.

Nonmajors course on Information Technology

With Prof. Michael Loui, developed a course, ECE 101, on information technology and engineering for non-engineering students. Digital information technology is introduced at several levels, including audio and video media, digital logic, and the Internet. The course is half lab-based, with students doing real engineering designs so that they also learn the process of technology development and the tradeoffs faced by engineers. The course will satisfy the General Education requirements for non-science-and-engineering students.

Digital Signal Processing Laboratory Textbooks

Developed the first textbook for a DSP-microprocessor-based laboratory course. This text spurred the development of hands-on DSP laboratory courses at many universities, and similar courses are now a mainstay of many electrical and computer engineering curricula in the U.S. and around the world.

Completed the first open-source, on-line DSP laboratory textbook as part of the Connexions project.

ECE 420: Digital Signal Processing Laboratory

Introduced a digital signal processing laboratory at the University of Illinois in 1989. Innovations in both the content and the teaching methods keep this laboratory at the forefront of hands-on DSP education. During the next few years, this laboratory and the students will participate in a NSF-supported research project for the development of next-generation DSP compiler technology. Over the years, equipment donations from Texas Instruments, Motorola, and Analog Devices have equipped the instructional laboratory with state-of-the-art real-time DSP platforms. This course has become one of the most popular laboratory courses in the ECE department (capped at 60 students per semester due to space limitations) and is now offered every semester. ECE 320 has been designated a Texas Instruments Elite Laboratory.

ECE 210: Analog Signal Processing

Assisted in the development of the course outline and the laboratories for this innovative sophomore course, which replaces the traditional sophomore circuits course and the junior signals and systems course. Taught the first full-scale offering of ECE 210 (110+ students). This course is required for all EE and CompE majors in the current undergraduate curricula at Illinois.

Email:dl-jones@uiuc.edu

Return to the Douglas Jones Homepage


Created by Douglas L. Jones. Last updated January 7, 2010