Signals, Information, Inference, and Learning (SIIL) Group

[I’ve recently moved to Duke University.  New website forthcoming.]

The Signals, Information, Inference, and Learning (SIIL) Group focuses on the theory, application, and practice of statistical signal and array processing. Our goal is to advance technology for existing and emerging next generation systems through innovative algorithm development, solid theory and analysis, performance bounding, and experimental validation of concepts. Multisensor / multichannel applications and topics of interest include but are not limited to:


  • Cooperative Radar-Communications (Shared-Spectrum / RF Convergence)
  • Cognitive MIMO Radar / Sonar
  • Machine and Deep Learning
  • Bayesian Sparse Vector Recovery
  • Information Theory–Estimation Theory Interplay
  • Geolocation (positioning, navigation, and timing (PNT))
  • Robust Adaptive Filtering / Beamforming
  • Adaptive Radar Detection and Estimation
  • Synthetic Aperture Radar (SAR) Image Change Detection
  • Robust Parameter Estimation and Bounds Under Model Misspecification
  • Communication Over Dynamic Channels
  • Adaptive MIMO Communications 
Christ D. Richmond

Christ D. Richmond

Associate Professor Director of SIIL Group

Shared-Spectrum / RF Convergence

The limited availability of frequency spectrum requires greater spectral efficiency to meet the increasing demands for communication (comm.) and data services. Thus, exploring the possibility of diverse RF systems coexisting within the same frequency band is a means of improving spectral efficiency. Radars coexisting in the same frequency band as comm. systems presents a set of new challenges for system design and analysis.  We are exploring both the fundamental theory enabling such coexistence, and the development of algorithms to realize new capabilities to support emerging technologies. Military, civilian, and industrial systems potentially will be affected by this move toward shared-spectrum. Coexistence also will be essential to facilitate the required comm. needs of self-driving automotive vehicles.

Recent Papers:

Cognitive Radar / Sonar

Bats and dolphins are known to use echo location to find food and navigate their environments, and likely inspired the original concept of radar and sonar. Closer examination reveals that bats and dolphins actually adapt their chirp waveform emissions as their task evolves. The concept of cognitive radar / sonar builds on this observation by instituting multiple levels of system agility that allow adapting waveform characteristics, integration times, transmit powers, platform geometries, etc. as the mission evolves. Such agility has demonstrated improved performance over conventional radar/sonar. When coupled with layers of artificial intelligence, cognitive systems are expected to pave the way toward the next generation of radar / sonar technology.  We are working to develop meaningful performance bounds for such systems, and exploring the incorporation of deep learning methods to better leverage the benefits of AI.

Recent Papers:

Machine and Deep Learning

The increase in large amounts of stored labeled data (sometimes called “big data”) by many technology companies (e.g. Google, Facebook, Amazon, etc.) has set the stage for techniques such as machine and deep learning to extract incredible amounts of useful information and provide new and exciting capabilities. Such techniques are amazingly adept at finding and exploiting hidden structure within data. We are very interested in applying these learning techniques to classical problems in radar/sonar and communications. 

Bayesian Sparse Vector Recovery

Many engineering applications involve signals residing in a low dimensional subspace, or sometimes signals are intentionally transformed into a lower dimensional subspace (e.g. via compressed sensing). Sparse vector recovery aims to determine this low dimensional subspace from noisy data measurements, and knowledge of a much larger (underdetermined) space known to include the lower dimensional signal subspace. When cast within a Bayesian estimation framework, a selection of the prior distribution can be made to reflect desired signal sparsity. Variational Bayesian methods attempt to improve computations by using a misspecified prior distribution, but one that yields better computational efficiency.  We are interested in quantifying the performance loss suffered versus the gains in computational savings. Emerging techniques based on message passing appear to achieve better computational efficiency, while not using a misspecified model. So, we are also exploring the potential of applying such techniques to other areas.

Recent Papers:

Information Theory-Estimation Theory Interplay

New intimate relationships between Information Theory and Parameter Estimation have surfaced over the last twenty years.  Perhaps the most prominent is the I-MMSE that relates mutual information and the mean squared error (MSE) of the minimum MSE estimator.  Such relationships provide new insights and perspectives on signal processing, and may hold the key to achieving true spectral efficiency in shared-spectrum / RF convergence systems.

Robust Adaptive Filtering / Beamforming

The optimal adaptive beamformer (ABF) maximizing output signal-to-interference plus noise ratio has filter weights that depend on the data covariance and signal array response vector. The effectiveness of practical application of this optimal beamformer, however, is limited by (i) data stationarity (needed for covariance estimation), and (ii) knowledge of the true signal array response vector. Robust ABF (RABF) methods embrace these two critical issues by reformulation of the ABF weight optimization to account for them and minimize their effects.  We are interested in various aspects of robust ABF including insights from random matrix theory, application of machine / deep learning methods, and the implications of information theory based universal schemes.​

Recent papers:

Adaptive Radar Detection and Estimation

Adaptive radar detection and estimation make use of multiple data measurements collectively to accomplish the tasks of target detection and parameter estimation in the presence of additive clutter, jamming, and noise of unknown covariance. Several algorithms have been developed and studied in detail, and continue to find wide use in many adaptive radar systems, including the adaptive matched filter (AMF), the generalized likelihood ratio test (GLRT), and the adaptive coherence estimation (ACE).  We are interested in applying these techniques in new areas and using them in the context of transfer learning.

Some Papers:

Synthetic Aperture Radar (SAR) Image Change Detection

Synthetic aperture radar (SAR) exploits the relative motion of the radar platform to synthesize an effective array aperture during a coherent processing interval that in combination with judicious waveforms and signal processing can create photograph-like images of stationary targets and other scatterers in the scenes of interest. SAR is attractive for several reasons, but one advantage is its ability to create these images regardless of weather and lighting conditions. SAR change detection seeks to determine if a scene has changed since last inspection by comparing SAR images taken from the same area at different times. Such can be valuable for agricultural surveys, forest observations, and natural disaster damage assessments.  We are interested in developing and applying SAR change detection algorithms to new areas, including synthetic aperture sonar.

Recent Papers:

Robust Parameter Estimation and Bounds Under Model Misspecification

Much of signal processing and statistical inference is data model-based. These models, when correct, allow determination of fundamental limits on detection, classification/association, parameter estimation, data compression, and information rate transfer. Data models, however, are rarely specified correctly, as modeling errors often persist at some level. Robust estimation methods provide algorithms and procedures that are less sensitive to modeling errors. Bounds on parameter estimation that account for the possibility that the assumed model could differ from the true data model have also been developed, and are referred to as misspecified parameter bounds. Such bounds yield insight into the complex trade space that robust estimation schemes attempt to navigate.  We are interested in developing robust estimation techniques for adaptive communications, and passive sonar.

Recent Papers:

Communication Over Dynamic Channels

Airborne communication channels can face some major challenges. The general nature of aeronautical channels, the high dynamic mobility of some airborne platforms, the complex scattering encountered over terrains traversed by these airborne platforms, and varied interference challenges (intentional and unintentional) are all central factors contributing to the complexity of some airborne communication networks. Opportunities exist to advance the state-of-the-art for these challenging environments, and we’re pursuing various approaches.

Recent Papers: