Research

This section gives an overview of the research activities in the DSPRL. The Current Projects list below contains the research projects that are ongoing. Descriptions of recent and older projects are listed in the Previous Projects section below.

Our broad range of interests is reflected in the Areas of Interest page. Lastly, recent research sponsors and collaborators are acknowledged on the Sponsors page.

Current Projects

Nonlinear / Non-Wiener Effects in Adaptive Algorithms

| synopsis | related publications |

Adaptation of Coupled Nonlinear Oscillator Arrays

| synopsis | related publications |

Instantaneous Frequency Estimation & Interference Rejection

| synopsis | related publications |

State-Space Structures for Fast Adaptation

| synopsis | related publications |

Tracking Analysis of Affine Projection Algorithms

| synopsis | related publications |

Previous Projects

Speech Coding Based on Direct Line Spectral Frequency Adaptation

| synopsis | related publications |

EEG Modeling for Brain-Computer Interface

| synopsis | related publications

Acoustic Source Localization in the Presence of Reverberation

| synopsis | related publications | additional online materials |

Reconstruction of Telephone Loop Topology

| synopsis | related publications |

Speaker Identification Based on Adaptive Forced Response Inverse Filtering

| synopsis | related publications |

• Bias removal from estimation of the parameters for summed, damped, complex exponentials in colored noise. What are the limits of performance and, can we develop a procedure that approaches the performance limits? Can we detect when the assumption of colored noise, rather than white noise, is appropriate?
• Investigation of parametric spectral estimators for multi-resolution spectral estimation. Can we trade off time-resolution and frequency-resolution? How do performance and computational requirements compare with those for the wavelet transform?
• Build a digital filter simulator for various filter structures, with varying desired wordlength for fixed and floating point arithmetic. Compare performance with linear stochastic system predictions, in terms of filter output signal-to-noise ratio. Implement on the Motorola 56000 DSP, and corroborate the simulator prediction.
• Development, and performance evaluation, of signal processing algorithms for extracting a mixed stochastic signal process from its wideband noise corrupted measurements. The signal process is typically masked by the wideband noise. After signal processing enhancement the signal component can be presented acoustically to a human interpreter.
• Investigate parametric modeling of radar returns, for their efficient coding and identification.
• Image reconstruction from partial information, which is noisy. The partial information may be from transforms (such as the MBR representation), or from a priori knowledge (such as knowledge of a few zero crossings). Which kind of partial information is most effective? How can we best code this information?
• Analyze procedures that can serve to generate a coherent carrier in a digital receiver. This involves subsampling, extremely narrowband digital filtering (multi-rate) to suppress modulation components, Hilbert transform filtering for removal of the negative frequency interference, and FFT-based or parametric spectral estimation.
• Develop and analyze procedures for fusing information from radar and infra-red sensors, for target detection and target tracking.
• Fast algorithms for adaptive beamforming for small acoustic arrays.
• Receiver-based estimation and detection for digital transmission over time-varying noisy channels.
• Highly parallel Recursive/Iterative Eigenspace decomposition for sensor array signal processing.
• Detection of narrowband processes in colored noise.
• Use of algorithm structure, to design digital filters with reduced sensitivity to coefficient quantization and finite wordlength effects, and to design fast converging system identification algorithms.
• Approximate stochastic partial realizations for insuring nonnegative definite covariance sequence models, and for reduced order modeling of stochastic systems. Such spectral information is needed in matched, Wiener, Kalman, and statistical digital filter designs.
• Effects of spectral estimators on performance robustness of spectral information based designs of matched, Wiener, and Kalman filters in terms of signal-to-noise ratio, mean-squared error, and divergence performance respectively.
• Time-varying parametric modeling for speaker independent word recognition.
• Iterative reconstruction of space-limited scenes (images) from noisy projections. The idea is to provide for compatibility of the a priori knowledge and the noisy measurements.
• Effects of segment averaging on the parametric Burg spectral estimator.
• Spectral analysis of periodically time-varying systems, such as switched capacitor filters. Here we investigate the effects of internal parameters (switch resistance, noise color, amplifier characteristics), overall structure, and clockjitter, on the output noise spectral distribution.
• Covariance sequence modeling for spectral estimation. The modal decomposition model contains all relevant information in a generalized and more accessible form than spectral densities.
• Nonlinear time series modeling for prediction of power demand.