580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can higher-order statistics and cumulants improve blind channel identification in multipath and non-Gaussian environments?
This research area focuses on leveraging higher-order statistics (HOS) and cumulants to enhance blind channel identification, specifically targeting frequency-selective, multipath, and non-minimum phase channels under non-Gaussian noise conditions. The approach addresses the limitations of conventional second-order statistics methods which assume Gaussianity and often fail in practical fading environments with multipath and impulse noise.
Key finding: This paper proposes a blind channel estimation algorithm based on third-order cumulants to estimate phase and magnitude of frequency-selective fading channels without training sequences, demonstrating robustness against... Read more
Key finding: This study evaluates eight distinct channel identification algorithms, demonstrating that higher-order cumulant based methods achieve accurate parameter estimation of broadband radio channels under various SNR and input data... Read more
Key finding: Introduces a robust, adaptive estimation algorithm for communication channels contaminated with impulsive noise, based on robust statistics and QQ-plot analysis. The method yields unbiased parameter estimates in contaminated... Read more
2. What role do sparse signal representations and structured perturbation handling play in high-resolution channel identification for multipath environments?
This theme investigates the application of compressed sensing and sparse recovery techniques to channel identification in multipath fading environments. It primarily addresses the 'off-grid' problem, where channel multipath parameters do not align exactly with predefined discrete dictionaries, causing performance degradation in traditional sparse methods. The research introduces optimization algorithms that adaptively refine channel parameter grids to improve detection and estimation accuracy in sparse, multipath communications.
Key finding: Proposes a novel algorithm combining particle swarm optimization (PSO) with orthogonal matching pursuit (OMP) to adaptively perturb dictionary grid points in sparse multipath channel estimation, effectively mitigating the... Read more
Key finding: Introduces the PSO-CAF technique, a transform domain method leveraging cross-ambiguity function and PSO to identify multipath channel parameters by clustering delay-Doppler components and performing local optimizations. The... Read more
Key finding: Presents an online adaptive algorithm based on k-sparse component analysis (k-SCA) assumptions to identify mixing matrices in underdetermined blind source separation problems encountered in channel identification. The method... Read more
3. How can prior knowledge and kernel-based methods enhance channel identification performance under nonlinear and binary measurement constraints?
This research direction explores the integration of kernel adaptive filtering techniques and the exploitation of prior system knowledge (e.g., transmission filter characteristics) in channel estimation frameworks. Emphasis is placed on enabling the efficient identification of channels with nonlinearities, sparse impulse responses, or quantized (binary) outputs. Kernel methods map input-output relationships into high-dimensional spaces, facilitating nonlinear estimation while maintaining computational tractability. These methods are combined with side information and novel recursive algorithms to improve accuracy and convergence.
Key finding: This paper extends blind channel identification methods by incorporating prior information about transmitter and/or receiver pulse-shaping filters, allowing the estimation process to focus on the actual propagation channel... Read more
Key finding: Develops a recursive kernel-based adaptive filtering algorithm for identifying finite impulse response in nonlinear systems with binary-valued outputs. The algorithm uses kernel functions for implicit high-dimensional feature... Read more
Key finding: This comprehensive study compares different blind identification algorithms including kernel methods that exploit reproducing kernel Hilbert spaces (RKHS) structure for nonlinear system modeling. Kernel algorithms are shown... Read more