New enhanced robust kernel least mean square adaptive filtering algorithm

Abstract

This paper studies an enhanced robust kernel least mean square (KLMS) adaptive filtering algorithm for nonlinear acoustic echo cancellation (NLAEC) in impulsive noise environment. Robust KLMS algorithm based on M-estimate theory shows robustness to simulated, Contaminated Gaussian (CG) impulsive noise. However, it fails to combat real-world impulsive noise which normally consists of a few consecutive impulsive samples. In this work, the linear prediction (LP) scheme is applied to the KLMS algorithm to detect and cancel the impulsive noise. The resultant LP-based KLMS (LPKLMS) algorithm thus can achieve improved robustness to the real-world impulsive noise which is frequently encountered in NLAEC and other applications alike.