Effects of Impulse Noise at Filter Input on Performance of Adaptive Filters Using the LMS and Signed Regressor LMS Algorithms

Abstract

This paper presents an adaptive filtering system where impulse noise is present at filter input. To study adverse, or favorable, effects of such impulse noise on adaptive filter performance, we develop transient and steady-state analysis of adaptive filters using the LMS algorithm (LMSA) and signed regressor LMS algorithm (SRA). Through analysis and experiment, we find that the SRA exhibits significantly higher robustness against the impulse noise than the LMSA. For the SRA, we find that mean square tap weight misalignment (MSTWM) decreases as the impulse noise variance increases, attaining a minimum before the filter diverges.