A Bias-Compensated Robust LMS Algorithm with Adaptability to Impulsive Noise and Noisy Input

Abstract

This thesis proposes a robust least mean square algorithm (rLMS) to eliminate bias due to noisy input data and to reduce the performance degradation due to impulsive noise. It is well known that the ordinary least mean square algorithm (LMS) affords biased estimates when applied to the parameter estimation problem with noisy input data. To eliminate this bias, a new estimation method for the input noise variance is presented and explained. A saturation-type error nonlinearity with impulsive noise in LMS adaptation is presented. In simulations, rLMS provided a smaller mean square deviation than did other algorithms. Because of the improved adaptability to noisy input data in the system and impulsive noise interference, the algorithm has improved precision in estimating the Finite impulse response of an unknown system. Therefore, rLMS can be applied extensively to adaptive signal processing.