M-estimate based Proportionate Normalized Adaptive Subband Filter with Combined Step-size

Abstract

The M-estimate based proportionate normalized subband adaptive filter (M-PNSAF) exists a performance compromise caused by the constant step-size. To effectively deal with this problem, we propose to take advantage of large and small step-sizes through the adaptation of the mixing factor to improve the M-PNSAF performance. Different from adaptively adjusting the mixing factor through a sigmoid function, the proposed algorithm minimizes the squared a posterior subband errors at each iteration to find the optimal time-varying mixing factor. Finally, simulation results in impulsive noises corroborate the effectiveness of the proposed algorithm.