The Robust Kernel Conjugate Gradient Least Mean p-Power Algorithm

Abstract

The traditional kernel adaptive filter (KAF) based on the minimum mean square error (MMSE) criterion suffers from poor filtering accuracy and slow convergence rate in Gaussian noises. To this end, the recursive KAF based on the least squares (LS) criterion can improve filtering accuracy and convergence rate in Gaussian noises with the increase of computational and storage burdens. However, both quadratic similarity measures based filters may suffer instability in non-Gaussian noises. To address these issues, a robust kernel conjugate gradient least mean p-power (KCGLMP) algorithm based on the mean p- power error (MPE) criterion is therefore proposed by combining the conjugate gradient optimization method with kernel trick. The proposed KCGLMP algorithm improves filtering accuracy and computational efficiency, simultaneously. Simulation results confirm the superiorities of the proposed KCGLMP algorithm over other conventional KAFs in non-Gaussian noises.