New partial update robust kernel least mean square adaptive filtering algorithm

Abstract

This paper studies a partial update (PU) robust kernel least mean square (KLMS) adaptive filtering algorithm which is particularly suitable for nonlinear acoustic echo cancellation (NLAEC) application. By exploring the data mapping property from the linear space to the high-dimensional feature space using polynomial kernel, the sequential PU scheme for conventional linear adaptive filters can be applied to the KLMS algorithm. This results in reduced computational complexity with moderate convergence rate loss. Moreover, in order to enhance the robustness of the KLMS algorithm to impulsive interference, the robust M-estimate scheme is incorporated into the kernel trick used in KLMS to develop a robust kernel least mean M-estimate (KLMM) algorithm. Finally, computer simulations are conducted to verify the advantages of the proposed work.