New approaches for the Kernel-based Adaptive Filter with Epanechnikov kernel

Abstract

Kernel Adaptive Filtering has proven to be an effective solution for nonlinear channel equalization, outperforming traditional linear filters. In this context, we highlight the Kernel Maximum Correntropy (KMC) filter, in which the use of the Epanechnikov kernel has shown to be a promising approach. However, this method presents two drawbacks: the numerical instability that caused divergence and the need of training constantly. In this paper, to address the first problem, a sliding window procedure was proposed. To address the second problem, a decision direct mode was implemented. Both versions showed a desirable behavior, with no performance loss. A study in noisy scenarios was also considered.