Adaptive kernel bandwidth method for kernel normalized LMS adaptive algorithm

Authors:

Abstract
In this paper, an adaptive adjustment method for the kernel parameter used in the kernel adaptive filters (KAFs) is proposed. The KAF is one of the linear-in-the-parameters (LIP) nonlinear filters, and is based on the kernel method used in machine learning. Typically, the Gaussian kernel function is used, but there is no effective method for automatically adjusting its parameter that influences the convergence characteristics of the KAFs. An adaptive adjustment method for this parameter is proposed in the paper. The proposed method uses the difference of ℓ 1 norms of the input signals for the unknown system and the adaptive filter as the criteria. The kernel parameter will be updated according to the differences. The qualitative results of the proposed method is shown by the computer simulations.

This publication has 21 references indexed in Scilit: