Quantized kernel least mean mixed-norm algorithm

Abstract

Quantized kernel least mean square (QKLMS) algorithm is an effective up-to-date adaptive nonlinear learning algorithm which also has good performance for kernel structure growing control. It achieves good results under Gaussian noise environment. In this paper, a new algorithm, quantized kernel least mean mixed norm (QKLMMN), is proposed for adaptive nonlinear learning with non-Gaussian additive noise statistical distribution models (including combination). As an alternative of conventional squared error criteria, mixed-norm criteria is utilized for our algorithm. A comprehensive convergence analysis is carried out. Experiments for nonlinear time series prediction and nonlinear system identification are conducted. Experimental results verified the effectiveness and superiority of our proposed algorithm compared with other kernel based adaptive nonlinear learning algorithms under non-Gaussian noise environment.