Kernel Least Mean Square With Maximum Correntropy Criterion

Abstract

We introduce a novel kernel least mean square (KLMS) algorithm for nonlinear input-output models, where the output is generated with respect to multiple inputs in a coupled fashion. The KLMS algorithm is proposed under maximum correntropy criterion for robustness. The mean square convergence has been carried out and the energy conservation relation is also established, which reflect the effects of the coupling parameter. A data-independent upper bound on the stepsize is derived to guarantee the convergence of the KLMS algorithm. Simulation results are provided to demonstrate the excellent performance.