Convex combination of quantized kernel least mean square algorithm

Abstract

In this paper, we propose an new kernel adaptive filter, namely convex combination of quantized kernel least mean square algorithm (CC-QKLMS). By applying the convex combination idea to QKLMS, the CC-QKLMS takes the kernel sizes as the combined variables, which can achieve a fast convergence rate and a low steady-state mean-square error (MSE). In addition, since the quantization method is incorporated in CC-QKLMS, a linear growing network structure is naturally avoided. Simulation results on channel equalization validate the better performance of the CC-QKLMS in terms of the convergence rate and steady-state MSE.