Consideration on the performance of kernel adaptive filters for the mixture of linear and non-linear environments

In this paper, we consider the characteristics of the kernel adaptive filters for the mixture of linear and non-linear environments. We first consider employing a linear kernel as one of the kernels in multi-kernel adaptive filters. It is pointed out that the convergence characteristics of the filter corresponding to the linear kernel is affected by the selection of the other kernels. Then, we propose a new structure which uses a linear and a multi-kernel adaptive filter simultaneously. In the proposed method, the a posteriori error of the linear filter is calculated before updating the multi-kernel adaptive filter. Then, the a posteriori error is used to calculate the update term for the multi-kernel adaptive filter. The effectiveness of the proposed method is confirmed by computer simulations.