Comparison of Cascaded LMS-RLS, LMS and RLS Adaptive Filters in Non-Stationary Environments

This paper proposes a cascaded LMS-RLS prediction filter for improved performance in non-stationary environments. In this proposed filter, an LMS filter with varying step-size is used as the initial filter for achieving faster convergence rate and then a RLS filter for obtaining improved convergence. Theoretical analysis shows that the proposed technique improves the behavior of the adaptive filter in steady state. Computer simulations using MATLAB SIMULINK validate its performance as the error is found to be significantly less compared to the error rates of individual LMS and RLS filters.