A variable step size strategy for sparse system identification

This work proposes a novel two-strategy approach to improve the performance of a recently-proposed (RZA-LMS) algorithm for sparse environments. The first and key strategy uses a novel h -norm-based variable step-size scheme to improve the performance of both the conventional and the RZA-LMS algorithms. The second strategy applies to the normalization process to remove the constraint on the step size imposed by RZA-LMS algorithm by widening the range of choice of the step size. The simulation results clearly show the superiority of our proposed schemes in both sparse and non-sparse environments.