Accelerating Variable Step-Size LMS Adaptation Algorithms Using A Dynamic Adaptation Gain

It has been shown experimentally that incorporating dynamic adaptation gain/step-size (DAG) in variable step-size least mean squares (VS-LMS) parametric adaptation algorithms provides a significant acceleration of the adaptation/learning transients. The DAG is a filter acting on the correcting term before updating the estimated parameters. In this paper, the analysis of the VS-LMS algorithms incorporating a dynamic adaptation gain is done both in a deterministic and stochastic environment. Design of the DAG filter is discussed. Simulation results (adaptive line enhancer, filter identification) provide, on one hand, further evidence of the advantages of incorporating a DAG in VS-LMS algorithms and, on the other hand, validate the theoretical results.