Mean weight behaviour of delayed LMS algorithm with leakage

Abstract

The mean weight model of the leaky delayed least mean square (LDLMS) algorithm using new simplifying assumptions, instead of the classical independence theory is presented. In this way, by assuming algorithm convergence, it is shown that with the proposed model it is possible to describe accurately mean weight behaviour. In particular, the proposed model permits the study of the adaptive filter behaviour under more practical situations. The theoretical predictions are in close agreement with Monte Carlo simulations.