Improved normalized least mean square algorithm using past weight vectors and regularization parameter

Abstract

In this paper we derive an improved minimization criterion for normalized least mean squares (NLMS) algorithm using past weight vectors and the regularization parameter. The proposed minimization criterion minimizes the mean square deviation (MSD) of currently updated weight vector and past weight vector. The result of the proposed NLMS algorithm approaches the conventional NLMS algorithm as the regularization parameter reduces to zero. The result also shows that as the regularization parameter decreases the convergence rate increases. © 2011 Springer-Verlag.