Generalized Normalized Gradient Descent Algorithm with Direct Update of the Step-Size Parameter

Abstract

A modified generalized normalized gradient descent (GNGD) adaptive algorithm is proposed. The original GNGD algorithm requires an indirect procedure for the computation of the step-size parameter, leading to an increased computational complexity. The proposed algorithm uses a direct update formula for this parameter, which further reduces the computational demands. Simulations in the context of echo cancellation prove the equivalency of the algorithms in terms of performances and improvements over normalized least mean square (NLMS) algorithm.