Modified LMS algorithm for fast identification of digital channels

Abstract

A new modified LMS algorithm, to adjust the coefficients of an adaptive digital filter, is presented. Starting from the basic LMS (least mean square) algorithm, the modified algorithm, LMSm, has been conceived in order to achieve, for a determined number of iterations, both properties in parallel: a faster convergence rate; a smaller final quadratic error than the LMS and nLMS algorithms. The underlying idea is its application to digital communication channel identification and later equalization, which requires a high frequency of operation. These properties have been analytically derived and its behaviour has been extensively verified. Both analytical and experimental results obtained by using the LMSm algorithm in standard digital channel models show the following important improvements: a higher velocity of convergence; a smaller mean squared error between the model identified by the system and the real channel parameters.