NEWTON/LMS ALGORITHM USING MODIFIED DCT

Abstract

A Newton/LMS algorithm using a modified DCT is proposed which uses an efficient technique of inverting the input autocorrelation matrix when the periodic pseudorandom sequence is used as the reference signal. The number of operations is greatly reduced and the computational results show fast convergence rate and low misadjustment error. The application of the algorithm to the case of nonperiodic reference signal is also described.