Stochastic analysis of the transform domain LMS algorithm for a non-stationary environment

This paper presents an improved statistical model for the transform-domain LMS algorithm operating in non-stationary environment from a time-varying plant. The stationary case is also considered as a particular case of the non-stationary one. The derived model takes into account a fixed-length sliding window for estimating the transformed input signal power. Small step-size conditions and Gaussian input data are assumed. Using model expressions, algorithm parameters as optimum step size, excess error, and misadjustment are obtained. Through numerical simulations, the accuracy of the proposed model is assessed.