A performance study of the RLS algorithm as a channel estimator in a nonstationary environment

The performance of an exponentially windowed RLS algorithm carrying out system identification of a nonstationary environment is considered. The two criteria used for assessing the performance of the algorithm are the steady state mean-squared error (MSE) and the initial rate of convergence. The effect of both filter order and level of nonstationarity is illustrated, with the LMS adaptive algorithm used as a reference for comparative purposes. The analysis presented by McLaughlin, Mulgrew and Cowan (1988) is used as the basis for the performance study and simulations are presented to allow comparison of the theoretical predictions of initial rate of convergence and steady state MSE with the result of computer simulations. These simulations also demonstrate that for a variety of practical situations the LMS rather surprisingly outperforms the RLS, in terms of steady state MSE