Parameter identification of quasi-stationary Rayleigh-faded time-varying digital channels

Novel solutions to the parametric identification of a frequency-selective time-varying digital Rayleigh channel from the received noisy data sequence are presented for both data-aided and non-data-aided (blind) cases. The random channel behaviour is modelled as a second-order autoregressive (AR) process. The classic higher-order Yule–Walker (YW) identification approach is based on the estimation of small-lag samples of the autocorrelation function (acf ) of the observed sequence. Due to the highly correlated nature of the channel process it gives unsatisfactory results. Two novel identification procedures are then proposed which are nonlinear in nature and exploit acf samples at larger lags thus avoiding the ill-conditioning problem of the YW approach. A third solution is also presented which is based on the averaging of the acf estimates at similar lags. This latter gives the best results in terms of estimation accuracy and robustness to noise and to channel-dynamic effects, especially for the blind case. It exhibits a reasonable computational complexity and constitutes a good candidate for practical applications.