On the use of mismatched Wiener filtering for the characterization of non-stationary channels

We develop a method for the determination of local regions in time in which a channel can be approximated as stationary. Contrary to previous results in literature relying on to some extent arbitrary measures and thresholds, we consider a realistic (flat fading) channel estimator and relate the size of local quasi-stationarity regions to the degradation of the mean squared channel estimation error due to mismatched statistics. As the evaluation of the mean squared error turns out to be difficult, we give an approximate expression. Using channel measurements, we exemplarily evaluate the local quasi-stationarity regions based on the actual and the approximate mean squared error, and we find that the results show strong similarities.