Mitigation of error propagation in decision directed OFDM channel tracking using generalized M estimators

We treat decision-directed channel tracking (DDCT) in mobile orthogonal frequency-division multiplexing (OFDM) systems as an outlier contaminated Gaussian regression problem, where the source of outliers are the incorrect symbol decisions. Existing decision-directed estimators such as the expectation-maximization (EM)-based estimators and the 2-D-MMSE estimator do not appropriately downweight incorrect/poor decisions while defining the channel estimator, and hence suffer from error propagation at high fade rates. We propose to use generalized M estimators for decision-directed channel estimation and apply extreme value theory in conjunction with the Huber's cost function for outlier detection and downweighting. Significant reduction in error propagation is obtained by identifying and downweighting those symbol decisions that act as outliers before using them for channel estimation. The proposed estimators clearly outperform the 2-D-MMSE and the EM-based estimators in terms of simulated error rate performance at high fade rates. These estimators also have performance comparable to (or better than) pilot-assisted Kalman-filter-based channel tracking schemes with only modest increase in complexity. Thus, a 6% to 12% increase in data throughput is obtained without loss in performance when compared with the pilot-assisted schemes, even when DDCT is applied over frame durations experiencing two to three fade cycles.