Robust Channel Tracking in Fast Fading MIMO channels

Channel estimation for Alamouti space frequency coded (ASFC) 2 times 2 and 2 times 1 MIMO OFDM systems, employing decision directed channel tracking (DDCT) for high fade rates and long frame durations is studied here. Errors in the regression matrix (i.e. equalisation errors) have a Gaussian distribution, but due to error propagation problem of DDCT at high fade rates and long symbol durations, large deviations from the Gaussian model is observed. Hence the problem of channel estimation can be formulated as an outlier contaminated Gaussian linear regression problem. Most of the classical estimators like Least squares (LS) and Expectation Maximisation (EM) algorithm perform poorly in presence of outliers. Moreover, for a MIMO system the source of outliers are the erroneous decisions of any of the multiple input channels. Here, we propose to use the EM algorithm to partition multiple input channel into independent estimations for each transmit-receive antenna pair and provide an iterative framework for solving the multiple regression problem. EM algorithm is made robust using the Huber´s M principle (for the E and the M steps) and it is shown analytically that mean square error (MSE) performance improves as compared to existing schemes which assume Gaussian noise. The simulated symbol error rate (SER) and MSE performances demonstrate the significant gains of the proposed robust EM algorithm over existing DDCT techniques at high normalised fade rates for the given system.