On the effect of channel estimation error with superimposed training upon information rates

Adopting affine precoding as the transmission strategy, we investigate the effect of channel estimation error on the mutual information between the input and the output of a frequency selective fading MIMO channel. For Bayesian receivers which perform joint channel and symbol estimation there is asymptotically no loss in information rate. In contrast, for the receivers which obtain the channel estimation and use it to estimate the symbols, the lower bound on information rate is maximized by enforcing a form of orthogonality between symbols and training