Bit-error rate optimized training for time-varying fading channels

We study time division multiplexed training (TDMT) for the temporally correlated Rayleigh flat-fading channel. For a class of causal and non-causal estimators, we derive the bit-error rate (BER) of TDMT assuming grey-coded M-PSK input. The high-SNR performance of the estimators is characterized by simplified expressions for the minimum attainable BER. It is shown that energy allocation techniques that are often assumed to boost performance of TDMT may be useful in this setting, but only for the simplest estimators and at moderate values of SNR.