How much training is optimal in adaptive PSAM over markov rayleigh fading channels

To mitigate the effects of time-varying fading, Pilot Symbol Assisted Modulation (PSAM) has been introduced, through which the transmitter periodically inserts known symbols or pilots in the data frame. The channel estimates obtained from these pilots are then employed in the coherent demodulation of data symbols. A novel PSAM scheme was introduced in that adapts the transmitter´s coded modulation strategy to the quality of the channel estimate at the receiver without requiring any channel-feedback from the latter. In our work, we study the performance of this non-feedback adaptive PSAM scheme using clustering techniques. More precisely, instead of using one pilot per data frame, we send a cluster of pilots consecutively in the beginning of every data transmission interval. The performance of this scheme is measured in terms of achievable rates using binary signaling and modeling the fading process as a Gauss-Markov process. We show through numerical computations that our new strategy provides higher achievable rates at certain levels of SNR and fading correlation, and we provide numerical answers to how much training is ¿optimal¿ for the strategy. Finally, we show how the methodology may be extended to high-order Gauss-Markov fading models, with which pilot clustering is expected to yield even higher benefits.