Channel Capacity Bounds in the Presence of Quantized Channel State Information

The goal of this paper is to investigate the effect of channel side information on increasing the achievable rates of continuous power-limited non-Gaussian channels. We focus on the case where (1) there is imperfect channel quality information available to the transmitter and the receiver and (2) while the channel gain is continuously varying, there are few cross-region changes, and the noise characteristics remain in each detection region for a long time. The results are presented for two scenarios, namely, reliable and unreliable region detection. Considering short- and long-term power constraints, the capacity bounds are found for log-normal and two different Nakagami-based channel distributions, and for both Max-Lloyd and equal probability quantization approaches. Then, the optimal gain partitioning approach, maximizing the achievable rates, is determined. Finally, general equations for the channel capacity bounds and optimal channel partitioning in the case of unreliable region detection are presented. Interestingly, the results show that, for high SNR's, it is possible to determine a power-independent optimal gain partitioning approach maximizing the capacity lower bound which, in both scenarios, is identical for both short- and long-term power constraints.