Optimization of Space-Time Block Codes for Noncoherent Reception in Fast Fading Channels

Noncoherent reception of a space-time block code (STBC) transmitted over a fast fading channel is considered. Reception method is based on a generalized likelihood ratio test. STBC codebooks optimized for this scenario were generated using the autoencoder neural network. For a baseline, codebooks obtained by adding pilot symbols to the Alamouti STBC were used. Error rate performance of the codebooks is analyzed in both block fading and fast fading conditions. It is shown that the autoencoder-generated codebooks outperform the Alamouti-based codebooks and, in the case of sufficiently long signals, they can provide not only transmit diversity but also additional time diversity.