Codebook Design for Noncoherent Communications Using Autoencoders

The problem of joint optimization of coding and modulation for noncoherent communications is considered. To minimize block error rate, an autoencoder network is used. The network architecture is designed to mimic the end-to-end operation of a communication system with noncoherent correlation receiver. Signal codebooks with various spectral efficiencies are obtained for several small block lengths. Achieved pairwise correlations between signals of codebooks are compared with some known results. Also, minimum distances and power efficiencies of the obtained codebooks are analyzed. The obtained results show that autoencoder networks can be successfully used to design signal codebooks tailored to specific operating conditions.