Direct Optimization of Codebooks for Code-Domain Nonorthogonal Multiple Access with Noncoherent Reception

Nonorthogonal multiple access (NOMA) is a technology that allows accommodating more users in a limited time-frequency resource and thus increasing overall spectral efficiency of a multiuser communication system. A common approach to NOMA requires the knowledge of the channel state information (CSI) at the receiver. To make CSI estimation possible, pilot signals should be transmitted by all users. This phase of the communication process consumes valuable resources of the system and should be minimized. Noncoherent reception does not require CSI knowledge allowing to get rid of dedicated pilot signals. In the paper, noncoherent NOMA is considered for uplink non-selective block fading channel. Direct optimization of codebooks is performed using an autoencoder neural network. The receiver is based on a generalized likelihood ratio test. Several codebooks are obtained with signal length from 4 to 8 samples, for 3… 7 users and 1…3 bits of data per user. Error rate performance of the obtained codebooks is analyzed, achieved combinations of power and spectral efficiency are presented. It is shown that the obtained codebooks can outperform codebooks with orthogonal separation of users.