LAUSR

As a candidate technology of B5G cellular Internet of Things(IoT), grant-free non-orthogonal multiple access (NOMA) has arisen worldwide concerns. However, the non orthogonality leads to the mutual interference between users which reduces the reliability of NOMA systems. Recent research in the design of NOMA transmission signal and multiuser detection are carried out separately which is difficult to achieve the global optimum. To solve this problem, we study the joint optimization of transmission and detection for NOMA system from an information bottleneck (IB) perspective. Specifically, we resort to deep learning to parameterize the transmitter and receiver of NOMA systems. Furthermore, the loss function for training is derived and analyzed based on information bottleneck and variational inference. Simulation results show that, the proposed method has better bit error rate (BER) performence with the same overloading factor compared to existing new techniques.