Joint Active User and Data Detection in Uplink Grant-Free NOMA by Message-Passing Algorithm

Grant-free non-orthogonal multiple access (NOMA) is highly expected to support massive connectivity and reduce the transmission latency for future wireless communications. In this paper, we present a joint active user and data detection with no priori knowledge of the active users relying on expectation propagation (EP) and Gaussian approximation (GA) algorithm. To detect the user activity, a structured spike and slab prior is introduced to present the sparsity of transmission signal. Further, the parameters unknown are learned via expectation maximization (EM), which improves the performance of active user detection. Specifically, the active user detection problem in NOMA is firstly formulated under EM framework by parameter learning, and then the transmission data can be detected accurately by message-passing algorithms (MPA). Simulation experiments demonstrate the superiority of our proposed EP-GA-EM algorithm both in the performance of reconstruction and the bit error rate (BER).