A Two-Stage Bayesian Receiver for Massive Grant-Free OFDM-NOMA with Interfering Pilot Symbols

Massive grant-free non-orthogonal multiple access (NOMA) enables low-latency data transmission for a subset of coexisting user terminals, which may not be known a priori, and operates in an uncoordinated manner. On the receiver side, the effectiveness of user data decoding techniques heavily relies on accurate initial estimates of user activity, channel conditions, and parameters, all of which are derived from pilot symbols. However, in practical systems where pilot symbols are shared among a large number of users, the process is complicated by pilot interference. This paper introduces a pilot-only clustered multiple sparse Bayesian learning approach to address this acquisition challenge. It employs a multiple measurement vector (MMV) model to handle the common support shared by the multi-antenna channel impulse responses of each user. The resulting estimates are then used to initialize a low-complexity Bayesian procedure, based on expectation propagation, to refine both user activity and channel estimations, as well as to facilitate multi-user detection and decoding by leveraging both pilot and data observations. Numerical simulations demonstrate the good performance of the proposed two-stage Bayesian receiver for massive grant-free NOMA under frequency-selective channels with antenna correlation.