Sparse Superposition Coding with Bayesian Detection for Correlated Unsourced Random Access

This paper considers a heterogeneous communication model combining (i) conventional unsourced random access, where individual messages are recovered at the receiver, and (ii) semantics-aware random access, where messages of the transmit-ting users are inherently correlated due to a commonly observed physical phenomenon. We leverage sparse superposition coding with Bayesian message passing decoder performing approximate inference on a graphical model that encodes the underlying correlations, and accounts for the structure of the applied sparse superposition code. We consider noncoherent operation without channel state information at the transmitter and at the receiver. We perform numerical simulations to characterize the trade-offs between the two operational modes.