Tanner-Graph-Based Massive Multiple Access—Transmission and Decoding Schemes

In this paper we consider two Tanner-graph-based transmission schemes for massive multiple access, which combine non-orthogonal multiple access (NOMA) and grant-based random access. Each user transmits two data streams repeatedly using several channel time-frequency resource blocks (RBs) and the transmission schedule is represented by a Tanner graph, where variable nodes and check nodes represent the transmitted signals and the RBs, respectively. In Scheme 1 each variable node represents a data stream of a user, whereas Scheme 2 employs rate splitting and each variable node represents the superimposed data streams of a user. On the receiver side, we first consider peeling decoders that serve both as pilot-based channel estimators and baseline decoders. We then develop message-passing decoders for both transmission schemes that can fully exploit the diversity afforded by the repetitive transmission across different RBs, as opposed to the peeling decoders. We also propose a neural decoder by deep unfolding the message-passing decoder and further performing a small number of training epochs using the pilots. Simulation results show that for Transmission Scheme 1, the message-passing decoder offers decoding performance improvement over the peeling decoder by orders of magnitude; and as a result, Scheme 1 substantially outperforms Scheme 2. Moreover, the neural decoder further improves upon the performance of the message-passing decoder, as more information is learned about the transmitted data over the training epochs.