Structured Sparse Bayesian Learning Based Multiuser Detectors for Uplink Grant-Free NOMA With Variable User Activities

In this work, we study the grant-free non-orthogonal multiple access (GF-NOMA) system to support the massive machine-type communications (mMTC), where the number of users is large, but normally, only a small portion of them activate within a random access opportunity. For the uplink of this system, we develop a structured sparse Bayesian learning based multiuser detection (SSBL-MUD) scheme. In a practical mMTC scenario, the user activity status could change in different time slots. To solve the multiuser detection (MUD) problem, we formulate the uplink GF-NOMA in multiple time slots with variable user activity factors based on multiple measurement vector (MMV) model. In order to leverage the sparse structural feature of the transmitted signal, we propose multiple sparse Bayesian learning multiuser detection (MSBL-MUD) and block sparse Bayesian learning multiuser detection (BSBL-MUD) schemes. Both of them are able to recover the transmitted information by joint user activity detection and data detection, and do not require any prior knowledge of the user sparsity. Simulation results demonstrate that the proposed algorithms significantly outperform the existing classical detectors.