Improved Bayesian Learning Detectors for Uplink Grant-Free MIMO-NOMA

Grant-free non-orthogonal multiple access (GF-NOMA) and multiple-input multiple-output (MIMO) techniques are key enablers for massive machine-type communications (mMTC) in 5G cellular Internet of Things. On the other hand, compressed sensing (CS) is widely accepted for multiuser detection, due to the sporadic traffic of mMTC. In this letter, we propose two Bayesian-based CS algorithms for uplink grant-free MIMO-NOMA. Exploiting the iterative expectation maximization (EM) and maximum ratio combining techniques, the spatially enhanced sparse Bayesian learning (SE-SBL) algorithm is developed to alternately update hyperparameter values and combined observation signals. Furthermore, we embed the generalized approximate message passing technique into the SE-SBL and propose a low-complexity computational approach. In particular, the aforementioned Bayesian algorithms do not require any prior knowledge of user activity level and noise power. Simulation results show that the proposed Bayesian methods exhibit a superior performance gain over the state-of-the-art.