Sparse Bayesian Learning based on Fast Marginal Likelihood Maximization for Joint User Activity Detection and Channel Estimation in Grant-Free NOMA

Abstract

Grant-free non-orthogonal multiple access (GF-NOMA) is a promising solution to solve the massive connectivity problem with low latency and signaling overhead. User activity detection (UAD) and channel estimation (CE) are two enabling technologies in GF-NOMA systems. In this paper, a correlation-enhanced sparse Bayesian learning algorithm based on fast marginal likelihood maximization (CSBL-FM) is proposed, which can improve the performance of the UAD and CE without prior knowledge of channel state information and sparsity. Firstly, a multi-frame sparse model is proposed so as to exploit the correlation and sparsity characteristics of single time slot and among multiple frames. Then, in order to accurately realize signal reconstruction, channel estimation process is described as sparse signal recovery process based on user indicators and training sequences. Next, based on the proposed multi-frame sparse model, the loss function is derived and optimized to detect active users by utilizing fast marginal likelihood maximization. Simulation results show that the proposed CSBL-FM algorithm is practical to be applied in GF-NOMA system by achieving the balance between high reconstruction performance and fast convergence speed.