Grant-Free Non-Orthogonal Multiple Access (GF-NOMA) is considered as a promising technology to support the massive connectivity of Machine-Type Communications (MTC). The design of efficient and high-performance multi-user detection (MUD) scheme is a challenging issue of GF-NOMA, especially when the number of active users is unknown and relatively high. This paper adopts Sparse Bayesian Learning (SBL) approaches to solve the MUD problem of GF-NOMA in MTC. The MUD problem within a certain access slot is formulated as a Single Measurement Vector (SMV) model and efficiently solved via SBL-based methods. To further improve the MUD performance, we set up a Multiple Measurement Vector (MMV) model and develop block SBL-based MUD methods, by exploiting the temporal correlation of user activity over successive access slots. Then to extend the usage of the aforementioned algorithms to the scenarios with relatively high, or quasi-sparse, user activity, we propose novel SBL-based MUD algorithms via post sparse error recovery methodology, for both the SMV and MMV problem models. Simulation results show that the proposed SBL-based MUD algorithms achieve substantial performance gain over traditional ones, especially when the number of active users is unknown and relatively high.

中文翻译：

---

基于贝叶斯学习的无资助 NOMA 系统多用户检测


无授权非正交多址 (GF-NOMA) 被认为是支持机器类型通信 (MTC) 大规模连接的一项有前景的技术。高效、高性能的多用户检测（MUD）方案的设计是GF-NOMA的一个具有挑战性的问题，特别是当活跃用户数量未知且相对较高时。本文采用稀疏贝叶斯学习（SBL）方法来解决MTC中GF-NOMA的MUD问题。某个接入时隙内的 MUD 问题被表述为单测量向量 (SMV) 模型，并通过基于 SBL 的方法有效解决。为了进一步提高 MUD 性能，我们通过利用连续访问时隙上用户活动的时间相关性，建立了多重测量向量 (MMV) 模型并开发了基于块 SBL 的 MUD 方法。然后，为了将上述算法的使用扩展到用户活动相对较高或准稀疏的场景，我们针对 SMV 和 MMV 问题模型，通过后稀疏错误恢复方法提出了新颖的基于 SBL 的 MUD 算法。仿真结果表明，所提出的基于 SBL 的 MUD 算法比传统算法取得了显着的性能提升，特别是当活跃用户数量未知且相对较高时。