Improved Message Passing Algorithms for Sparse Code Multiple Access

• Published in: IEEE transactions on vehicular technology 2017-11, Vol.66 (11), p.9986-9999
• Main Authors: Dai, Jincheng, Niu, Kai, Dong, Chao, Lin, Jiaru
• Format: Article
• Language: English
• Subjects: 5G mobile communication; Algorithms; Complexity; Computation; Computational complexity; Computer simulation; Convergence; Convergence speed; design framework; Detectors; Lookup tables; Mathematical analysis; Message passing; MPA; Multiuser detection; Nodes; NOMA; Nonorthogonal multiple access; Overloading; Scheduling; scheduling strategies; SCMA; Table lookup
• ISSN: 0018-9545; 1939-9359
• DOI: 10.1109/TVT.2017.2741525

Sparse code multiple access (SCMA) is one of the nonorthogonal multiple access techniques for the 5G system. SCMA can provide different levels of overloading to meet the diverse traffic connectivity requirements. However, its relatively high computational complexity of multiuser detection is still a significant concern for practical implementation, even when the sparse structure has already been employed. In this paper, a design framework for an improved SCMA multiuser detector is proposed based on the message passing algorithm (MPA). As the primary SCMA detector, two aspects of MPA are simplified and optimized. First, we introduce a lookup table (LUT) scheme to reduce the computational complexity of the max* operation in the MPA. In contrast to the extensive Jacobian approximation, the proposed LUT method can guarantee the stable convergence of the MPA for SCMA. Second, a series of novel scheduling schemes are proposed to speed up the convergence. A single scheduling MPA (SS-MPA) method is given to enhance the convergence performance of MPA, where the soft messages in the function nodes and variable nodes are serially calculated and synchronously updated. To further improve the throughput of the MPA, a multiple scheduling MPA (MS-MPA) is proposed. In this method, multiple detectors are used to calculate in parallel and update the node messages by different orders. Since the scheduling strategies of message update are optimized, both the SS-MPA and MS-MPA can converge more quickly than the conventional MPA. Theoretical analyses and simulation results regarding the error performance and convergence properties of the above schemes are included.