Generalized message passing detection of SCMA systems based on dynamic factor graph for better and flexible performance-complexity tradeoff

Abstract

Multiuser detection based on the message passing algorithm (MPA) has been considered for sparse code multiple access (SCMA) systems. Recently, some complexity-reduced MPA detectors have been proposed, among which the MPA detector based on dynamic factor graph (DFG-MPA) has been shown to outperform other MPA detectors with comparable complexities. However, all these MPA detectors are somehow not very flexible in terms of performance-complexity tradeoff, i.e., the granularities of computational complexity reduction are relatively large. In this paper, a generalized scheme of DFG-MPA, termed as GDFG-MPA, is proposed to make a better and more flexible performance-complexity tradeoff. The proposed scheme features two aspects: (1) instead of banning a message update forever, a banned message update at some iteration is allowed to be updated at later iterations; (2) different numbers of message updates are banned from updating at different iterations. Optimization of GDFG-MPA can be made by allocating banned message updates among iterations. Numerical results have demonstrated that compared to DFG-MPA the proposed GDFG-MPA can achieve much better performance at the same computational complexity or achieve the same performance with much lower complexity. Moreover, the proposed GDFG-MPA is more flexible in tuning the performance and complexity tradeoff.