Single edge based belief propagation algorithms for MIMO detection

Abstract

In this paper, two low-complexity belief propagation (BP) based detectors are proposed for multiple-input multipleout (MIMO) system. The factor graph is leveraged to represent the MIMO channels, and based on which our algorithms are developed. Unlike the existing complicated standard BP detectors that consider all the edges when updating the messages, our algorithms only focus on single edge, which largely reduce computational complexity. In particular, we propose a novel Gaussian approximation with feedback information (GF) mechanism to enable the proposed single edge BP detector. In order to further improve the detection performance, we also propose to integrate the linear MIMO detector into the initial GF based single edge BP detector, where the pseudo priori (PP) information obtained from linear detector is judiciously exploited. Convergence and complexity analyses, along with the numerical simulations, verify that the proposed single edge BP detectors outperform the existing BP detectors in performance while with low complexity.

A theoretical analysis of belief propagation based detection in MIMO systems

“…More manifest relationship between BP and JML and other analyses about why BP based detection can achieve high performance remain as future works. However, the proposed approach provided a method to analyze and redesign some improved BP based detection algorithms such as the one in [9], which will be presented in another paper.…”

“…Since the BP detection algorithm on the factor graph in Fig.1 is different with the BP based decoding in some extent, thus we give its theoretical analysis in the following contex. What's more, the authors in [9] proposed some improved BP based detection algorithms recently, the theoretical analysis proposed in this paper can be further developed to analyze the algorithms in [9] and to design other advanced BP based detection algorithms.…”

“…Nevertheless, theoretical analysis about why BP can achieve such good performance is not given in [8]. Since BP based detection is different from BP based decoding in some extent, especially some improved BP based detection algorithms [9], we therefore give the analysis of the relationship between BP and JML via an optimization framework of Bethe approximation to free energy and JML detection in this paper. The performance of BP is demonstrated to be associated with a parameter c. When c → 0, the Bethe Lagrangian pseudo-dual function F # Bethe turns out to be the negative of Lagrangian function L JML , and thus the stationary points of BP is the critical points of joint maximum likelihood detection (JML) optimization problem.…”

Graph‐based low complexity detection algorithms in multiple‐input–multiple‐out systems: an edge selection approach

“…We develop our method proposed in [18], and propose a novel edge selection policy. Before the details are addressed, we give some discussion about the FG model.…”

“…In particular, only means are used as feedback considering that on Gaussian graphical models, the means converge to the correct values whereas the variances are generally incorrect [20]. Moreover, since only binary modulation simulation results are given in [10,14,18] we consider high-order modulation here. Mapping between bit soft output and modulation symbols when computing the feedback information is a new issue arised in our approach, and we propose a novel mapping rule to cope with this.…”