QR-decomposed generalized belief propagation for MIMO detection
Akihide David Shigyo, Shogo Tanabe, Koji Ishibashi · International Symposium on Information Theory and its Applications · 2016
In this paper, we propose a QR-decomposed generalized belief propagation (QR-GBP) for multi-input multi-output (MIMO) detection to eliminate an error floor and achieve near-optimal bit error rate (BER) performance. MIMO detection via belief propagation (BP) exhibits the even worse performance than maximum-likelihood detection (MLD) since factor graphs defined by typical MIMO channels are fully-connected. Generalized belief propagation (GBP) enables to calculate exact marginal probabilities on a valid region graph consisting of sets of multiple nodes in a factor graph. However, factor graphs of MIMO channels cannot be directly transformed into region graphs. In this paper, we apply QR-decomposition to the channel matrix to construct the valid region graph. We further propose truncated region graph to reduce the computational complexity of GBP-based MIMO detection. Numerical results show that GBP-based MIMO detection with valid region graph can achieve near-optimal performance close to MLD, and low-complexity detection with truncated region graph can eliminate the error floor even at high signal-to-noise ratio (SNR) region.