Based On The Confidence For The Spread Of Mimo Detection Algorithm And Theory Analysis

MIMO detection problem is one of the most important problems in MI-MO technology, and is the integrant module of the receiver. Most detection algorithms proposed so far can not trade-off or self-adaption smoothly between complexity and performance. This thesis discusses about the belief propaga-tion (BP) based MIMO detection algorithm. The factor graph model of the MI-MO channels and VBLAST architecture is researched, and several BP based detection algorithms with low complexity and high performance is proposed, which can trade-off between complexity and performance and have asymptotic optimality. Moreover, this thesis uses some tools in statistic physics and in-formation geometry to conduct the theoretical analysis about the BP detection algorithm, and optimization framework is adopted for both ML and BP. Fur-thermore, LDPC coded MIMO systems are also discussed, and BP based joint decoding and detection algorithm is proposed.Firstly, this thesis established the factor graph model and the pairwise markov random field model of the MIMO channels and VBLAST architecture, the character of their complete graph is analyzed. Then the standard BP (SBP) based detection algorithms on the graph models are analyzed respectively, the design of message and message passing rules are presented, and simulation re-sults are given. The results show that SBP on pairwise markov random field model achieves low performance just as linear detector, while SBP on factor graph based on log-likelihood-ratio can achieve high performance nearly the same as ML detector. However, the complexity of SBP on factor graph is even higher than ML.Secondly, this thesis proposes an edge selection approach for BP based detection algorithms on factor graph. Single edge based and partly edge based schemes are proposed, and two rules (fixed number and dynamic number of selected edges) for partly edge based scheme are also proposed. The simula-tion results show that the proposed BP detector can achieve high performance with low complexity, trade-off between complexity and performance and have asymptotic optimality. For further improve the proposed BP detector, the lin-ear detector cascaded BP detector is proposed to initialize the message of BP detector. Moreover, the complexity and convergence analysis are also given, in which the average mutual information is used as an index for convergence, and simulation result shows that the proposed BP detectors are almost information lossless.Thereafter, the theoretical analysis of BP based detection algorithm is giv-en in this thesis. Some tools in statistic physics and information geometry such as Bethe free energy, and optimization framework for both BP and ML are used for discussing the relationship between BP detector and ML detector here. The stationary points of BP are with the minimum Bethe free energy. Based on this conclusion, the constrained optimization function of ML are related with that of BP through Bethe free energy. And the Lagrange function of BP is equal to the opposite of the pseudo-dual Lagrange function of MLFinally, the LDPC coded MIMO systems are considered and BP based joint decoding and detection algorithm is proposed. The factor graph model of LDPC and VBALST architecture is analyzed. What’s more, soft informa-tion design and message passing rules are also discussed. Simulation result show that the proposed algorithm have higher performance than conventional schemes.