Research On Markov Chain Monte Carlo Method For MIMO Systems

Multiple-input multiple-output(MIMO) is one of the key technologies to improve spectral efficientcy and quality-of-service, which has been used in lots of wireless standards, such as 3GPP LTE/LTE-Advanced、IEEE 802.16e/802.16 m WiMax, HSPA+ and so on. Meanwhile, in order to match the requirements of the explored data traffic consumption, massive access and various new services, large-scale MIMO with tens to hundreds of antennas is proposed to be an important candidate for the next generation wireless communication systems. Since the performance and complexity of the communication system is mainly dependent on the signal processing at receiver, one of the main challenges for MIMO systems, especially for large-scaled MIMO system, is the tradeoff between performance and complexity for the detection algorithms. Among the plenty of detection algorithoms, Markov Chain Monte Carlo(MCMC) algorithom, which has a linear or at most a polynomial complexity, can be used as a promising lowcomplexity solution to MIMO systems. Thus, we focus on the high performance and low complexity MCMC-MIMO detection algorithms in this thesis.Firstly, the background of MCMC statistical algorithm is brelifly explained. Then we discuss four typical sampling methods which have been widely applied to solve the integration of high-dimensional in Bayesian problem. The effects on the convergence of Markov Chain, including initial value, jumping distribution and its standard deviation, and the parallel/depth tradeoff of MCMC samplers are also well analyzed. In this thesis, bit-wised and symbol-wised MCMC algorithms are applied to MIMO detection. The theoretical analysis, formula deduction and practical calculation procedures of MCMCMIMO are also presented. Besides, we also analyze the issues of MCMC algorithm, the stalling problem in high SNR regime and higher complexity than traditional linear detector.According to the analysis of the mechanism for the disadvantages of traditional MCMC, we proposed sereval technologies to improve the performance and reduce the complexity. And most of them can be jointly used to optimize the MCMC detector. Results show that MCMC algorithms based on Max-Log updating with enhanced techniques, say biased processing, reinitialized processing and LLR clipping, can achieve 1-2dB performance gains with 60% complexity of MMSE-PIC. Meanwhile, the symbol-wised MCMC detectors based on Max-Log updating or 2-best updating can achieve the performance of quasi-ML with complexity moderately increased. Besides, in massive MIMO, bit-wised MCMC algorithm with enhanced techniques outperform MMSE detector at the complexity of biased MMSE based on three stage of Nuemann series expansion.Finally, we propose a novel MCMC-MIMO detection algrithm based on stochastic computation with pipeline and dual partition calculation scheme, which can achieve higher throughput and lower-complexity than traditional methods. Matrix-matrix and matrix-vector multiplication in preprocessing in MCMC-MIMO canbe greatly simplified with the weighted sequence generator and dual partition stochastic multiplication. The pipeline technique enables preprocessing module and Gibbs sampling module perform simultaneously. Simulation results show that compared with 12-bit fixed point MCMC detector, the complexity of MCMC-MIMO detector based on stochastic computation can be further reduced by 58% with 0.5dB performance gain.