基于Max-Log更新的马尔科夫链蒙特卡洛MIMO检测增强算法

针对传统的马尔科夫链蒙特卡洛（MCMC）算法，提出了一种基于Max-Log更新的MCMC-MIMO检测算法。该算法采用了基于Max-Log更新的采样，可以有效产生收敛于后验概率（APP）分布的比特样本列表集合，同时可避免计算传统MCMC算法中的每比特概率分布。但是该检测算法在高信噪比下，采样过程会陷入锁死到局部最优态。在此基础上，提出了3个增强技术:1）抖动处理，对给定置信区间内的更新进行抖动处理；2）条件下重新初始化，对处在潜在锁死态的采样序列进行重新初始化；3）修剪饱和处理，利用球形译码算法中的修剪饱和技术来处理MIMO检测输出的对数似然信息（LLR）。仿真结果显示，基于Max-Log更新的MCMC增强算法能有效地解决陷入锁死的问题，从而提高系统性能并降低系统的计算复杂度。在复杂度为MMSE-PIC检测算法的90%的基础上，性能提高了2 dB。

An Enhanced MCMC Algorithm for MIMO Systems Based on Max-Log Updating

In this paper, an enhanced Markov chain Monte Carlo (MCMC) algorithm based on max-log updating is proposed for multiple input multiple output (MIMO) system. The max-log updating can generate the list vectors to simply the complexity of the calculation of the extrinsic log-likelihood ratios (LLRs) efficiently. Meanwhile, it avoids calculating probability distribution per bit in conventional MCMC. However, the proposed MCMC detection suffers from the so called "stalling" problem, where the Markov chain may be trapped into local optimal state. Thus, we also propose three enhancement technologies:1) biased processing, i.e., updating randomly in a given biased interval; 2) reinitialized processing, i.e., reinitialize the Markov chain under the sub-optimal states; 3) clipped processing, i.e., reprocessing the LLR with clipping. Simulation results show that the proposed algorithm can remedy the "stalling" problem efficiently with reduced complexity, and can achieve 2 dB performance gains with 10% less complexity than MMSE-PIC.