时频双选信道中基于变分推理的迭代接收技术

Given the ever increasing demand for multimedia service and the high mobility of terminals, the effect of doubly selective fading caused by multipath and Doppler spread on the transmitted signals becomes more prominent. To this end, the optimal receiver has to perform joint channel estimation and data detection. In this study, we propose to use variational inference, which has been successfully applied in machine learning, to the receiver design in time and frequency doubly selective channels (DSCs). By minimizing the Kullback-Leibler (KL) divergence between the target function and the postulated one, variational inference approximates the complicated multi-variant function with a relatively simple one which is much easier to factorize. For DSCs parameterized by basis expansion model (BEM), joint channel estimation and data detection algorithms based on variational inference in the presence of carrier frequency offset (CFO) and phase noise (PHN) are going to be investigated. The effects of using different conditional independent constraints will be compared in terms of performance and computational complexity. On the factor graph representation of DSCs, the application conditions and the rules of using hybrid massage passing based on variational inference will be studied. The joint iterative channel estimation, data detection and decoding algorithms based on variational massage passing (VMP) are investigated and the performances of which are going to be analyzed using both extrinsic information transfer (EXIT) chart and the "semi-analytical" methods. Thanks to the systematic and highly structured features of the variational inference, the studies in this proposal can be extended to the problems of interference mitigation, multiuser detection, etc, and contribute to both the theoretic analysis and practical implementation of iterative receivers.

随着人们对"移动性"和"多媒体业务"需求的不断增长，由多径和多普勒效应引起的时间和频率双选择性衰落对无线传输的影响越来越显著，因此其最佳接收机必须进行联合信道估计和检测。本课题将已在机器学习领域得到成功应用的"变分推理"引入时频双选信道的接收机设计，通过最小化与目标函数之间的KL距离，将包含多变量的复杂函数用相对简单且易于因式分解的函数近似。对采用基扩展模型参数化的时频双选信道，研究受到载波频偏和相位噪声影响下基于变分推理的联合信道估计和检测算法，分析不同独立性约束条件对算法性能和复杂度的影响。在时频双选信道的因子图模型上，研究基于变分推理的混合消息传递的规则和条件，设计基于变分消息传递的迭代估计、检测和译码算法，并通过EXIT图和"半分析"方法分析其性能。由于变分推理具有系统性和结构化的特点，本课题的研究可扩展至干扰抑制、多用户检测等问题，对迭代接收机的设计具有重要的理论意义和应用价值。

未来移动通信要求更高的传输数据速率，发送信号的带宽往往大于传输信道的相干带宽，引起的频率选择性衰落造成符号间干扰。另外，高速移动的通信设备带来的多普勒效应，使得通信信道随时间快速变化，接收机需要联合同步、信道估计与符号检测。考虑到双选信道参数未知且远大于观测数目，本课题采用复指数基（CE-BEM）展开模型对信道进行建模，以减少待估计的信道参数的个数。课题将已在统计物理等领域得到广泛应用的变分推理方法引入接收机设计，通过最小化与目标函数之间的KL 距离，将包含多变量的复杂函数用相对简单且易于因式分解的函数近似。首先，利用变分推理的方法和因子图上的消息传递规则，提出了一种联合双选信道估计与符号检测算法。借助于快速傅里叶的高效性，提出的算法复杂度低，仅与观测数据的长度成对数增长。然后，研究了在变分推理框架下，平均场、Bethe等不同的独立性约束条件下的多种混合消息传递算法，分析了算法的实现复杂度、误码率性能以及信道的估计性能。另外，对于受较强相位噪声影响的通信系统，提出了一种联合信道系数估计，相位噪声估计和迭代译码算法，消除了未知动态信道和相位噪声对通信系统的影响。提出了联合载波相位恢复和译码算法，并借助半分析模型分析了不同信道编码系统下的载波相位恢复性能。此外，根据相关领域的研究发展情况，本课题还将提出的算法扩展应用到了下一代通信系统接收机设计中，解决了超奈奎斯特（FTN）波形调制模式下，数据符号通过双选信道时的符号检测问题，提出的算法实现复杂度与信道记忆深度无关，有利于大压缩因子和信道延时扩展较长情况下的实际应用。本课题的研究成果，对于复杂信道下基于变分推理方法的接收算法设计具有较好的理论研究意义，对于实际中低复杂度接收机设计具有一定的参考价值。