面向大规模MIMO系统的快速卷积混叠盲辨识技术研究

Massive Multi-Input Multi-Output (MIMO) is one of the key technologies for the 5th Generation (5G) Mobile Communication Systems since it can improve the data transmission rate and increase the system capacity substantially. Its most advantages are based on the fast and accurate acquisition of the channel state information (CSI). The conventional estimation methods based on pilot training sequence lead to severe pilot contamination, and the increase of the antenna number makes the computational complexity of channel identification grow rapidly, both of which will affect the accurate estimation of CSI seriously. This project aims at the fast and accurate acquisition of CSI. It uses the dynamic convolutive mixing model to fit the massive MIMO system, and seeks to provide a useful solution to the acquisition of CSI, meanwhile ensuring the channel capacity and spectrum utilization. Under this framework, the tensor decomposition theory is firstly exploited to analyze the identifiability of the convolutive MIMO systems. Secondly, the mutually correlated source signals are preprocessed in the transmitting side by some properly designed precoders, so that the correlation among the coded signals is removed and consequently their spatial diversity is increased. Finally, the precoding mechanism, combined with compression sensing, and Z transform technology, is utilized to develop fast blind identification algorithms for the convolutive massive MIMO systems, which can reduce the amount of data that the receiving antennas need to deal with and decrease the computational complexity of the CSI estimation. The research achievements of the project provide an important reference for the future wireless communication systems.

大规模多输入多输出(MIMO, Multi-Input Multi-Output)天线技术能极大提高通信传输速率、增加系统容量，是5G网络的关键技术之一，其众多优势都基于信道状态信息(CSI, Channel State Information)的快速准确获取。传统的导频训练序列获取方法会导致严重的导频污染，且天线数量的增加使得信道辨识的计算复杂度迅速上升，二者均会严重影响CSI的准确获取。本项目以CSI的快速准确获取为目标，采用动态卷积模型拟合大规模MIMO系统，在保证系统容量和频谱利用率的前提下估计CSI。首先采用张量分解理论分析系统的盲可辨识性；然后在发射端引入设计合理的预编码器，去除发送信号间的互相关性以提高其空间多样性；最后结合预编码机制、压缩感知、Z变换等，设计快速卷积信道盲辨识算法以减少接收端需处理的数据量、降低CSI估计的运算复杂度。课题的研究为5G技术发展提供重要的借鉴。

大规模多输入多输出(MIMO, Multi-Input Multi-Output)天线技术是5G网络的关键技术之一，其优势有赖于信道状态信息(CSI, Channel State Information)的快速准确获取。传统的导频训练方式存在计算数据量大、精确度低等问题；基于此，本项目采用动态卷积模型拟合大规模MIMO系统，借助盲信号分离框架进行信道辨识，继而进行源信号估计或重建。具体研究工作包括：从盲可辨识性理论角度出发，研究在动态卷积模型下以密度聚类为基础的源信号数目估计问题；讨论在高阶MIMO FIR场景下源信号的准确重构问题；针对MIMO FIR模型下的复值信道状态信息辨识问题，将预编码器与导频序列相结合，在预编码器参数估计的基础上通过张量分解进行信道辨识；研究在大样本场景下，利用知识蒸馏进行模型压缩、提高计算效率的算法。项目的研究成果对大规模MIMO技术的进一步发展有所助力。