Toward High-Dimensional Data Analysis In Massive MIMO Systems: A Random Matrix Theory Inspired Perspective

Massive multiple input multiple output(MIMO)is a vital technique for future communication systems.It can boost the development of many research areas,i.e.,wireless communication,smart grids,and smart health care.Specifically,MIMO systems help to transmit/collect massive streaming data for the above fields.With the explosive growth of the data,there are new possibilities for deep understanding and analysis of problems in these fields.On the other hand,in the face of the explosion of the data(dimensions and quantities),the classical modelbased data modeling and analysis methods encounter difficulties in terms of effectiveness and robustness.In this thesis,the random matrix theory,the key method of high-dimensional data modeling and analysis,is used to solve the corresponding problems.The specific work and results achieved include:· This thesis firstly introduces and analyzes the development of massive streaming data analysis.Based on the random matrix theory,we further present the background,the motivation,and the basic principle of the related data modeling and analysis.· Based on the theory of random matrix,the Eigen-Inference based channel state information(CSI)related unknown parameter estimation and the rotation invariant estimation based channel reconstruction are respectively proposed.With the above methods,a linear quantization precoding scheme based on Bussgang's theorem is proposed to solve the precoding problem of large-scale MIMO systems with low-resolution DACs and imperfect CSI.Besides,the simulations verify the robustness of the proposed scheme.· Based on the nonconvex optimization technique,a nonlinear precoding technique with low computational complexity is proposed,improving the performance of the linear quantized precoding method proposed in the former Chapter.It is proved that the proposed algorithm has convergence under weak assumptions.The efficiency and reliability of the proposed method are verified by simulation data.· Based on the semi-supervised deep learning,a new data equalization scheme for largescale frequency-selective MIMO systems based on low-resolution DACs is proposed under the condition of a small number of pilots.Taking Bregman divergence as the evaluation criterion,it is analyzed and proved that unlabeled data can improve the generalization performance of the proposed method.The effectiveness of the proposed method is then verified by numerous case studies.· Based on the theory of random matrix and the massive streaming data collected by the phasor measurement unit,a new abnormality detection method based on multidimensional covariance matrix test is proposed.The computational efficiency of the proposed method is further improved by the principal component calculation and redundant calculation elimination.Besides,the efficiency of the above method is verified by the artificial and real data.· A method of automatic extraction of temporal and spatial features of massive streaming EEG data based on deep learning has been proposed.Combined with the above methods,a new EEG data classification method is further constructed with the majority voting algorithm.The accuracy and reliability of the proposed method have been verified by the real EEG data provided by the Shanghai Mental Health Center.