大规模多输入多输出(Multiple-input Multiple-output，MIMO)技术可以有效 提高空间分辨率和自由度，减少用户间的干扰，成为下一代移动通信的核心技术 之一。精确信道状态信息(Channel State Information，CSI)的获取是挖掘大规模 MIMO 潜力的关键。毫米波通信和密集组网的引入，对信道估计提出了以下挑战: 非理想硬件的使用对导频信号造成严重的失真，毫米波信道容易发生阻塞从而使 通信中断，多小区环境下导频污染现象严重。针对上述问题，本文研究了非理想 硬件下的毫米波大规模 MIMO 信道估计与突变检测，以及多小区大规模 MIMO 系 统导频污染抑制。主要成果如下。第一，考虑毫米波大规模 MIMO 系统往往会采用低功耗、廉价的硬件，本文 研究了收发端存在硬件损伤下信道估计。利用毫米波信道的有限散射特性，将硬 件损伤下的信道估计问题建模成测量矩阵受扰动的稀疏恢复问题，提出鲁棒稀疏 贝叶斯学习(Robust Sparse Bayesian Learning，R-SBL)信道估计算法。进一步，为 了降低计算复杂度，提出鲁棒快速稀疏贝叶斯学习(Robust Fast Sparse Bayesian Learning，R-FSBL)信道估计算法。仿真结果表明，与现有方法相比，所提方法可 以获得更高的信道估计精度和更低的误码率。第二，针对毫米波系统容易因车辆和行人的阻塞而发生通信中断问题，本文研 究了盲毫米波大规模 MIMO 信道突变检测。利用信道矩阵的低秩特性，将信道突 变检测问题建模成低维子空间突变检测问题，提出基于累积和(Cumulative Sum， CUSUM)的信道突变检测算法。进一步，推导出信道突变检测算法的高效递归形 式。仿真结果表明，所提方法可以保证较低的检测延迟和虚警率。第三，针对多小区大规模 MIMO 系统的导频污染问题，本文研究通过非正交 导频设计提高信道估计精度。将非正交导频设计建模成最小化信道估计均方误差 和问题，提出使用线性化交替方向乘子法(Linearized Alternating Direction Method of Multipliers，LADMM)算法对该问题进行求解，同时考虑每个导频符号只能 从 4-QAM 星图集合中选取的特殊情况。推导出使用非正交导频以及最大比合并(Maximum RatioCombining，MRC)下，用户上行遍历速率的下界。基于此，将非 正交导频设计建模成最大频谱效率和问题。仿真结果表明，与现有方法相比，所 提方法可以获得更高的信道估计精度和上行可达和速率，并且计算复杂度更低。

As one of the key technologies of next-generation mobile communications, mas- sive multiple-input multiple-output (MIMO) can effectively improve the spatial resolu- tion and degree of freedom, and reduce interference between users. The acquisition of accurate channel state information (CSI) is the key to tapping the potential of massive MIMO. The introduction of millimeter wave (mmWave) communication and dense net- working poses the following challenges to channel estimation: the use of non-ideal hard- ware causes serious distortion to pilot signals, the mmWave channel is prone to blockage and thus the communication is interrupted, and the phenomenon of pilot contamination in a multi-cell environment is serious. In response to the above problems, this dissertation studies mmWave massive MIMO channel estimation and abrupt changing detection un- der non-ideal hardware, as well as multi-cell massive MIMO system pilot contamination suppression. The main results are as follows.First, considering that mmWave massive MIMO systems often use power-efficient and inexpensive hardware, this dissertation studies the mmWave massive MIMO channel estimation with hardware impairments at transceivers. By exploiting the limited scattering characteristics of mmWave channels, the problem of channel estimation with hardware impairments is modeled as a sparse recovery problem where the measurement matrix is disturbed, and a robust sparse Bayesian learning (R-SBL) channel estimation algorithm is proposed. Furthermore, in order to reduce computational complexity, a robust fast sparse Bayesian learning (R-FSBL) channel estimation algorithm is proposed. Simulation results show that compared with the existing methods, the proposed method can obtain higher channel estimation accuracy and lower bit error rate.Second, for the problem that the mmWave communication is prone to interrup- tion due to the blocking of vehicles and pedestrians, this dissertation studies the blind mmWave massive MIMO channel abrupt changing detection. By exploring the low-rank characteristics of channels, the channel abrupt changing detection problem is modeled as a low-dimensional subspace abrupt changing detection problem, and a cumulative sum (CUSUM)-based channel abrupt changing detection algorithm is proposed. Furthermore, the recursive form of the channel abrupt changing detection algorithm is deduced. Sim- ulation results show that the proposed method can guarantee lower detection delay and false alarm rate.Third, for the pilot contamination problem of multi-cell massive MIMO system, thisdissertation studies improving channel estimation accuracy through non-orthogonal pilot design. The non-orthogonal pilot design is modeled as the problem of minimizing the total mean square errors (MSEs) of the channel estimators of all BSs. The linearized alternating direction method of multipliers (LADMM) algorithm is proposed to solve the pilot design problem, and the case that the choice of each pilot symbol restricted to a 4- QAM constellation is considered. A lower bound of uplink ergodic capacity in the case of cellular networks using non-orthogonal pilots and maximum ratio combining (MRC) is derived. Based on this, a new non-orthogonal pilot design problem that maximizes the total spectral efficiency of all users in the cellular network is established. Simulation results show that, compared with the existing pilot design method, the proposed pilot design method can obtain higher channel estimation accuracy and uplink achievable rate, and lower computational complexity.