毫米波大规模多输入多输出（multi-input and multi-output，MIMO）技术能大幅提升通信系统的传输速率、频谱效率以及能量效率，是支撑未来移动通信网络最关键的传输技术之一。然而，由于毫米波器件制造工艺水平的制约，以及系统的硬件成本和能量消耗等方面的限制，实际中通常采用低成本、低功耗、性能受限的器件部署毫米波大规模MIMO系统。硬件性能受限将导致系统性能大幅降低，亟待研究开发与之匹配的传输技术。本文针对硬件性能受限下毫米波大规模MIMO系统的关键传输技术和优化问题进行了研究，具体如下：针对硬件性能受限大规模MIMO系统，分析了低精度量化误差和非完美射频链路对信道估计性能的影响，推导出上行传输可达速率的近似闭式表达式。在此基础上，揭示出低精度模拟—数字转换器（analog-to-digital converter，ADC）和非完美射频链路之间的一种性能互补关系。解决了一比特量化毫米波大规模MIMO系统的上下行信道估计和下行预编码设计问题。对于上行传输过程，提出了一种基于压缩感知的信道估计算法，通过估计信道角度域的稀疏参数来重构信道矩阵。在下行传输过程中，利用信道角度域参数上下行的互易性，并且通过优化调整一比特数字—模拟转换器（digital-to-analog converter，DAC）输出信号的幅度来获取更多的自由度，提出了一种基于近端梯度法的信道估计方案，同时设计出基于近端梯度法的预编码技术。研究了混合精度量化毫米波大规模MIMO系统的ADC精度分配、信道估计以及导频设计联合优化问题。利用深度学习的理论，设计出一种模型驱动的神经网络，解决了混合精度量化信道估计问题。设计出一种可直接从网络权重获得优化导频符号的导频设计网络。提出一种通过可导函数生成选择向量的天线选择网络，实现混合精度ADC的分配优化。最终，通过一种基于自编码器的结构对网络进行联合训练，解决了混合精度ADC分配、信道估计以及导频设计联合优化问题。研究了毫米波/太赫兹大规模MIMO感知通信一体化技术，分析了大天线阵列、宽带传输场景下由于硬件限制所产生的波束偏移和波束分裂效应，设计出基于波束偏移和波束分裂效应的用户感知方案。利用波束偏移效应，使不同子载波的波束能够同时指向不同的方向，实现感知范围内的同时覆盖以及用户方位的感知，引入波束分裂效应，扩大感知范围并提升感知准确性。
Millimeter-wave (mmWave) massive multi-input multi-output (MIMO), a key technology for the next generation mobile communication systems, has several appealing advantages, e.g., increasing transmission rate, improving energy and spectral efficiency, and enhancing cellular coverage. However, owing to the semiconductor manufacturing challenges and the limitations of hardware cost and power consumption, low-cost hardware with impairments will be deployed in practical mmWave massive MIMO systems, which causes severe performance degradation. Hence, this thesis studies the key technologies and optimization methods for mmWave massive MIMO systems with hardware impairments, as follows.The impacts of hardware impairments on the channel estimation performance are investigated, and an approximate tractable expression for the uplink achievable rate of massive MIMO systems with hardware impairments is derived. Meanwhile, the appreciable compensations between low-resolution analog-to-digital converters (ADCs) and radio frequency (RF) impairments are illustrated.To address the uplink channel estimation problem for mmWave massive MIMO systems with one-bit ADCs, a compressed sensing-based algorithm is devised to recover the angular domain sparse parameters of mmWave channel. By exploiting the latent degree of freedom (DoF) offered by the amplifiers, we develop a proximal gradient descent-based method for downlink channel estimation and preccoding with one-bit digital-to-analog converters (DACs).Aiming at the mmWave massive MIMO systems with mixed-resolution ADCs, a deep learning-based joint pilot design, channel estimation and mixed-ADCs allocation method is proposed. Specifically, a pilot design neural network whose weights directly represent the optimized pilots is devised, and a model-driven network is developed as the channel estimator. Besides, a novel antenna selection network for mixed-ADCs allocation is designed to further improve the channel estimation accuracy. Then, an autoencoder-inspired end-to-end architecture is adopted to jointly optimize these networks.By taking advantage of joint beam-squint and beam-split effect, we propose a novel integrated sensing and communications scheme for mmWave/terahertz (THz) massive MIMO systems. Specifically, with the beam-squint effect, the base station (BS) steers the beams of different subcarriers towards distributive directions simultaneously, and then the users in these directions can be sensed. Moreover, the beam-split effect can be introduced and exploited to expand the sensing range and improve the sensing accuracy.
