大规模 MIMO 技术通过在基站部署大量天线可以以显著提高的频谱效率同时服务多个用户，因而它是目前 5G 公认的一项关键技术。然而，在香农奈奎斯特采样框架下，大量天线会引入高维度的信号处理，这会导致大规模 MIMO 系统过高的训练开销、计算复杂度、成本和功耗。为了解决这些问题，本文借助全新的压缩感知理论，通过挖掘和利用大规模 MIMO 系统中信道或信号存在的稀疏性，提出了一系列大规模 MIMO 中基于压缩感知理论的无线传输技术方案，其性能明显优于香农奈奎斯特采样框架下设计的传统方案。本文的创新点和贡献如下：首先，传统下行信道估计方案在 FDD 大规模 MIMO 中需要极高的导频开销。为解决这一问题，通过利用延时域大规模 MIMO 信道的结构化稀疏性，我们在结构化压缩感知理论下提出了重叠导频及用户处对应的信道估计方法。提出方案所需的导频开销仅和多径分量较小的数目有关，明显低于传统方案所需的开销。其次，传统信道反馈方案在 FDD 大规模 MIMO 系统中需要极高的反馈开销。为解决这一问题，我们利用角度域大规模 MIMO 信道的分布式稀疏性，基于分布式压缩感知理论，提出了一种低开销的自适应信道反馈方案。该方案由基于压缩感知的自适应信道状态信息获取和闭环信道追踪两部分构成，既可确保基站精确获取下行信道状态信息，又可大幅降低信道反馈开销。再次，毫米波大规模 MIMO 技术通常采用远小于天线数的射频链路数来降低系统成本和功耗，而传统的信道估计方案无法直接用于这种收发机结构。为解决这一问题，我们利用毫米波信道的准直射传播特性，提出了一种基于自适应压缩感知理论的高精度上行多用户信道估计方案。通过采用自适应的观测矩阵，本方案可有效解决连续入射角和出射角带来的信道估计能量泄露问题。最后，为了降低大规模 MIMO 系统中大量射频链路引入的高功耗和高成本，我们提出了一种大规模空间调制 MIMO 系统上行传输方案。通过利用上行空间调制信号内在的分块稀疏性，我们提出了基于分块稀疏压缩感知理论的上行多用户信号检测算法，该方案可以以低复杂度获得准最优的信号检测性能。本文通过挖掘大规模 MIMO 系统中信道或信号存在的稀疏性，从压缩感知理论这一新视角出发，在信道估计、信道反馈、信号检测等大规模 MIMO 系统关键技术上提出了全新的解决方案，为大规模 MIMO 系统从理论走向实践提供了相关的理论分析和具体传输方案。
By exploiting hundreds of antennas at the base station (BS) to simultaneously serve a set of users, massive multiple-input multiple-output (MIMO) can improve the spectrum efficiency by orders of magnitudes, and it has been considered as a key 5G technique. However, the large number of antennas will introduce high-dimensional signal processing, which will bring the prohibitively high training overhead, computational complexity, system cost, and power consumption. To solve these challenging problems, in this thesis, by exploiting the inherent sparsity of channels or signals in massive MIMO systems, we propose a series of emerging compressive sensing (CS)-based schemes. These schemes outperform the conventional counterparts designed under the framework of Shannon-Nyquist sampling theorem. To be specific, the contributions of this thesis lie in the following several parts.First, conventional channel estimation schemes will suffer from the prohibitively high pilot overhead in frequency division duplexing (FDD) massive MIMO. To solve this problem, by exploiting the structured sparsity of delay-domain massive MIMO channels, we propose the superimposed pilot design at the BS and the associated reliable channel estimator at the user under the framework of structured compressive sensing (SCS). For the proposed scheme, the required pilot overhead is only proportional to the small number of dominated paths, which is much smaller than the pilot overhead required by its conventional counterparts.Second, conventional channel feedback schemes will suffer from the high channel feedback overhead in FDD massive MIMO systems. To solve this problem, by leveraging the distributed sparsity of angle-domain massive MIMO channels over system bandwidth, under the framework of distributed compressive sensing (DCS), we propose an adaptive channel feedback scheme with low feedback overhead. The proposed feedback scheme consists of the DCS-based adaptive CSI acquisition and the following closed-loop channel tracking, which can ensure the BS to reliably obtain the high-dimensional CSI from the low-dimensional non-orthogonal pilot signal.Third, millimeter-wave (mmWave) massive MIMO typically employs the much smaller number of radio frequency (RF) chains than that of antennas for reduced cost and power consumption, which will lead the conventional channel estimation schemes to be not directly used in such systems. To solve this problem, by using the near line-of-sight of mmWave channels, under the framework of adaptive CS, we propose an uplink multi-user channel estimation scheme. By designing the reference signal and the associated channel estimator, the proposed scheme can effectively solve the power leakage due to continuous angle of arrival or departure, and it can estimate the mmWave massive MIMO frequency selective fading channels.Finally, to reduce the cost and power consumption from large number of RF chains required by large number of antennas in massive MIMO, we propose a massive spatial modulation (SM) MIMO uplink transmission scheme and the associated multi-user detector at the BS. By leveraging the block-sparsity of the uplink SM signals, under the framework of block-sparse CS, the proposed MUD is capable of achieving the near-optimal performance with low computational complexity.From the novel perspective of CS theory, this thesis investigates several key techniques including channel estimation, channel feedback, and signal detection in massive MIMO by exploring and exploiting the sparsity of channels or signals. We hope our proposed scheme can provide the theoretical analysis and specific transmission schemes for massive MIMO from theory to application.
