Advanced Precoding and Detection Techniques for Large MIMO Systems

Multiple-input multiple-output (MIMO) transmission has been at the core of wireless communication research for the past two decades. Driven by the explosive increase of data demand, the development of MIMO systems has entered a large-scale realm where there are dozens of or even more than a hundred antennas and users. The large number of antennas can significantly boost the system throughput and robustness against noise. However, the physical realization of such a large MIMO system can be very complicated and expensive. On the one hand, optimal signal processing algorithms usually have complexities that increase rapidly in the numbers of antennas and users. On the other hand, large number of antennas means increased hardware overheads, such as those of power amplifiers and D/A converters. This thesis considers efficient precoding and detection algorithms that can reduce implementation complexity and cost. Specifically, the thesis consists of the following three parts:;In the first part, we consider a fundamental problem in MIMO communication, namely MIMO detection. The traditional lattice decoding methods, as well as its efficient approximations by lattice reduction aided (LRA) methods, relax the symbol bounds in detection and thus suffer from performance loss. We propose a systematic adaptive regularization approach to lattice decoding to alleviate the adverse effect of symbol bound relaxation, which is based on the study of a Lagrangian dual relaxation (LDR) of the optimal maximum-likelihood (ML) detector. We find an intriguing relationship between lattice decoding and ML, which was not reported in the previous literature. Simulation results show that the proposed LDR approach can significantly outperform existing lattice decoding and LRA methods.;In the second part, we consider the vector perturbation approach which is a promising technique to achieve near-sum capacity and allows simple user processing in the multiuser multiple-input single-output (MISO) downlink scenario. However, the conventional vector perturbation designs can have very high per-antenna powers, which causes significant difficulty to power amplifier implementations. To tackle this problem, we propose a vector perturbation design with per-antenna power constraints (VP-PAPC). The resulting optimization problem is an integer program which requires a computationally demanding enumeration process. Lagrangian dual relaxation is used to transform the VP-PAPC problem into standard integer least square problems which may have efficient approximations. Simulation results show that the proposed method can effectively reduce the power back-off caused by high per-antenna power in conventional vector perturbation.;In the last part, we consider constant envelope (CE) precoding in the single-user MISO downlink scenario. CE precoding is recently proposed as a mean to utilize cheap but power-efficient power amplifiers in very large MIMO systems. We provide complete solutions to some fundamental signal processing issues in CE precoding which were only partially solved in the previous literature. In addition, we enhance CE precoding with antenna subset selection for transmit optimization and implementation cost reduction. Simulation results reveal that the proposed method only exhibits moderate power loss compared to non-CE beamforming but have the advantages of CE transmission and fewer active transmitting antennas. II.