Massive multiple-input multiple-output (MIMO) is a key ingredient of the fifth-generation (5G) cellular communication networks with high energy and spectral efficiency. When the number of base station (BS) antennas is very large, it can provide satisfying performance with simple linear processing. But massive antenna arrays bring high computational complexity and deployment cost. In order to reduce the complexity and cost, this thesis deeply studies the project named ``Research on High-Efficiency Transmission Schemes for Massive MIMO Systems" from the aspects of linear processing and antenna technology. By making full use of statistical channel state information (CSI), multiple high-efficiency transmission schemes and performance analysis are given. The contents are the channel predictor based on polynomial fitting for massive MIMO systems, joint optimization of number of antennas and transmit power based on maximizing energy efficiency in massive MIMO systems, uplink power control with truncated polynomial expansion (TPE) receiver in massive MIMO systems, low-complexity two-stage beamforming for three-dimensional (3D) massive MIMO and transmission algorithm of exploiting BS antenna tilt for simultaneous wireless information and power transfer (SWIPT) in 3D massive MIMO systems. The detailed contents and main contributions are listed as follows:
1. In massive MIMO systems with delayed CSI, we propose a low-complexity channel prediction based on polynomial fitting to reduce the computing time and process latency. The predictor based on polynomial fitting needs very few CSI samples to estimate the current one. Compared with conventional Wiener predictor, polynomial predictor does not need to estimate the channel statistics, and avoids matrix inversion. We analyze the approximate SINR which only depends on statistical CSI and is numerically validated tight. We derive the approximate gap of per-user average rate between perfect CSI, Wiener predictor and polynomial predictor. We also analyze the normalised mean square error (NMSE) of prediction. We investigate the performance under a more practical and more general angle of departure (AoD) spectrum model which has a concentration direction and a spreading factor. When the concentration direction and spreading factor are relatively small, the performance is satisfying with a proper prediction order.
2. In the massive MIMO system with minimum mean-square-error (MMSE) receiver, we propose an iterative algorithm to jointly optimize the number of BS antennas and user transmit power for maximizing the energy efficiency. Firstly, we derive the approximate achievable rate with MMSE recevier. This approximation is very tight with the actual value via simulations. Accordingly, the optimization problem is formulated, aiming at maximizing the energy efficiency subject to the target of achievable rate and the constraint of transmit power. We transform the optimization problem, which is a fraction programming problem, into a linear programming problem. We propose an iterative algorithm to jointly optimize the number of BS antennas and user transmit power. The algorithm only requires channel statistics instead of large-dimensional instantaneous channel realizations. So, the algorithm only has to run once in multiple channel coherence periods which has low complexity.
For comparison, we investigate the zero-forcing (ZF) scheme. Simulations show that our proposal with MMSE receiver outperforms ZF receiver in terms of energy-efficiency and can significantly reduce the number of BS antennas and BS deployment cost.
3. We propose an uplink power control algorithm for uplink massive MIMO systems with TPE receiver, alternatively optimizing transmit power and polynomial coefficients. TPE receiver can avoid the matrix inversion in ZF and MMSE receivers with low complexity. Firstly, we derive the associated approximate SINR. Accordingly, the optimization problem is formulated aiming at minimizing the total transmit power of all users, subject to per-user SINR target and per-user transmit power constraint. The proposed algorithm depends only on channel statistics. For comparison, we investigate the uplink power control with ZF and MMSE receivers. When the number of BS antennas is large, the complexity of TPE receiver is the lowest. Simulations reveal that when the SINR target is low, the performance of TPE receiver is very close to that of ZF and MMSE receivers, when the SINR target is high, we can add more TPE terms or antennas to reduce the performance gap.
4. We propose low-complexity two-stage beamforming for 3D massive MIMO systems with a uniform rectangular array (URA), where the second-stage prebeamforming is the Kronecker product of horizontal and vertical discrete Fourier transform (DFT) prebeamforming. DFT prebeamforming splits the cell. We propose a low-complexity user grouping algorithm based on statistical CSI. DFT prebeamforming and instantaneous CSI form the low-dimensional effective channel. With the low-dimensional effective channels, signal-to-leakage-and-noise-ratio (SLNR) precoding is adopted for each group. Because inter-group interferences mainly come from adjacent groups, SLNR precoding is used to combat the intra-group and adjacent-group interferences. For comparison, we study the SLNR precoding considering all-group and intra-group interferences. The approximate SINR of our proposal is derived using deterministic equivalent method. Simulations validate that the approximations are tight, and our proposal achieves a good balance between performance and complexity.
5. For 3D massive MIMO systems, we propose a transmission algorithm which exploits BS antenna tilt for SWIPT. The BS applies matched filtering (MF) precoding and users adopt power splitting (PS) technique for information decoding (ID) and energy harvesting (EH). In order to improve the SWIPT efficiency, we exploit the vertical domain and jointly optimize the BS antenna tilt, allocated power and PS ratios. The optimization problem aims at minimizing the transmit power subject to SINR and harvested power targets. We propose an iterative algorithm which only requires statistical CSI. Simulations show that the performance of the proposal is close to that of brute-force search, and outperforms the conventional systems with an adjustable tilt.