Research On Detection Algorithms For Massive MIMO Systems

Multiple-input Multiple-output(MIMO) wireless systems can achieve both higher spectral efficiency and improved link reliability by transmitting multiple data streams concurrently and within the same frequency band. Thus, MIMO technology has been adopted in many wireless standards, including 3GPP LTE and IEEE 802.16 m WiMAX. However, due to the constantly increasing demands for higher data rates, these systems are already approaching their throughput limits. Massive MIMO, which makes a clean break through the use of very large number of antennas, is expected to be one of the key enabling technologies for the next-generation of wireless communications. The price to pay are the increased complexity and energy comsumption of the signal processing. In this dissertation, we focus on the high performance and low complexity algorithms for Massive MIMO detections.First, we discuss the opportunities and challenges provided by massive MIMO systems. We start with the analysis of the channel hardening and favorable propagation, which indicate that the channel becomes more and more deterministic as the number of antennas increase and the sum-capacity can achieve its largest possible value. Followed by the discussion of the spectral efficiency and energy efficiency, which shows the use of large antenna arrays can improve the spectral and energy efficiency with orders of magnitude. For the challenges, we illustrate the channel reciprocity and propose a reciprocity calibration method. The phenomenon of the pilot contamination in the multicellular system, which constitutes an ultimate limit on performance in massive MIMO system, is also analyzed.Secondly, we propose three optimized low-complexity linear detection algorithms based on minimum mean square error(MMSE) method for Massive MIMO system. To this end, we introduce the biased MMSE detection to simplify the calculation of the log likelihood ratio(LLR) in the massive MIMO system. Followed by the illustration of three proposed detection algorithms, which are derived from three different techonologies applied to the biased MMSE detection. The first technology employs the truncated Neumann series(NS) expansion, which can reduce the complexity by avoid matrix inversion. The second technology transforms the MIMO detection into solving the linear matrix by the Jacobi iterative method(JIM). The third one further transforms solving the linear matrix into minimizing the quadratic function with the use of conjugate gradient(CG) method. We also study the performance and complexity of these detectors, which demonstrate that the CG-based biased MMSE detection achieves the best trade-offs.Finally, we propose two novel non-linear detection algorithms, i.e, Markov Chain Monte Carlo(MCMC) and Belief Propagation(BP). We propose both bit-wise and symbol-wise MCMC algorithms based on Max-Log(ML) updating, which draw samples in log-domain directly. Two enhancement techniques, biased processing and normalized strategy, are also proposed to overcome the so called “stalling” problem for MCMC based algorithms. To further improve the performance, we derive a Max-Log with Second-Max(MLSM) updating method for the symbol-wise MCMC. Followed by the illustration of the BP detection, where we propose a binary BP(B-BP) algorithm for high order modulation. However, the performance loss of B-BP is serious except for 4-QAM modulation. Thus, we also propose a log non-binary BP(Log-NB-BP) algorithm. The simulation results of MCMC algorithm show that both bit-wise MCMC-ML and symbolwise MCMC-ML can achieve performance gains than minimum mean square error based on parallel interference cancellation(MMSE-PIC), and the symbol-wise MCMC-MLSM can approach the performance of quasi-MAP detector-single tree search(STS). The simulation results of BP algorithm demonstrate that Log-NB-BP can achieve better performance with lower complexity compared with the B-BP detection.Based on the results obtained throughout this dissertation, we find that the linear biased-MMSE detectors based on NS, JIM and CG are feasible for the multi-user Massive MIMO, where the service antennas have a large excess over active terminals. The MCMC and BP algorithms are feasible for the point-to-point Massive MIMO, where the antenna configurations in the transmitter and receiver are in the same order.