Research On Low-complexity Precoding And Signal Detection For 5G Massive MIMO Systems

With the popularity of intelligent terminals access network,and the explosion growth of data traffic,wireless network capacity will meet with an unprecedented challenge.To address this challenging,5th Generation mobile communications systems(5G)can increase wireless network capacity by increasing node density,developing new frequency bands and improving spectrum efficiency.By making full use of spatial dimension,Massive Multiple-Input Multiple-Output(Massive MIMO)technology can significantly improve system capacity,spectrum efficiency,energy efficiency and link stability performance,therefore,it is generally recognized as one of the key candidate technology for 5G network.It is generally known that precoding and detection technologies are the core of Massive Multi-User Multiple-Input Multiple-Output(MU-MIMO)systems,because they can effectively reduce the interference between users,and significantly enhance the system performance.With the increasing number of base station antennas,precoding and signal detection techniques will involve very high complexity.To address this issue,low-complexity precoding and signal detection strategies are design in this paper.The main contributions are listed as follows.In the first part,for downlink under-loaded massive MIMO systems,a new low-complexity precoding scheme with high parallelism and hardware-efficiency is proposed.Firstly,by leveraging the matrix decomposition and the matrix polynomial series accumulation method,a high parallelism Regularized Zero-Forcing(RZF)precoding is designed with the consideration of many practical factors,such as,small scale fading,large scale fading,delay and implementation complexity.Secondly,by constructing the difference equation and utilizing Random Matrix Theory(RMT)tools,the solution process of the matrix polynomial factor is simplified significantly.Then,a simple closed expression of optimization coefficient is deduced.Furthermore,according to different Massive MIMO systems configuration and Signal to Noise Ratio(SNR),the matrix polynomial terms can be flexibly selected to achieve a better trade-off between system sum rate and hardware overhead,when base station uses power allocation strategy.Meanwhile,the proposed precoding could enable pipelining to execute the diagonal matrix-vector and hollow matrix-vector multiplication.Next,the case that more UTs can be served in under-loaded massive MIMO systems is taken into account.Based on jointing Preconditioning Steepest Descent(PSD)algorithm and Diagonal Neumann Series(DNS)iteration,a high parallelism and fast convergence precoding scheme is proposed,which is named Preconditioning Steepest Descent-Diagonal Neumann Series(DNS-PSD)algorithm.In the proposed scheme,the complex massive MIMO precoding can be transformed into the solution of the linear equation.Then,by using DNS-PSD iteration method,the problem of solving linear equation can be transformed into solving the minimum of a quadratic function.In addition,by utilizing the excellent properties of DNS matrices,a good preconditioning matrix of Steepest Descent(SD)algorithm is designed.Then,based on the merger of the DNS iterative method,the proposed DNS-PSD iterative precoding not only greatly improves the convergence rate,but also guarantees a wide range convergence.In the second part,for uplink under-loaded massive MIMO systems,a novel signal detection mechanism is proposed to fast update the inverse of a matrix after a small perturbation when a user is added in or removed from the systems frequently.The core module of the signal detection mechanism includes the update matrix module and the initial input inverse matrix module.Firstly,by leveraging block matrix irnverse lemma,two update algorithms named Inflate expansion and Deflate compression are presented.Secondly,benefiting from the linear iterative theory,RMT and the matrix theory tools,two on-the-fly detection schemes named Zero-Forcing Matrix Decomposition Polynomial Expansion update(ZF-MDPE-update)and Zero-Forcing Successive Over-Relaxation update(ZF-SOR-update)detections are proposed to update a zero-forcing(ZF)detection quickly.As the proposed detection schemes can directly and fast update the detection matrix after a small perturbation,the complexity of signal detection decreases drastically.Therefore,the contradiction between the performance and the computational complexity is further reduced.Theoretical analysis and simulation results demonstrate that without recomputing the entire high-dimension inverse matrix,the proposed update detection schemes not only have significantly lower complexity,but also achieve the near-optimal performance of the exact ZF detection.In the final part,for uplink over-loaded massive MU-MIMO systems,we propose a signal detection scheme which has low-complexity and high-spectrum efficiency.Firstly,the equivalent real-valued signal model is established.Based on this module,the performance of widely linear signal detector is analyzed.Then the mapping relationship between the Widely-Linear Minimum Mean Square Error(WL-MMSE)detection and the conventional Minimum Mean Square Error(MMSE)detection is derived.Secondly,based on the Chebyshev polynomials accelerated Symmetric Successive Over-Relaxation(SSOR)non-stationary iterative algorithm,we propose a low-complexity and high spectrum-efficiency WL-MMSE signal detection for uplink over-loaded massive MIMO systems.In the algorithm design,the Chebyshev polynomials are utilized to construct a novel secondary iteration of SSOR.This algorithm breaks up the traditional idea of using the stationary iterative process to reduce the signal detection complexity for uplink massive MIMO systems.And puts forward the idea of using non-stationary iterative process which can noticeably speed up the convergence rate and enhance system performance.Based on the real-valued signal model,the proposed detection provides obvious advantages in system sum rate and Bit Error Ratio(BER)performance compared with both the recently reported methods(e.g.,linear iterative detections,optimal polynomial expansion(PE)detections)and the conventional detections(e.g.,MMSE,Matched Filter(MF)).Additionally,the proposed detection has low storage cost and is also insensitive to the spatial correlation.