Research On Hybrid Precoding In MmWave Massive MIMO Systems

Millimeter-wave(mmWave)communication provide technical support to realize the high-rate data transmission.However,the signal is faced with serious path loss when propagating through high-frequency electromagnetic wave.Supported by hybrid precoding,massive multi-input multi-output(MIMO)technology can provide antenna array gain to compensate the serious path loss of mmWave communication.The main problem of the existing hybrid precoding schemes is the imbalance between the spectral efficiency and the complexity.Therefore,this dissertation studies the hybrid precoding scheme with high spectral efficiency and low complexity,aiming at reducing the complexity of the algorithms with high spectral efficiency and improving the spectral efficiency of the algorithms with low complexity.The main research work and innovations of this dissertation are as follows:(1)In order to solve the problems of the existing hybrid precoding schemes based on codebook,such as the difficulty of obtaining codebook consisting of array response vectors,the poor correlation between orthogonal codebook and channel,and the high complexity of orthogonal matching pursuit(OMP)algorithm,a low-complexity hybrid precoding scheme based on codebook is studied.Firstly,a random codebook is designed.Compared with the codebook consisting of array response vectors,the random codebook can avoid the overhead of departure angles and arrival angles estimation,and has a higher correlation with the channel than the orthogonal codebook.Secondly,based on the designed random codebook,a hybrid precoding algorithm based on generalized OMP(gOMP)is proposed,which can reduce the complexity of OMP algorithm.Simulation results show that the proposed random codebook is close to the codebook consisting of array response vectors and much higher than orthogonal codebook in terms of spectral efficiency or sum-rate.In addition,the spectral efficiency or sum-rate of the proposed gOMP algorithm is close to that of the OMP algorithm,and the complexity is significantly reduced.(2)Aiming at reducing the high complexity of the existing non-codebook hybrid precoding scheme,a low-complexity non-codebook hybrid precoding scheme is studied.For single-user scenarios,an equivalent single-hidden layer neural network architecture is proposed,and a hybrid precoding scheme based on adaptive gradient back propagation is proposed inspired by the backpropagation algorithm.Simulation results show that the spectral efficiency of the proposed algorithm is close to the full-digital precoding,and the complexity is lower than that of the competing algorithms.Further,for multi-user scenarios,a low-complexity hybrid scheme is proposed based on successive interference elimination(SIC)and adaptive gradient(AG).First,through the simplification of the expression of the sum-rate,the optimal analog precoder is obtained based on SIC.Then,the AG algorithm is put forward to design the analogand digital-combiner.Simulation results show that when the signal-to-noiseratio(SNR)is higher than a certain value,the proposed algorithm has higher sum-rate and lower complexity than the competing algorithm.(3)In order to solve the problems of the existing hybrid precoding schemes based on low-resolution analog phase shifters(APSs),such as the high complexity of algorithms with high spectral efficiency and the low spectral efficiency of algorithms with low complexity,a high spectral efficiency and low complexity hybrid precoding scheme based on low-resolution APSs is studied.Inspired by cross-entropy(CE)algorithm,a hybrid precoding algorithm based on tree-coding-aided adaptive-cross-entropy(TC-ACE)is proposed.The main idea is to randomly generate some candidate analog processing matrices according to a series of pre-defined probability distributions,and select some elites to update the probability distributions.In order to derive a closed-form solution of the probability distributions,tree code is designed to map each element of the analog processing matrix to a binary number.The updated probability distribution continues to produce a candidate analog processing matrix that repeats the process of selecting elite candidates to update the probability distribution.Through iteration,the optimal solution of analog processing matrix can be obtained with sufficiently high probability.Simulation results show that the complexity of the proposed scheme is lower than that of the algorithms with high spectral efficiency,and the spectral efficiency or sum-rate of the proposed scheme is better than that of the algorithm with low complexity.(4)Aiming at the problems of high complexity of traditional hybrid precoding algorithms,low spectral efficiency of traditional hybrid precoding algorithms under imperfect channel conditions and high power consumption of the fully-connected hybrid precoding architecture,a low-complexity hybrid precoding scheme based on convolutional neural network(CNN)for partiallyconnected hybrid precoding architecture is studied.The nonlinear relationship between the analog matrices and the channel matrix is mapped to the CNN network,which will be trained by regarding the channel matrix and the analog matrices as the input and the label respectively.In the prediction phase,the channel matrix is input into the optimized CNN network,and the optimal analog matrices can be obtained quickly.The simulation results show that the optimal analog matrices can be obtained quickly by using the CNN method.In terms of the sum-rate,the CNN scheme is superior to the competing algorithm under imperfect channel condition.In terms of the energy efficiency,the CNN scheme outperforms the fully-connected hybrid precoding.