Research On Massive MIMO Hybrid Beamforming Based On Neural Network

The use of high frequency bands in the 5G era brings high bandwidth and high rates to communication systems,but also accelerates signal attenuation due to its own characteristics.In order to solve the problems caused by high frequency bands,massive MIMO beamforming technology has gradually developed and matured.For this technology,it has been found that if all digital beamforming codecs are used at both ends of the transceiver,it will cost a lot.Therefore,relevant research scholars propose to divide beamforming into two parts: low-dimensional baseband digital beamforming and high-dimensional RF analog beamforming to form a hybrid beamforming(HBF)architecture.Therefore,the expenses will be effectively reduced.This paper uses this architecture to establish a relevant communication system model.Then this paper explores how to better solve the corresponding beamforming matrixs at both ends of the transceiver when maximizing the spectrum efficiency of the system.In a point-to-point single-user massive MIMO hybrid beamforming communication system,this paper first introduces a traditional element iteration algorithm to solve the related beamforming matrix and carries out experimental simulations.Then,based on the traditional algorithm,this paper proposes a neural network model based on unsupervised learning to train the correlation beamforming matrix.At the same time,this paper also uses the neural network based on supervised learning to train the correlation beamforming matrix by using the calculation results of the traditional iterative algorithms.Finally,the simulations are used to compare and analyze the results of different methods.The simulation results show that the neural network method based on unsupervised form used in this paper has more obvious advantages.After solving the single-user situation,this paper studies the massive hybrid beamforming system under the multi-user situation.This paper first uses a traditional zero-forcing algorithm to solve the related beamforming matrixs.However,the limitations of the traditional algorithm and the system performance obtained by the algorithm still have a certain gap with the best pure digital beamforming system.So,this paper continues to use the neural network based on unsupervised form to train the beamforming matrix.Finally,a simulation comparison shows that the neural network used in this paper is superior to the traditional algorithm in terms of maximizing the spectral efficiency of the system.At the same time,the neural network can more closely approximate the system performance achieved by the best pure digital beamforming.In short,a series of related studies on hybrid beamforming show that the neural network model proposed in this paper can obtain or even exceed the system performances achieved by traditional algorithms.In addition to this,the advantages brought by neural networks greatly reduce the complexity of the problems.