Deep Learning Based Detection And Channel Tracking For MIMO Systems

With the advent of the information age,mobile communication technology has gradually become the main communication method for daily communication and interflow.How to maximize spectrum utilization,expand system capacity,and increase data transmission speed has become an urgent need for research and solution in wireless communication problem.The emergence of Multiple Input Multiple Output(MIMO)technology has solved the problems of channel capacity and transmission rate to a certain extent.In recent years,deep learning has demonstrated good performance in many areas,such as data mining,machine translation,and natural language understanding.Therefore,it is very necessary to apply data-driven deep learning methods to signal detection,channel estimation,modulation recognition,and other communication fields.This article uses deep learning tools to study two key points in MIMO technology——signal detection and channel state information acquisition.The research content includes the following two aspects:1)MIMO signal detection algorithm based on BD-NetAt present,the existing MIMO detection algorithm has many problems such as high calculation complexity,strong assumptions,and difficulty in obtaining noise information.The DetNet network based on deep learning does not require noise statistics during signal detection,and has close to optimal detection performance,but there are problems that the network training parameters are large,the network convergence speed is slow,and it is assumed that the channel is known.The BD-Net(BiLSTM-Detection Network)network uses the projection gradient descent method to optimize the maximum likelihood estimation of the transmitted signal,and uses the fitting and generalization characteristics of the neural network to design a neural network with a bidirectional LSTM as the basic unit.The number of iterations of the projection gradient descent algorithm corresponds to the number of network layers.The simulation results show that,while reducing the computational complexity,BD-Net improves the detection accuracy of the transmitted signals under different modulation modes to a certain extent.2)Time-varying channel tracking algorithm based on CsiBD-NetThe DetNet needs to know the channel state information during the detection process.Aiming at the problem of signal detection when the channel state is unknown,CsiBD-Net network is proposed.CsiBD-Net fix all hyperparameters in pre-training BD-Net,and transform the channel state information from a known quantity to a trainable parameter.Use the training sequence to obtain the rough estimate channel as the initial value of the trainable channel parameter in CsiBD-Net.By continuously fine-tuning the channel parameter during the online training process,real-time tracking of the slowly varying time-varying channel is realized,and the channel tracking value is used to realize the estimation problem of transmission signal.In addition,in order to further improve the accuracy of signal detection,an improved combined network based on CsiBD-Net is proposed.The combined network combines the two network structures of CsiBD-Net and pre-trained BD-Net.The simulation results show that the signal detection accuracy is improved while ensuring the channel tracking convergence speed.