Deep Learning-based Channel Estimation In Millimeter-wave Massive MIMO Systems

Millimeter-wave massive MIMO,as a key technology of 5G communication systems,has attracted widespread attention because of its ability to increase system capacity and increase spectrum utilization.Channel estimation is an important means to reflect the communication quality,and its accuracy affects the performance of the communication system.The traditional channel estimation method relies on the pilot signal and the prior statistical characteristics of the channel,resulting in higher requirements for the quality of the pilot.The application of deep learning methods to channel estimation has important research value.There is no need to know the system channel characteristics,and the estimation accuracy can be effectively improved by learning the data generated by the system.Aiming at the problem that the pilot channel estimation method will lead to excessive pilot overhead and poor estimation accuracy,a two-stage channel estimation scheme based on deep neural network is proposed.In the first stage,the training data containing the pilot is input into the network model,and the Dropout strategy is used to prevent over-fitting.The second stage uses the trained model to input the test and verification data and estimate the channel parameters.The experimental results show that compared with the four algorithms of Least Squares(LS),Minimum Mean Square Error(MMSE),Orthogonal Matching Pursuit(OMP)and Long ShortTerm Memory(LSTM),in the two cases of the same number of pilots and estimation accuracy,the deep neural network method improves the estimation accuracy by about 1.58 d B and reduces the pilot overhead by about 32.4% respectively.And in the case of different numbers of pilots,the estimation accuracy of the four algorithms that use a larger number of pilots is lower than that of the deep learning method that uses fewer pilots.In view of the sparse characteristics of millimeter-wave massive MIMO channels,the system is affected by noise factors resulting in low channel estimation accuracy,a convolutional neural network channel estimation method based on attention mechanism is proposed.Firstly,consider the channel matrix as a two-dimensional image.Then,construct an attention mechanism network and embed it in a convolutional neural network to enhance the saliency of noise features in the image.Finally,the noise is extracted through the network model and the denoised image is obtained,that is,the channel matrix is estimated.The simulation results show that,the estimation accuracy of the proposed convolutional neural network channel estimation method based on the attention mechanism is improved by about 1.86 d B on average,compared with the Least Square(LS),Minimum Mean Square Error(MMSE),CNN and Denoising Convolutional Neural Network(Dn CNN)algorithms.