| The massive Multi-input Multi-output(MIMO)technology can multiple increase the system capacity and improve the spectrum efficiency by using multiple input and output antennas.For multi-user massive MIMO systems,precoding the transmitted signal can effectively reduce Inter-User Interference(IUI)and improve spectral efficiency(SE),while channel state information(CSI)is an important part of formulating a precoding scheme.In a fast time-varying channel,the CSI obtained by the base station through the uplink channel estimation will soon expire,thereby significantly reducing the performance of downlink precoding.Therefore,CSI prediction is very necessary.This paper proposes a downlink channel prediction method based on deep learning(DL),and its contribution can be summarized as:1.This paper provides a DL-based Time Division Duplex(TDD)scheme,in which CSI is obtained through a DL-based predictor instead of a traditional pilot-based channel estimator.2.This paper designs a mixed architecture of Recurrent Neural Network(RNN)and Deep Neural Network(DNN),which called MRDNN based-channel prediction method.The main idea is to use DNN to interpolate the channel response at a small number of pilot subcarriers to obtain the channel response at non-pilot subcarriers,and use RNN to perform channel prediction on the time series of known channel response values.3.This paper evaluates the performance of the proposed MRDNN-based channel predictor through simulation experiments.The results show that the prediction accuracy of the channel predictor based on MRDNN is better than other traditional channel predictors in low and medium mobility scenarios.According to the known downlink CSI,the system performance can be further improved by precoding.Therefore,this article proposes a DL-based downlink precoding method,and its contribution can be summarized as follows:1.This paper proposes a channel emulator(CE-DNN)based on DNN to simulate the signal transmission process in the actual channel.2.This paper proposes a precoding method based on the Cascaded DNN architecture of the joint training of two DNNs.Specifically,the Cascaded DNN uses CE-DNN-assisted precoding DNN(P-DNN)for joint training,and then divides the CDNN to obtain a P-DNN with adaptive precoding capabilities.3.This paper provides the performance analysis of the CE-DNN-based channel simulator and the P-DNN-based precoder.The results show that the CE-DNN-based channel simulator does not rely on accurate CSI,and the P-DNN-based Compared with traditional precoders,the precoder still has higher precoding performance when using low-precision CSI. |
