A CNN-based Massive MIMO Communication Systems

As an effective means to improve system reliability and spectrum effectiveness,massive Multiple Input Multiple Output(MIMO)technology is regarded as one of the main implementation methods of the fifth-generation communication system.Although massive MIMO technology has many performance advantages,there are still many challenges to be solved in the implementation process of precoding technology,channel estimation and signal detection.Therefore,in order to fully explore the potential advantages of massive MIMO,this thesis introduces deep learning into the design of communication system,and conducts research on the design scheme of massive MIMO system based on deep learning.The specific research work is as follows:In order to improve the adaptability of the existing end-to-end communication system and verify the feasibility of the MIMO system physical layer scheme based on deep learning,this chapter combines the physical layer representation of the MIMO system and the Orthogonal Frequency Division Multiplexing(OFDM)modulation and demodulation process to form an end-to-end autoencoder system.Meanwhile,a MIMO-OFDM system based on autoencoder is proposed.Through end-to-end joint training,the overall performance of the system is optimized.The simulation results show that the training model has good adaptability and generalization ability under different channel environments and space-time coding schemes,which verifies the feasibility of the scheme.Meanwhile,the proposed system can realize a signal mapping method suitable for the current channel environment according to the potential characteristics of source information,and improve the transmission reliability of the system.Aiming at the problems of poor performance of existing massive MIMO signal detection methods and low accuracy of channel estimation methods,a massive MIMO channel estimation and signal detection system based on Convolutional Neural Network(CNN)is proposed.This scheme introduces a CNN-based autoencoder into the MIMO system,and combines the channel estimation network to achieve channel equalization.The proposed system realizes signal mapping and transmission under different scales of MIMO systems to obtain the best receiving performance.Simulation experiments show that in the massive MIMO uplink scenario with a small number of users,the proposed system can match the maximum likelihood detection performance in the case of low signal-to-noise ratio,and achieve better signal detection performance than traditional nonlinear methods in the case of high signal-to-noise ratio.Compared with the traditional method,the proposed channel estimation method can obtain a performance improvement of about 5.5 d B under long term evolution extended typical urban model.Aiming at the high complexity of signal processing at the receiving end of massive MIMO,a CNN-based massive MIMO precoding system is proposed.The proposed system introduces a precoding module at the transmitting end,precoding the modulated signal in combination with channel information,and uses an end-to-end learning method to enable the transmitting end to find the best link for signal transmission.At the same time,the receiving end only needs to demodulate the signal to recover the source information.Simulation experiments show that in the massive MIMO downlink scenario with a small number of users,the proposed system can select the best transmission link based on the provided channel information,and exhibits better error performance than traditional nonlinear precoding schemes.