Massive MIMO Channel Estimation Based On Deep Learning

The performance of the communication system is affected by the channel estimation technology.By obtaining accurate channel parameter estimation,the receiver of the system can accurately demodulate the transmitting signal.The massive Multi Input Multi Output(MIMO)system significantly improves the system capacity and increases the spectrum utilization,which is one of the core technologies of 5G.However,the antenna number of this system is different from that of the traditional MIMO system,so the traditional channel estimation technology is not suitable for the current system.Deep learning technology has been successful in image,video,voice and other processing fields.In this paper,deep learning technology and channel estimation technology in communication are integrated to provide a new thinking direction for channel estimation technology.In the research process,the fusion of OFDM,MIMO-OFDM,massive MIMO system channel estimation technology and deep learning technology is discussed.First,the channel estimation of OFDM system based on deep learning is proposed in the traditional OFDM system.In the OFDM system,the system is modeled as an autoregressive process.The LS algorithm is used to obtain preliminary channel estimates as the model data set.Then,according to the deep learning convolutional neural network technology,a one-dimensional volume matching the system is designed.The integrated neural network outputs the optimal autoregressive coefficients through iterative training,and finally uses this output value to monitor channel frequency domain changes in real time.Compared with the traditional LS and LMMSE algorithms,the OFDM system uses the new channel estimation method with better BER performance.Then,the channel estimation of MIMO-OFDM system based on deep learning is proposed.What the MIMO-OFDM research method has in common with the OFDM system is that the model data set is obtained by the LS method.The difference is that the MIMO-OFDM system also needs to consider the influence of the air space,and is used in the construction of the neural network.The feedforward neural network performs iterative training to track the relevant parameters of the channel.Finally,through simulation analysis,compared with the commonly used LS method and LMMSE method,the estimated mean square error is greatly reduced,so that the BER performance of the receiving end is improved.Finally,a massive MIMO channel estimation based on deep learning is proposed.For the scenario where the pilot length in a massive MIMO system is less than the number of antennas,a new solution is proposed by combining deep learning techniques with channel estimation techniques.The scheme is divided into two parts.First,a pilot and channel estimator are obtained by combining a two-layer neural network and a deep neural network,and then another deep neural network is used to improve the accuracy of the channel estimation technique.Through simulation analysis,based on in-depth study of massive MIMO channel estimation and LMMSE channel estimation algorithm in SNR range from 0 to 25 d B,the bit error rate and mean square error(MSE)of the former are lower,and analysis of the two algorithms is obtained its pilot length and MSE,and the relationship between the system capacity lower the system pilot length setting advice can be obtained.