Deep Learning Based Wireless Channel Characteristics Prediction In High-Speed Railway Scenarios

In recent years,the stability and speed of high-speed trains continue to improve with the continuous advancement of technology,China’s achievements in the field of highspeed railways have attracted worldwide attention,and people’s travel increasingly depends on the fast and convenient high-speed railway transportation system.The application of 5G in high-speed railway broadband mobile communication technology that inject new vitality into the passage of railway transportation has become an important part of the intelligent development of high-speed railway.Because the channel state information obtained by traditional channel estimation algorithms lacks reference value in high-speed environments,the combination of artificial intelligence technology to study wireless channel prediction methods has become a new research goal.Deep learning has gradually become one of the research directions of many researchers in the field of intelligent communication in the future,and it is natural to extend the method of deep learning to channel prediction.Therefore,this paper studies the application of deep learning in channel prediction in high-speed railway mobile scenarios,and proposes more research methods in wireless channel prediction.This article first analyzes the different fading characteristics of the large-scale and small-scale in the process of wireless channel propagation,and determines the main content of the experimental research.Then it introduces the knowledge of deep learning,uses feedforward neural networks to illustrate the basic principles of neural networks,and introduces several deep learning frameworks most commonly used by researchers now,and clarifies the tools used in the experiments in this article.From the introduction of wireless communication and deep learning,the overall cognition of using deep learning methods to predict channel characteristics is obtained.The cutting simulation scenario of high-speed railway is established by using the principle of random propagation graph to obtain the data set required for training.Then it is proposed to use Long Short-Term Memory(LSTM)to construct a channel prediction model,fully combining the characteristics of LSTM network to learn long time series,and processing time-dependent channel state information of Single Input Single Output(SISO)communication systems.This paper uses three evaluation indicators that RMSE,MAPE and R2 to evaluate the results of neural network prediction models,and use coordinate-based Back Propagation Neural Network,Deep Neural Network,Recurrent Neural Network prediction methods to conduct comparative analysis.Finally,the problem of high-speed mobile Multiple Input Multiple Output(MIMO)channel prediction is studied.First,the principle of Convolutional Neural Network(CNN)is explained,and CNN is used to learn the spatial topological relationship of multiple antennas in the MIMO communication system to obtain research ideas for learning spatial correlation.Then the MIMO multi-channel prediction problem is modeled,and a ConvLSTM neural network prediction model that combines the characteristics of LSTM and CNN is proposed,and multi-dimensional antenna information is added to the learning of the neural network.The paper reflects the characteristics of the input data through autocorrelation analysis and similarity analysis,which provides reference and verification for the comparison and selection of neural network hyperparameters.Finally,two prediction modes are used to compare the performance of different prediction methods,and it is found that the proposed prediction model is more accurate and efficient,which provides a new method for channel prediction of high-speed railway MIMO communication systems.