Research On Channel Estimation Technology Based On Deep Learning For Internet Of Vehicle Communication

Intelligent Transportation Systems(ITS)are the foundation for supporting unmanned and intelligent transportation such as vehicle formation,vehicle road coordination,and autonomous driving,and are also the core of future development in the transportation field.Among them,wireless communication in the Internet of Vehicle(Io V)can improve the safety and comfort of driving,and can improve the quality of service for users,thus becoming a key technology of ITS.On the one hand,in order to ensure communication quality in the Io V scenario,it is necessary to address the issues of channel fading and noise interference at the receiver.Currently,common channel estimation and signal detection methods have certain guarantees for improving communication quality,but the modular processing methods of traditional communication systems cannot ensure end-to-end optimization.On the other hand,due to the current channel state information(CSI)loss of real-time when the vehicle is moving at high speeds,it is necessary to predict CSI at future times.In this thesis,channel estimation and channel prediction algorithms in Io V communication have been studied,and the main contents are as follows:(1)This thesis characterizes the theoretical basis of channel estimation and prediction in wireless communication.This thesis introduces large scale fading and small scale fading,and presents two channel models based on the types of fading,including analysis and characterization.Then,it introduces pilot based estimation algorithms in wireless communication: Least Square estimation,Linear Minimum Mean Square Error estimation and interpolation algorithms,as well as channel estimation algorithms based on deep learning,and analyzes the shortcomings of traditional channel estimation algorithms.Finally,we introduce the auto regression(AR)prediction algorithm and the neural network based channel prediction algorithm.(2)This thesis proposes an estimator based on end-to-end learning.Unlike traditional communication systems,the proposed method combines channel estimation and signal detection,eliminating the need for modular data processing.Specifically,this article designs a deep neural network based on the generative adversarial network and residual network,which can achieve end-to-end recovery of received signals.In experiments,the network was applied to additive white Gaussian noise channels,multipath channels,dual vehicle communication channels,and Rayleigh fading channels,respectively.The TR38.901 channel model is selected in this thesis to generate mobile scene channel data.We use the toolkit in MATLAB,and set relevant parameters to generate channel data.The simulation results show that the channel estimation method based on end-to-end learning has higher accuracy,it is suitable for unknown channels and has robustness.(3)Considering the mobility in Io V scenario,this thesis proposes a channel prediction method based on attention mechanism and long short-term memory(LSTM).The existing channel prediction methods have the problems of falling into local optimization and fixed network model.Therefore,this thesis combines the attention mechanism and neural network to calculate the correlation between data,and can achieve global optimization.The simulation results show that the channel prediction method based on the attention mechanism outperforms the AR and LSTM algorithms.