Research On Channel Characteristics Of High-speed Railway And Channel Prediction Based On Artificial Intelligence

Due to the sustained development of Chinese high-speed railways,the railway system is facing the severe challenge of greater passenger flow and more number of trains at the same time.Furthermore,the passenger’s demand for rail travel is no longer that just point-to-point transportation,and the needs for working and entertainment of passengers during the journey must also be taken into consideration.For these reasons,the Globle System for Mobile Communications-Railway(GSM-R)system based on the second-generation mobile communication system used in the current high-speed railway system gradually appears to be inadequate.Thus the research of a new generation of high-speed railway communication system is imperative.As the basis of wireless communication system design,accurate wireless channel modeling can provide important guidance for technical evaluation,system simulation,and network coverage planning during system design.This article first conducts a survey on the current research of high-speed railway channel modeling based on the measurement,and then based on the current research progress,relying on the measurement data from high-speed railway environment,the typical channel parameters are extracted and modeled.In the process of channel parameter extraction and channel modeling,it is found that the traditional channel modeling method based on measurement data have to face more and more problems such as the increase of the center carrier frequency and bandwidth,the higher speed of receiver and sometimes measurements are difficult to carry out.In order to avoid the difficults caused by channel measurement.Therefore,this article explores some AI(Artifical Intelligence)-based channel prediction methods.There are two methods proposed in this paper which designed to use scant measurement data to obtain more unmeasured data.The research points of this article are as follows:(1)Based on measurement data,the relationships of channel parameters with the distance between the receiver and the base station are explored;the different statistical modeling methods for some key channel parameters are compared and the cross-correlation between different channel parameters are compared.(2)To solve the problem that the selection of test points in the channel measurement process is not completely consequent.This artical proposed using the neural network to learn the characteristics of the measured data,and using the learned network to generate more channel data in the interval of measurement points to solve the problem of insufficient measurement data.This paper analyzed and compared the prediction results of different channel parameters with different types of neural networks,different numbers of neurons in hidden layer and different ratios of training data set.At last the most suitable neural network for this prediction method was selected.(3)Channel measurement often requires a lot of manpower and material resources.When the range that needs to be measureded is reduced,the cost of measurement will also decrease.Therefore,this paper proposes a time series prediction method based on Long Short-Term Memory(LSTM)network,and improves the traditional prediction method according to the characteristics of the predicted channel parameters.After that,the channel was predicted by the improved method with the measured data,and the prediction effect was analyzed.