Research On Building Of High-Speed Railway Broadband Channel Database And Data Mining

With the rapid development of high-speed railway techniques,high-speed railway technological system is gradually becoming mature.High-speed railway board-band wireless mobile communication system is currently playing an indispensable role in railway service to fulfill the demand of passengers.The propagation characteristics and states of wireless channel is vital in terms of quality of wireless communication system.In order to fulfill the requirement of service of the fifth-generation system,it is of great importance for system to perform a method to recognize the propagation scene precisely and the performance will be improved by using some adaptive techniques.And there exist relevance between different propagation scenarios,it is of great essence to reveal the underlying channel characteristics and analyze time-variant channel of high-speed railway.Nowadays,current researches mainly focus on LOS/NLOS identification,correlation of large-scale parameters and small-fading.However,few studies referring to propagation scenes recognition and analysis on channel relevance in high-speed railway propagation scenes.As stated above,we propose a novel propagation scene recognition model and investigate the relevance analysis of propagation scenes for high speed railway channels.Firstly,we divided and select the typical scenes then extract time-frequency-space dispersion features from measured data that obtained from channel sounding of Beijing-Tianjin intercity railway.Then board-band wireless channel datasets are created,and split into two sets: training set and testing set,each feature matrix corresponds to a scene.After that,we introduce several commonly used machine learning algorithms and feature fusion schemes then propose a weighted score fusion based LSTM network which takes characteristics of wireless channel into consideration.After training process,multiple evaluation metrics including accuracy,confusion matrix,f-score,area under ROC curves are employed.The results show that the proposed model outperforms than other fusion schemes and commonly used machine learning algorithms,provides with an accurate and efficient recognition application for high-speed railway communication systems.Furthermore,we propose a method to evaluate the relevance between different scenes.We employ high-dimensional Wasserstein distance that is widely used to measure the distance between two high-dimensional distribution functions,as an indicator that reflects the numerical and macro relevance between two scenes' channel features.Likewise,correlation matrix collinearity parameter that used to measure th e linear correlation between two matrices,is regard as the micro perspective of relevance of different scenes.From the results we find that the multi-link scenario and suburban scenarios are the most relevant from two perspectives.It is of great importance for people to make research on time-variant channel and communication system design from the relevance analysis.