Wireless Channel Modeling Based On Machine Learning In The High-speed Railway Scenario

As a product that adapts to social progress and technological development,high-speed railway(HSR)has the characteristics of fast running speed,large passenger carrying capacity,low pollution and strong safety.The wireless communication system of the high-speed railway needs to carry the dedicated communication service of train control and the communication service of the passengers on the train.Therefore,in order to provide more reliable communication services,it is necessary to study the wireless communication system of the HSR.Research on wireless channels is a prerequisite for wireless communication system research.An accurate understanding of the propagation characteristics of wireless channels can provide a reliable basis for the design and optimization of wireless communication systems.Through wireless channel modeling,the channel characteristics can be modeled,and then provide a theoretical basis for the evaluation of key technologies of the wireless communication system,prototype construction and network deployment.HSR will experience different scenarios during its operation,and each scenario has different characteristics.At present,domestic and foreign scholars have made a lot of research with respect to HSR wireless channel modeling of different scenarios,but there are still some deficiencies.Therefore,in view of the problem of poor accuracy caused by overfitting in wireless channel modeling in HSR scenario,this dissertation uses machine learning algorithms to carry out wireless channel modeling research on several common HSR scenarios,and proposes an accurate modeling method of HSR wireless channel based on machine learning.The specific content and research results of this dissertation include:(1)First of all,in view of the problem of modeling error caused by not considering the difference of Rician K-factor in the process of HSR wireless channel modeling,based on the measured data of Rician K-factor in different HSR scenarios,the K-Means clustering algorithm in machine learning is used to cluster the Rician K-factor,and obtains its typical value in different distance ranges.Simulation platform can be used to obtain the results of the Rician K-factor clustering with clear boundaries,and the typical values of Rician Kfactor in different segments.Taking into account the influence of different Rician K-factor typical values on the model in the modeling process,the modeling error caused by not considering the difference of Rician K-factor can be avoided.(2)Secondly,aiming at the problem of poor generalization ability caused by overfitting phenomenon in HSR wireless channel modeling,based on the measured data of path loss in different HSR scenarios,using the theory of model evaluation and selection in machine learning,a precise modeling method of least squares regression fit based on crossvalidation is proposed to verify whether the model overfit and then obtain a more reliable channel.model.Path loss models,generalized error values and reliability validation results of HSR wireless channel in different scenarios can be obtained by simulation.The results verify that the algorithm proposed in this dissertation has a good modeling effect,avoids the phenomenon of overfitting or underfitting in the modeling process of HSR wireless channel,and can obtain a more reliable model.(3)Finally,in order to further analyze the effect of generalization error on ergodic capacity properties of HSR wireless channel,Nakagami-m fading is introduced to approximate Rician fading and Rayleigh fading.The formula of ergodic capacity is further simplified by combining Meijer G function.The mathematical relationship of the ergodic capacity depending on the Rician K-factor,the signal-noise ratio of the receiver,and the path loss generalization error are established after the ergodic capacity analysi s of the HSR wireless channel.The simulation platform can be used to obtain the lower traversal capacity curve of HSR wireless channel in different scenarios.The results show that the existence of generalization error will adversely affect the ergodic capacity of HSR wireless channel.