Research On Data-Driven Based Wireless Channel Modeling

As a ubiquitous connectivity scenario,vehicle-to-everything(V2X)is an important infrastructure and technology path to achieve intelligent transportation.The development of V2 X technology can greatly increase the gross domestic product and strengthen our country’s core competitiveness in technological innovation.It can also make a significant contribution to building an energy-saving,efficient and safe intelligent transportation system in China.At present,cellular-V2X(C-V2X)technology is being promoted in China,and wireless channel is an important basis for the design and performance evaluation of any communication system.In order to accelerate the development of next-generation Internet of vehicles systems and the validation of prototypes,it is urgent to carry out research on C-V2 X channels.However,the current researches on C-V2 X channels still use the traditional modeling method,which lack the support of automatic learning rules and are difficult to cope with the changing trend of wireless channel data with massive,time-varying,and diverse features.To better understand the propagation characteristics of C-V2 X channels,an efficient modeling method with adaptive and self-learning capabilities is required,while data-driven models have this advantage.Therefore,this thesis studies the technical scheme and feasibility of applying data-driven model to C-V2 X channel,thus providing theoretical and technical support for applying data-driven model to solve the modeling problem of time-varying fading channel.A channel measurement system suitable for dynamic environment is designed in this thesis for the channel measurement of C-V2 X communications.The measurement system can support the channel measurement of multi-scene and multi-antenna for the Internet of vehicles below 40 GHz band and up to 500 MHz bandwidth.Then,the system based on the temporary C-V2 X standard is used to carry out the measurement of typical and complex scenes such as highways,urban roads,tunnels,and roundabouts.A large amount of data from actual measurements will provide data support for the modeling and research on characterization of C-V2 X channels.Aiming at the C-V2 X channels modeling,a data-driven wireless channel modeling scheme is designed in this thesis,which is used to predict the channel characteristics of highway and urban scenarios.In this thesis,the back propagation neural network,radial basis neural network,extreme learning machine and wavelet neural network are used as the core algorithms of the data-driven model,by which the path loss,root mean square delay spread and Doppler spread of the channels are predicted.The prediction performance of four neural networks is comprehensively evaluated and the feasibility of data-driven channel model is proved.Furthermore,the spatial correlation and spatial consistency of the two scenarios are studied,which is helpful for the design of useroriented beamforming and tracking techniques in vehicle-to-infrastructure(V2I)systems.Aiming at the intelligent perception of C-V2 X communication scenarios,a multiscene recognition algorithm based on machine learning is studied in this thesis.The backpropagation neural network,extreme learning machine,random forest,and support vector machine are taken as intelligent classification algorithms,and multiple channel features are combined to identify the highway,urban road,tunnel,and roundabout scenarios.Three evaluation indicators are used to comprehensively evaluate the performance of different classification models under different configurations.In this thesis,the recognition accuracy and robustness of the model are verified in multiple scenarios,thus providing a theoretical basis for the design of the intelligent scene recognition system for the Internet of Vehicles.