Research On Lightweight Scenario Identification Algorithm Based On Channel State Information In Vehicle-to-Vehicle Networks

In recent years,with the rapid development of the 5th Generation Mobile Communication Technology(5G),the Intelligent Transportation Systems(ITS)has gradually become a research hotspot in vehicle-to-vehicle networks.As a key technology applied in ITS,scenario identification plays an essential role in the field of channel modeling,resource allocation,intelligent driving and so on.This paper studies the design of lightweight scenario identification algorithm based on channel state information(CSI)in vehicle-to-vehicle networks.The main contents of this paper are as follows:Firstly,this paper introduces the background and theoretical knowledge of scenario identification in vehicle-to-vehicle networks.The scenario identification algorithm based on statistical features and machine learning in vehicle-to-vehicle networks is designed on the basis of previous studies,which extracts and select statistical features of CSI,and uses machine learning classifier for identification.Simulation results show that compared with the algorithm based on CSI module features proposed by previous studies,this algorithm achieves higher identification accuracy.Secondly,in order to reduce the time complexity of CSI feature extraction,the convolutional neural network is used to extract and select the statistical features of CSI.In addition,effective CNN and scenario identification algorithm based on lightweight residual convolutional neural network(Lite Res CNN)in vehicle-to-vehicle networks are designed by using depthwise separable convolution and residual module.Simulation results show that compared with the algorithm based on statistical features and machine learning,this algorithm achieves higher accuracy and has lower time complexity and spatial complexity.Besides,compared with the algorithm based on traditional CNN,the accuracy of this algorithm is basically the same,but the spatial complexity decreases significantly.Finally,in order to further improve the lightweight performance of scenario identification algorithm,the unstructured pruning algorithm based on absolute weight values and the structured pruning algorithm based on convolutional kernel clustering are applied to compress Lite Res CNN respectively.Besides,knowledge distillation is utilized to compensate for the precision loss during the pruning and improve the performance of the pruned model.The lightweight scenario identification algorithm based on model pruning and knowledge distillation(KDP-Lite Res CNN)in vehicle-to-vehicle networks is designed.Simulation results show that compared with structured pruning algorithm,unstructured pruning algorithm has less impact on model performance.Also,compared with the algorithm based on Lite Res CNN,this algorithm based on KDP-Lite Res CNN achieves approximately the same accuracy but has lower spatial complexity.In conclusion,this paper designs a lightweight scenario identification algorithm based on KDPLite Res CNN in vehicle-to-vehicle networks,which has high recognition accuracy,fast computing speed and takes up fewer computing resources and memory.It can be deployed on edge devices with limited computing power and memory in vehicle-to-vehicle networks and realize fast and accurate scenario identification.