Research On V2I Resource Allocation Algorithm For UUDN Vehicle-road Cooperative Perception Scene

Vehicle-road collaborative perception is a typical application scenario of 5G communication.Through the sharing of perception and dynamic map information between vehicle and roadside facilities,many applications such as queue driving,coordinated lane change,and intersection management can be realized.It not only provides strong support for fully autonomous driving of L5,but also enables the construction of smart cities.However,compared to the traditional 5G communication network,the Internet of Vehicles uses high-speed moving cars as the main body of communication.Unique factors such as unstable channel status and frequent switching make some traditional communication methods not suitable for the vehicle communication environment.In addition,vehicle awareness applications also pose great challenges to vehicle communication networks due to the requirements of throughput,time delay and reliability of communication.In order to solve the above problems,it is planned to improve the communication performance of vehicle-to-infrastructure communication(V2I)in the scenario of vehicle-to-infrastructure communication from two levels of network architecture selection and resource allocation.In terms of architecture selection,the UUDN architecture based on 5G ultra-dense networking is adopted to change the cell-based user access method,so that multiple base stations can provide communication services for a single user,alleviate the performance degradation caused by handover,and improve the service quality of vehicle users.In terms of resource allocation,considering the two stages of dynamic map information delivery by the roadside server and vehicle perception information uploading,centralized resource allocation schemes based on vehicle location and distributed resource allocation schemes based on perception are designed respectively for different communication requirements.The specific innovation work is summarized as follows:The downlink V2 I transmission in the vehicle-road cooperative sensing scenario is mainly for large-capacity dynamic map information.The characteristics of this type of information are large data packet capacity,random data packet generation,and periodic delivery once generated.Taking into account the instability of the vehicle communication network and the characteristics of data packets,this paper proposes a resource allocation algorithm based on deep reinforcement learning in the UUDN architecture.UUDN provides diversity gain and reduces performance degradation caused by switching.The central processor collects vehicles Information and select the transmission time slot and transmission power for the V2 I link in a centralized form.In addition,considering the unfavorable vehicle communication environment and the occasional objective situation of link congestion,the advantages of dynamic resource allocation of reinforcement learning are used to ensure the timeliness of the map information containing vehicle control information in a long period of time through the information age.In the process of resource allocation,the resources are tilted to the link that has not received data packets for a long time.By setting a reasonable reward function,the algorithm can achieve convergence under the specified number of training times.The simulation results show that compared with other algorithms under similar scene settings,it can guarantee V2 I link throughput,reliability,information age performance optimization and power consumption optimization.Uplink V2 I transmission in the collaborative sensing scenario is the prerequisite for generating dynamic map information.The vehicle needs to upload the sensing information of the on-board sensor to the roadside through the V2 I uplink for processing by the central processing unit.This type of information has a small amount of data but strict latency constraints.In order to ensure the efficient transmission of such information,this paper adopts a distributed architecture and proposes a resource allocation algorithm based on multi-agent reinforcement learning.Among them,each vehicle as an agent makes the decision of V2 I uplink frequency band allocation in a distributed manner by observing the local environment.The simulation results show that the proposed algorithm has higher Qo S performance by comparing other resource allocation schemes.