Multiagent Federated Deep-Reinforcement-Learning-Enabled Resource Allocation for an Air–Ground-Integrated Internet of Vehicles Network

Abstract

As an emerging architecture for the future 6G Internet of Vehicles (IoV), the air–ground-integration network has become a paradigm to achieve reliable interconnection everywhere. Unmanned aerial vehicles (UAVs), as low-altitude aerial platforms, can cooperate with ground infrastructures with the advantages of high flexibility and low cost. However, wireless resource allocation for vehicle-to-UAV (V2U) communications has encountered various challenges, such as air–ground spectrum sharing, dynamic topology, fast-changing channels, and time-sensitive services. In this article, we propose a multiagent federated learning and dueling double-deep Q-network (D3QN)-based resource allocation, namely, Fed-D3QN, to jointly optimize channel selection and power control to meet the low latency and reliability requirements of IoV services. Simulation results demonstrate that the proposed Fed-D3QN algorithm has good stability in the highly dynamic air–ground integration network. Additionally, it reduces the total delay of vehicle-to-infrastructure links and improves the transmission success rate of V2U links.