GNN-Based Resource Allocation for Digital Twin-Enhanced Multi-UAV Radar Networks

Abstract

Mutual interference has been a critical issue in multiple unmanned aerial vehicles (multi-UAV) networks. As an advanced technology, digital twin (DT) maps physical entities into virtual domain, enables real-time monitoring and dynamic updates, thereby enhancing the adaptability and performance of multi-UAV networks. In this letter, we investigate joint spectrum allocation and power control for a multi-UAV radar sensing network, where multiple unmanned aerial vehicles (UAVs) simultaneously perform radar sensing separately to detect targets and avoid collision. By modeling the multi-UAV network as a graph, we employ graph neural network (GNN) to capture environmental features, construct the DT network, and address resource allocation issues. In particular, we propose a message-passing neural network based spectrum allocation method and a graph attention network based power control method to maximizing the minimum radar echo signal-to-interference-plus-noise ratio (SINR) among all UAVs. Simulation results show that the proposed DT-enhanced GNN based resource allocation method can significantly improve the minimum SINR and extend the sensing coverage.