Graphic Neural Network based GPS Spoofing Detection for Cellular-Connected UAV swarm

Abstract

The cellular-connected Unmanned Aerial Vehicles (UAVs) are emerging as integral components of the 5G and beyond system due to their mobility and flexibility. Compared to a traditional single UAV, a flock of UAVs established as a UAV swarm can implement diverse distributed applications economically and efficiently, such as cooperatively smart agriculture, joint search and rescue, and supplementing temporary network connections. However, the GPS spoofing attack can manipulate UAV swarm locations and distort UAV swarm topology, which threatens the security of swarm communication and control. This paper proposes a Graphic Neural Networks (GNN) based GPS spoofing detection approach for cellular-connected UAV swarms. Especially, we propose a system in which the GNN model is used to detect GPS spoofing attacks by analyzing the similarity between the swarm GPS topology and communications topology. To evaluate the proposed neural networks, we use a bipartite graph and Hungarian algorithm to build a UAV swarm simulator. The results show that GNN can efficiently compute topologies’ similarity and detect GPS spoofing attacks. For instance, for a UAV swarm consisting of 10 UAVs, GNN detects the spoofing with accuracy over 90% and computation time of fewer than 10 milliseconds using Intel Core 1.6 GHz processor.