Edge Intelligence Enabled Soft Decentralized Authentication in UAV Swarm

Abstract

With the increased deployment of the Unmanned Aerial Vehicles (UAVs) in both military and civilian fields, the authentication of the UAV surveillance and controlling data becomes critical due to the severe consequences of any forged data. With the highly dynamic operation environment, a flying UAV network may not be supported by the infrastructure network on the ground for security provision. Hence, it is vital to improving network security by utilizing on-site resources within a flying UAV swarm. In this paper, we utilize the physical-layer fingerprints to increase the difficulty for the attackers to impersonate the legitimate UAVs. A decentralized authentication scheme is proposed to avoid the single-point failure at the cluster head (CH) caused by the imperfect estimations. To mitigate the high computational cost of the decentralized authentication and to further improving the authentication accuracy, a situational-aware authentication customization algorithm is proposed at each UAV to compute the reliability of different attributes. Only the UAV with reliable attributes observations will contribute to the decentralized authentication process. Moreover, a soft authentication decision algorithm, which is compatible with customized regression models at each UAV, is proposed to further improve the system robustness. Hence, the proposed authentication algorithm can be customized at the system level and node level to maximize the overall authentication accuracy under a minimal extra computational cost based on the decentralized process. The simulation results demonstrate that our proposed scheme significantly increased the accuracy by comparing to the other state-of-the-art machine learning-aided physical-layer authentication schemes.