Edge Intelligence Enabled Distributed and Collaborative Authentication in UAV Swarms
Abstract
Unmanned Aerial Vehicles (UAVs) have been widely deployed in various fields with many benefits such as cost reduction, safety improvement and service coverage enhancement. Unlike the other mobile ad hoc networks, the UAV swarm, which is a flying ad hoc network, may operate in a hostile environment or experience rapid network topology change which brings high vulnerability by using cloud-based centralized security provisioning techniques. Hence, securing the UAV networks with the on-site authentication resources becomes a vital aspect to accomplish the mission. The on-site authentication resources, such as the cross-layer attributes, can be utilized to form a unique characteristic of each UAV. Alternatively, decentralized authentication techniques have also been considered where multiple collaborative nodes are utilized to fuse a final authentication decision. Although the decentralized authentication techniques usually have a better security performance, they may increase the computational overhead and decrease the efficiency. Hence, limiting the computational overhead becomes a critical challenge when designing more sophisticated authentication schemes for UAV swarms.
In this thesis, a linear discriminant analysis-based centralized authentication mechanism is first proposed to enhance the security performance with limited computational overhead by eliminating the non-informative attributes. Then, to compensate for the single-point failure of the centralized authentication schemes, a collaborative authentication mechanism is proposed to enhance the performance by utilizing the soft edge authentication decisions. Ultimately, we define a novel concept of Security-of-Service (SoS) which is further utilized to minimize the complexity of the collaborative authentication. Instead of utilizing all authentication resources to reach a maximized security performance which creates a higher overhead, the SoS aims to only promise the exact authentication requirement by utilizing a minimum amount of authentication resources. The simulation results demonstrate that our proposed scheme is robust across the changing environment and can fulfill the SoS with limited authentication resources.