Reinforcement Learning Based PHY Authentication for VANETs

Mobile edge computing in vehicular ad hoc networks (VANETs) suffers from rogue edge attacks due to the vehicle mobility and the network scale. In this paper, we present a physical authentication scheme to resist rogue edge attackers whose goal is to send spoofing signals to attack VANETs. This authentication scheme exploits the channel states of the shared ambient radio signals of the mobile device and its serving edge such as the onboard unit during the same moving trace and applies reinforcement learning to select the authentication modes and parameters. By applying transfer learning to save the learning time and applies deep learning to further improve the authentication performance, this scheme enables mobile devices in VANETs to optimize their authentication modes and parameters without being aware of the VANET channel model, the packet generation model, and the spoofing model. We provide the convergence bound such as the mobile device utility, evaluate the computational complexity of the physical authentication scheme, and verify the analysis results via simulations. Simulation and experimental results show that this scheme improves the authentication accuracy with reduced energy consumption against rogue edge attacks.