Countering Intelligent UAV Eavesdroppers via IRS-Assisted UAV: A Game-Theoretic Approach

This paper investigates an uplink secure communication system where an uncrewed aerial vehicle (UAV) equipped with an intelligent reflecting surface (IRS) assists transmissions between ground users (GUs) and a base station (BS). The system faces a critical challenge from a mobile UAV eavesdropper (UAE) that adaptively adjusts its trajectory to maximize eavesdropping. To counter this threat, the legitimate UAV is designed with a dual role: carrying the IRS to enhance legitimate links while simultaneously acting as an active jammer to disrupt the UAE. We formulate the problem as maximizing the secrecy energy efficiency (SEE) of the legitimate UAV through joint optimization of its trajectory, jamming power, GU transmit power and scheduling, and IRS phase shifts, while the UAE optimizes its trajectory to maximize its eavesdropping rate. This adversarial interaction is modeled as a non-cooperative game, for which we define and prove the existence of a Nash Equilibrium (NE). To solve the resulting high-dimensional, coupled optimization problem, we develop a multi-agent deep reinforcement learning (MADRL) framework where the UAV and UAE act as independent agents. Simulation results validate the convergence of the proposed scheme to the NE and demonstrate that the dual-role UAV design significantly improves secrecy performance by effectively suppressing eavesdropping.