Protocol-Aware Jamming and Self-Play Defense for Active RIS-Assisted UAV Communications

Abstract

Active reconfigurable intelligent surfaces (RISs) mounted on unmanned aerial vehicles (UAVs) can restore blocked wireless links by combining aerial repositioning, controllable reflection, and element-level amplification. However, this amplification creates a dual-use vulnerability: a jammer whose signal reaches the active RIS may also be amplified and redirected toward the receiver. This paper investigates this threat in active RIS-assisted UAV communications under a protocol-aware mobile jammer that jointly adapts its trajectory, transmit power, beam direction, and attack mode, switching between pilot contamination (PC) and data-phase interference (DPI). PC corrupts channel acquisition and induces persistent RIS phase misconfiguration, whereas DPI creates an amplified RIS-mediated interference path during payload transmission. We analytically characterize the pilot-induced phase error and coherent-combining gain reduction, and define rate-loss and energy-normalized metrics to compare PC and DPI. We then formulate the defender–jammer interaction as a Markov game and develop a competitive multi-agent reinforcement learning defense based on centralized training and decentralized execution (CTDE) with adversarial self-play. Simulations show that fixedmodality defenses are brittle under attack-mode switching, while the proposed self-play defense remains robust against both fixed and adaptive jammers, achieving a worst-case rate $6.3 \times$ higher than the strongest fixed-modality baseline.