Sensing-Aware Beamforming and Interpretable Analysis for Pilot-Spoofing Resilience in ISAC Systems

Abstract

Detection-based defenses against pilot spoofing in integrated sensing and communication (ISAC) systems leave a structural gap. An adaptive eavesdropper that tunes its power below the detection threshold corrupts the uplink channel estimate and leaks signal to itself, yet evades every detector commonly proposed in the literature. This paper proposes a sensing-aided subspace projection defense. The defense operates continuously on the received signal covariance, suppresses the dominant interferer without a detector trigger, and reuses the eigenvalue decomposition already performed by the ISAC sensing subsystem. Evaluated against an adaptive attacker built from a reparametrized conditional generative adversarial network (cGAN), the defense recovers 8.1 dB of SINR at the legitimate user and attenuates information leakage toward the eavesdropper by 7 to 12 dB inside the stealth zone where conventional detectors fail. A triangulated interpretable analysis across three independent models yields design insights that position array sizing and sensing subsystem design as a joint optimization, and reveals a regime-dependent, non-monotonic dependence of post-defense SINR on array size under AoA-aware null-steering defenses with finite sensing precision.