This work introduces a two‑pronged framework to improve the robustness of zero‑day jamming evaluation and detection. An online detection system is built around a graph attention network (GAT) that captures global temporal‑spectral structures, coupled with Dirichlet‑process means (DP‑means) clustering to jointly classify known jamming types and discover unseen zero‑day strategies within a single learning objective. In parallel, an inference‑driven reinforcement‑learning (RL) jammer is proposed as an adversarial benchmark. Treating the target receiver as a black box, the jammer infers the detector’s state via hypothesis testing and optimizes the trade‑off between attack impact and stealth. Simulations show the RL jammer achieves 33% higher attack efficacy and 67% greater stealth than prior baselines, while the proposed detection framework improves detection accuracy by roughly 20% over existing methods.
Review