AI and the World's Hard Problems · Theoretical computer science · arXiv · 2026-10-21
Generative Diffusion Models for Secure UAV-ISAC with Real-Time Target Tracking
The work addresses surveillance missions in which a dual-function drone with integrated sensing and communication tracks a mobile target, where an eavesdropper can learn in real time whether the target is detected inside a protected region by observing changes in event-triggered communication traffic. It formulates the threat as traffic-composition covertness and proposes a keyless diffusion-based transmission scheme that superposes a message-agnostic safeguard waveform from a conditional denoising diffusion implicit model, trained so that triggered and non-triggered slots become statistically indistinguishable. Numerical results show that with multiple transmit antennas and an eavesdropper SNR of 5 dB, a learned eavesdropper stays near the blind guess of 0.5, a detector knowing the safeguard model reaches only 0.71 to 0.73 inside Bob's main lobe, and the average position PCRB for tracking varies only between 0.59 and 0.70 m².