Deep Learning-Based Secure Content Fetching in UAV-Aided and RIS-Assisted Under Coordinated Aerial and Ground Eavesdropping

This work addresses the problem of secure content fetching in UAV-aided and reconfigurable intelligent surface (RIS)-assisted multiple-input single-output (MISO) networks, with the presence of coordinated aerial and ground eavesdropping. We formulate the problem as a delay minimization task, which is nonconvex. To address this challenge, we decouple the problem into two interrelated subproblems: content placement, and joint optimization of active beamforming, phase shift, and UAV trajectory. For content placement, we efficiently handle the task by applying Karush-Kuhn-Tucker (KKT) conditions. For the optimization of active beamforming, phase shift, and UAV trajectory, we introduce a novel framework called BiL-GANDI, which combines a generative adversarial network (GAN) and a diffusion model. To the best of our knowledge, this is the first work to apply this combination in secure content fetching for UAV-aided and RIS-assisted networks. The GAN leverages its adversarial mechanism to detect and mitigate security vulnerabilities, while the diffusion model iteratively generates solutions from the input distribution, optimizing the active beamforming, phase shift, and UAV trajectory. Unlike existing deep learning-based approaches that require extensive datasets for training, our proposed frame-work effectively functions even in the absence of sufficient training data. Simulation results demonstrate the efficacy of our approach in reducing content fetching delays while ensuring security against coordinated eavesdropping attacks. Moreover, the proposed BiL-GANDI surpasses well-known methods in the literature.