Robust Generative-Augmented DRL for Multi-Beam Jamming of Drone Swarms

The increasing deployment of unmanned aerial vehicle (UAV) swarms introduces critical security risks through adversarial surveillance and attacks on infrastructure. Traditional counter-UAV (C-UAV) systems relying on kinetic interceptors or omnidirectional electronic warfare face limitations in scalability, precision, and energy efficiency. While phased-array beamforming enables targeted RF jamming, challenges persist in early detection, spectrum interference, and adaptive responses to evolving swarm tactics. To address these challenges, we propose a generative-augmented adaptive jamming framework grounded in deep reinforcement learning (DRL) and generative AI. Our system model incorporates drone mobility, multi-beam phased-array jamming, and a probabilistic jamming success metric based on the Jamming-to-Signal-plus-Noise Ratio (JSNR). We formulate the problem as a Markov Decision Process (MDP) optimized via Proximal Policy Optimization (PPO) and integrate three generative modules: a variational autoencoder (VAE) for scenario diversification, a diffusion model for predictive swarm tracking, and a GAN-based adversarial perturbation module. Simulations demonstrate that our method outperforms conventional DRL baselines, achieving 30% higher jamming accuracy, 2x lower latency, 80% less power consumption, and over 4× higher success rates in high-noise environments.