Particle swarm optimized deep learning for jamming detection and throughput enhancement in cognitive radio networks

Cognitive Radio Networks (CRNs) are essential for improving spectrum efficiency in 5G and beyond; however, their open and adaptive nature makes them highly vulnerable to jamming attacks. The purpose of this work is to develop an anti-jamming framework that jointly addresses jammer detection and frequency selection under adversarial conditions. The novelty of this work lies in jointly addressing strategic anti-jamming decision-making and data-driven jammer detection within a single unified framework. Motivated by the limitations of standalone game-theoretic, evolutionary, and deep learning approaches in dynamic adversarial environments, we propose a hybrid anti-jamming framework that integrates game theory, deep learning, and particle swarm optimization (PSO). Game theory is employed to model the strategic interaction between secondary users and jammers, enabling utility-aware frequency hopping (FH) decisions, while a PSO-driven deep neural network (DNN), termed DeepSwarm, is designed for accurate and robust jammer detection. The strength lies in leveraging PSO to enhance the convergence speed and robustness of the DNN in dynamic jamming environments. Simulation results demonstrate that DeepSwarm achieves 98.10% accuracy, 98.30% recall, 98.10% precision, and a 98.05% F1-score, outperforming SVM, linear regression, and stacking baselines. Furthermore, FH guided by the proposed detection framework improves channel utilization and increases normalized throughput by up to 32% compared to static selection under varying jamming probabilities. These findings confirm the scalability and effectiveness of the proposed framework for securing CRNs in adversarial environments.