DISTRIBUTED GRAPH NEURAL NETWORK-BASED ANTI-JAMMING FOR SCALABLE AND ENERGY-CONSTRAINED IOT NETWORKS

Abstract

This paper presents a distributed graph neural network (GNN) framework for anti-jamming in large-scale IoT systems with dynamic interference and strict energy constraints. The proposed approach enables each node to make decisions using only local observations, eliminating reliance on centralized coordination while maintaining robustness against adaptive jammers. Extensive simulations show that the framework achieves rapid learning with approximately 2× faster convergence than conventional deep reinforcement learning methods. Under strong jamming, it improves link quality by up to 5 dB in SINR and increases spectral efficiency by 20%–35%. In addition, the energy- aware design reduces power consumption by up to 30%, leading to a network lifetime extension of around 25%. The proposed method also demonstrates strong scalability, sustaining stable performance across networks of up to 1000 nodes with significantly lower signaling overhead. These results indicate that distributed GNN- based learning is a promising solution for reliable and energy-efficient IoT communications in adversarial environments. Keywords: Distributed graph neural networks; Anti-jamming communications; Internet of Things; Energy-efficient networking; Scalable wireless networks.