As wireless networks evolve toward cell-free massive MIMO architectures, ensuring secure and reliable communication becomes increasingly challenging. One critical threat is jamming attacks, which can disrupt communication in complex and dynamic network environments. This paper proposes a novel jamming detection framework based on dynamic graph modeling and graph neural networks (GNNs). The wireless network is represented as a sequence of time-evolving graphs, where nodes correspond to access points and user devices, and edges capture communication links based on metrics such as distance and signal quality. To detect jamming, the authors combine graph convolutional networks (GCNs) for spatial feature extraction with Transformer-based attention mechanisms to capture temporal patterns across graph sequences. This architecture enables the system to identify anomalies in network behavior caused by jamming activity.
The framework is evaluated using simulated scenarios with mobile users, varying channel conditions, and intermittent jamming. Results show high detection performance, with accuracy exceeding 99% in non-fading conditions and strong robustness even in more challenging fading environments. An important finding is the impact of channel conditions on detection: while stable channels improve reliability, fading introduces additional complexity that can mask jamming behavior. The study also highlights the importance of training strategies, showing that exposure to diverse jamming patterns improves generalization. Overall, this work demonstrates how AI-driven, graph-based approaches can enhance security in next-generation wireless systems, offering a promising direction for resilient and adaptive jamming detection in future 6G networks.