A Spatiotemporal Transformer Framework for Robust Threat Detection in 6G Networks

Abstract

6G networks provide high data rates, low latency, and massive connectivity but face security challenges due to the integration of communication, sensing, and AI. Traditional threat detection systems struggle to handle the complex interactions between dynamic network topologies and high-speed data flows in 6G environments. To address this, we propose a Spatiotemporal Dual-Stream Transformer framework that utilizes parallel graph-based and sequence-based attention mechanisms for real-time detection of threats such as cross-domain lateral attacks, large-scale DDoS, and sensor exploitation. Experimental results in a simulated 6G environment show an anomaly detection accuracy of 93.6% and an end-to-end inference latency of only 8.2 ms, while prototype testing achieves a 92.4% detection rate for 0-day exploits. These results establish a technical foundation and provide critical insights for the evolution of intelligent, secure 6G networks.