Anomaly Detection in 5G Networks Using Transformer-Based Autoencoder

This paper introduces a novel anomaly detection approach for 5G networks using Transformer-Based Autoencoders (TAE) to address the complex security challenges intrinsic to these environments. As 5G technology becomes increasingly widespread, its associated security vulnerabilities also escalate, particularly when applied to critical energy systems. Marine Corps Air Station (MCAS), Miramar, is one such case study, in which 5G non-standalone (NSA) cellular communications is used to connect disparate energy devices across the facility. In order to enhance the cyber resiliency of such an energy network, we must implement effective anomaly detection methods using machine learning. The 5G NSA setup has limitations compared to the 5G standalone architecture, particularly in its ability to detect cyber anomalies in energy traffic. This study employs a transformer architecture that enhances the anomaly detection capabilities of traditional autoencoders by more effectively capturing complex data interrelationships. Extensive experiments on a simulated 5G dataset validate the efficacy of our approach, which achieves high detection accuracy without the need for outlier removal, thereby effectively identifying and mitigating a variety of security threats. The model's ability to maintain stability without extensive preprocessing underscores its potential for enhancing the cyber resilience of 5G networks, specifically as it pertain to the MCAS Miramar use case.