Communication Dans Un Congrès Année : 2025

SHIELD: Scalable and Holistic Evaluation Framework for ML-Based 5G Jamming Detection

Résumé

Jamming remains a significant threat to the reliability and security of 5G networks, despite extensive investigation in the existing literature. This work addresses the scalability and robustness gaps found in previous approaches, introducing SHIELD-a scalable and holistic framework designed to evaluate jamming interference and support machine learning-based detection techniques without relying on costly external hardware. To validate our approach, we develop a realistic 5G testbed including a power-modulated jammer positioned between commercial off-the-shelf Android devices and an SDR-based radio access network. Our experimental results demonstrate that this jamming setup generates complex interference patterns that challenge detection methods proposed in prior work. We then propose a novel jamming detection methodology that, by synchronously collecting native logs from both the User Equipment (UE) and the Next-Generation Node B (gNB), captures a comprehensive view of network behavior in both normal and jammed states. SHIELD overcomes the shortcomings of existing detection methods-which typically fail under subtle, long-term interference-by employing a robust preprocessing pipeline that extracts multi-layer features through interpolation and sliding-window aggregation. We assess several lightweight yet accurate classifiers, including SVM, KNN, Gradient Boosting, and Random Forest, to determine detection performance across diverse real-world scenarios. Our evaluation shows that while current methods can achieve high accuracy-often exceeding 90%-in controlled scenarios, their performance can drop below 70% when exposed to varying conditions. In contrast, our proposed log-based framework maintains accuracy levels around 94% on unseen data, offering a scalable, cost-effective, and robust approach for large-scale 5G deployments.

Fichier principal
Vignette du fichier
_LNCS__ARES__SHIELD__Proceeding_.pdf (4.05 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05055784 , version 1 (05-05-2025)
hal-05055784 , version 2 (20-05-2025)

Licence

Identifiants

  • HAL Id : hal-05055784 , version 2

Citer

Jiali Xu, Aya Moheddine, Valéria Loscrì, Alessandro Brighente, Mauro Conti. SHIELD: Scalable and Holistic Evaluation Framework for ML-Based 5G Jamming Detection. 20th International Conference on Availability, Reliability and Security (ARES), Aug 2025, Ghent (BE), Belgium. ⟨hal-05055784v2⟩
492 Consultations
732 Téléchargements

Partager

  • More