Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

One-Class Classification as GLRT for Jamming Detection in Private 5G Networks

Matteo Varotto, Stefan Valentin, Francesco Ardizzon, Samuele Marzotto, Stefano Tomasin

TL;DR

This work addresses jamming detection in private 5G networks by framing watchdog IQ monitoring as a one-class problem and approximating the GLRT with a CNN trained on legitimate data and synthetically generated attack samples. The detector combines a two-class CNN with a uniformly distributed artificial dataset $\mathcal{D}^*_1$ to emulate jammer statistics, supported by theoretical grounding that links this training to GLRT performance. Across multiple jamming scenarios and time-window sizes, the CNN with synthetic data significantly outperforms a convolutional autoencoder baseline, achieving larger separation between false-alarm and misdetection curves and demonstrating practical gains for physical-layer security in industrial private networks. The results suggest a robust, attacker-independent approach for wireless intrusion prevention systems in private 5G deployments, with potential for broader adoption in secure communications.

Abstract

5G mobile networks are vulnerable to jamming attacks that may jeopardize valuable applications such as industry automation. In this paper, we propose to analyze radio signals with a dedicated device to detect jamming attacks. We pursue a learning approach, with the detector being a CNN implementing a GLRT. To this end, the CNN is trained as a two-class classifier using two datasets: one of real legitimate signals and another generated artificially so that the resulting classifier implements the GLRT. The artificial dataset is generated mimicking different types of jamming signals. We evaluate the performance of this detector using experimental data obtained from a private 5G network and several jamming signals, showing the technique's effectiveness in detecting the attacks.

One-Class Classification as GLRT for Jamming Detection in Private 5G Networks

TL;DR

This work addresses jamming detection in private 5G networks by framing watchdog IQ monitoring as a one-class problem and approximating the GLRT with a CNN trained on legitimate data and synthetically generated attack samples. The detector combines a two-class CNN with a uniformly distributed artificial dataset to emulate jammer statistics, supported by theoretical grounding that links this training to GLRT performance. Across multiple jamming scenarios and time-window sizes, the CNN with synthetic data significantly outperforms a convolutional autoencoder baseline, achieving larger separation between false-alarm and misdetection curves and demonstrating practical gains for physical-layer security in industrial private networks. The results suggest a robust, attacker-independent approach for wireless intrusion prevention systems in private 5G deployments, with potential for broader adoption in secure communications.

Abstract

5G mobile networks are vulnerable to jamming attacks that may jeopardize valuable applications such as industry automation. In this paper, we propose to analyze radio signals with a dedicated device to detect jamming attacks. We pursue a learning approach, with the detector being a CNN implementing a GLRT. To this end, the CNN is trained as a two-class classifier using two datasets: one of real legitimate signals and another generated artificially so that the resulting classifier implements the GLRT. The artificial dataset is generated mimicking different types of jamming signals. We evaluate the performance of this detector using experimental data obtained from a private 5G network and several jamming signals, showing the technique's effectiveness in detecting the attacks.
Paper Structure (15 sections, 1 theorem, 7 equations, 6 figures, 3 tables)

This paper contains 15 sections, 1 theorem, 7 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. System Model
  3. Security Scenario
  4. Detector Design via Machine Learning
  5. Data Collection and Augmentation
  6. Laboratory Setup
  7. Dataset Creation
  8. Jamming Detection Models
  9. Convolutional Neural Network (CNN)
  10. Convolutional Autoencoder (CAE)
  11. Numerical Results
  12. Training Process
  13. Performance Metrics
  14. Performance Results
  15. Conclusions

Key Result

Theorem 1

glrt_st[Th. 1] A nn trained with a mse loss function over the two class dataset $\mathcal{D} = \{\mathcal{D}_0, \mathcal{D}^\star_1\}$, obtain one-class classifiers equivalent to the glrt, when a) the training converges to the configuration minimizing the loss functions of the two models, and b) the

Figures (6)

  • Figure 1: Considered security scenario: blue arrows indicate legitimate cellular communications and red arrows indicate the jamming signals.
  • Figure 2: Example of two iq bitmaps.
  • Figure 3: Two types of virtual dataset for training the dl models.
  • Figure 4: Performance comparison in terms of accuracy between the two models in the case: n$=256$, noise: uniform.
  • Figure 5: Performance comparison in terms of accuracy between the two models in the case n$=256$, noise: uniform over a frame. Lines and colors are those of Fig. \ref{['comparison_uniform']}.
  • ...and 1 more figures

Theorems & Definitions (1)

  • Theorem 1