Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detecting 5G Narrowband Jammers with CNN, k-nearest Neighbors, and Support Vector Machines

Matteo Varotto, Florian Heinrichs, Timo Schuerg, Stefano Tomasin, Stefan Valentin

TL;DR

This work addresses the vulnerability of 5G NR to narrowband jammers that disrupt critical control subchannels. It proposes a WIPS that detects jamming at the physical layer using spectrogram representations of IQ samples and a pre-trained binary classifier, comparing CNN, PCA-assisted KNN, and PCA-assisted SVM on experimental data. CNN achieves perfect jam detection on the test set, while KNN and SVM achieve near-perfect accuracy with substantially faster inference times when using PCA for dimensionality reduction. The findings demonstrate that spectrogram-based ML can provide rapid, reliable jam-detection with potential deployment in an invisible watchdog, enabling timely mitigation in industrial 5G deployments.

Abstract

5G cellular networks are particularly vulnerable against narrowband jammers that target specific control sub-channels in the radio signal. One mitigation approach is to detect such jamming attacks with an online observation system, based on machine learning. We propose to detect jamming at the physical layer with a pre-trained machine learning model that performs binary classification. Based on data from an experimental 5G network, we study the performance of different classification models. A convolutional neural network will be compared to support vector machines and k-nearest neighbors, where the last two methods are combined with principal component analysis. The obtained results show substantial differences in terms of classification accuracy and computation time.

Detecting 5G Narrowband Jammers with CNN, k-nearest Neighbors, and Support Vector Machines

TL;DR

This work addresses the vulnerability of 5G NR to narrowband jammers that disrupt critical control subchannels. It proposes a WIPS that detects jamming at the physical layer using spectrogram representations of IQ samples and a pre-trained binary classifier, comparing CNN, PCA-assisted KNN, and PCA-assisted SVM on experimental data. CNN achieves perfect jam detection on the test set, while KNN and SVM achieve near-perfect accuracy with substantially faster inference times when using PCA for dimensionality reduction. The findings demonstrate that spectrogram-based ML can provide rapid, reliable jam-detection with potential deployment in an invisible watchdog, enabling timely mitigation in industrial 5G deployments.

Abstract

5G cellular networks are particularly vulnerable against narrowband jammers that target specific control sub-channels in the radio signal. One mitigation approach is to detect such jamming attacks with an online observation system, based on machine learning. We propose to detect jamming at the physical layer with a pre-trained machine learning model that performs binary classification. Based on data from an experimental 5G network, we study the performance of different classification models. A convolutional neural network will be compared to support vector machines and k-nearest neighbors, where the last two methods are combined with principal component analysis. The obtained results show substantial differences in terms of classification accuracy and computation time.
Paper Structure (18 sections, 6 equations, 8 figures, 3 tables)

This paper contains 18 sections, 6 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. System Assumptions
  3. Main assumptions
  4. Studied Jammer
  5. Experimental Setup
  6. Dataset Creation
  7. Cases and Labeling
  8. Machine Learning Models
  9. Convolutional Neural Network
  10. Principal Component Analysis
  11. $k$-Nearest Neighbors
  12. Support Vector Machines
  13. Classification Time
  14. Experimental Results
  15. Convolutional Neural Network
  16. ...and 3 more sections

Figures (8)

  • Figure 1: RF spectrogram measured over a bandwidth of $120$ MHz (x-axis) at center frequency $3750$ MHz for 20 s (y-axis), showing a 5G NR downlink signal of $100$ MHz bandwidth and a signal of an SSB jammer of $7.2$ MHz bandwidth. When the jammer is active, the 5G downlink transmission stops.
  • Figure 2: Experimental setup: (1) 5G UE and the SDR frontends for the (2) jammer, (3) watchdog, and (4) gNB. The PCs for (2) to (4) are not shown. The shown distances between the devices are for illustration purposes only.
  • Figure 3: Cumulated ratio of explained variance in % for a varying number of principal components. The $0.01$- and $0.99$-quantiles are based on 100 random datasets and marked in gray
  • Figure 4: FA and MD probabilities as a function of the threshold $\tau\in [0,1]$ (with $y$ axis values normalized to $1$) for the considered jammer
  • Figure 5: CDF of the classification time performed by the CNN with supervised learning
  • ...and 3 more figures