International Journal On Cyber Situational Awareness (IJCSA)
ISSN: (Print) 2057-2182 ISSN: (Online) 2633-495X
DOI: 10.22619/IJCSA
Published Semi-annually. Est. 2014
Editor-in-Chief:
Dr Cyril Onwubiko, Director – Artificial Intelligence, Blockchain & Cyber Security, Research Series, London, UK; IEEE DVP
Associate Editors:
Professor Frank Wang, Professor of Future Computing, Chair IEEE Computer Society, UK&RI, School of Computing, University of Kent, Canterbury, UK
Professor Karen Renaud, Professor of Cyber Security, University of Strathclyde, Glasgow, Scotland, UK
GPS Jamming Signal Classification with CNN Feature Extraction in low Signal-to-Noise Environments
Carolyn J. Swinney and John C. Woods
Abstract:
The Global Positioning System (GPS) is a satellite constellation which gives users access to position, navigation and timing services. Many industries not only benefit from this but are reliant on it. Although illegal, GPS jamming devices have the power to cause major disruption to many services including financial, power distribution and communication systems. Recent testing assesses Global Navigation Satellite System (GNSS) jammers as being very dangerous to aircraft and Unmanned Aerial Vehicles (UAVs) especially those flying at low height. GNSS is also critical to the safe operation of Connected and Autonomous Vehicles (CAV) such as driverless cars. Timely detection of an attack is deemed to be enough to ensure the safety of the vehicle. Detection and classification of GNSS jamming signals is necessary to enable this. This paper considers feature extraction using a Convolutional Neural Network (CNN) when representing the signal as a graphical image. The JamDetect dataset is produced containing 6 different types of commercial jamming signals. Features are extracted using a CNN before a machine learning classifier is trained for classification. Results show that representing the signal in the graphical form of Power Spectral Density (PSD) is the least susceptible to noise. CNN feature extraction with machine learning classifier Logistic Regression using PSD produces 82.7% (+/-0.7%) at -20dB SNR and 100% accuracy at -10dB SNR. The results using PSD graphical signal representation are significant for when it is necessary to detect and classify GPS jamming signals in low SNR environments.
Keywords: Convolutional Neural Network; Deep Learning; GNSS Jamming; Machine Learning; Classification; RF Signal Analysis; Transfer Learning; Feature Extraction; JamDetect.
ISSN: (Print) 2057-2182 (Online) 2633-495X
Volume 6. No. 1
DOI: 10.22619/IJCSA.2021.100135
Date Published: Feb. 2022
Reference to this paper should be made as follows: Swinney, Carolyn J., Woods, John C. (2021). GPS Jamming Signal Classification using CNN Feature Extraction in low Signal to Noise Environments. International Journal on Cyber Situational Awareness, Vol. 6, No. 1, pp1-21.