GNSS Spoofing Detection by Supervised Machine Learning with Validation on Real-World Meaconing and Spoofing Data-Part...

Area of Science:

• Navigation Systems
• Signal Processing
• Machine Learning

Background:

• Global Navigation Satellite System (GNSS) signals, particularly open service (OS) signals, are vulnerable to meaconing and spoofing attacks.
• These threats pose significant risks to Safety-of-Life (SoL) applications that depend on uncorrupted GNSS data.
• Existing detection methods often focus on pre- or post-correlation signal analysis.

Purpose of the Study:

• To investigate the efficacy of supervised machine learning algorithms for detecting GNSS meaconing and spoofing.
• To propose and evaluate the use of a specific algorithm, Classification Support Vector Machine (C-SVM), at the GNSS receiver level.
• To enhance detection capabilities by incorporating real-world attack datasets into the training process.

Main Methods:

• Utilizing Classification Support Vector Machine (C-SVM) at the GNSS receiver level to analyze correlations among measurements and observables.
• Training the C-SVM model with a combination of laboratory-generated and real-world spoofing and meaconing datasets.
• Conducting cross-validation experiments where datasets from one attack type (meaconing or spoofing) were used to validate the other during training.

Main Results:

• Enriching the C-SVM training dataset with real-world attack data significantly improves the detection of GNSS signal manipulation attempts.
• The C-SVM-based approach demonstrated promising results in identifying both meaconing and spoofing attacks.
• Comparative analysis across four experimental setups confirmed the benefits of dataset enrichment and the C-SVM's potential.

Conclusions:

• Supervised machine learning, specifically C-SVM, offers a viable and promising approach for detecting GNSS signal manipulation.
• Training C-SVM with diverse, real-world datasets is crucial for robust detection of sophisticated attacks.
• The C-SVM methodology shows potential for integration into federated learning frameworks for enhanced, distributed GNSS security.