Machine Learning Based Strategy to Detect Meaconing attacks in GNSS : Method And Analysis Using FGI Dataset

Abstract

Global navigation satellite systems (GNSS) signals, which are used to guide modern navigation systems, can be subtly manipulated while preserving their authenticity. This thesis explores the use of machine learning (ML) for detecting meaconing attacks in GNSS to overcome fundamental vulnerabilities in navigation systems dependent on precise positioning. After conducting a review of GNSS signal processing, spoofing techniques and current methods of detection, this research examines the behaviour of the receiver level tracking loops during meaconing. The study focuses on the carrier-to-noise density ratio (C/N₀), Doppler frequency, delay-lock loop (DLL) discriminator, code phase, multi-correlator distortions. The study uses raw in-phase and quadrature (I/Q) tracking information collected from Finnish geospatial institute (FGI) and processed with the FGI-GSRx software defined receiver on MATLAB, to detect different patterns in these features during meaconing attacks. Sliding window segmentation is applied to capture temporal dynamics. ML models, including random forest (RF) and support vector machines (SVM) were trained to distinguish between authentic and attacked signals. The proposed framework shows high detection performance across GPS and Galileo constellations, with results indicating strong accuracy, low false alarm rates, and consistent performance under blocked chronological validation While individual features show varying sensitivity to meaconing, combining these features provides a strong discriminatory capability. Additionally, several analysis including feature correlation, dimensionality assessment, and satellite generalization were performed and confirmed robustness of the approach. The results suggest that tracking loop features combined with ML, offer a reliable and scalable solution for real time detection of meaconing attacks in GNSS receivers.