Principal Component Analysis-Based Optimal Feature Design for GNSS Spoofing Detection

Global navigation satellite system (GNSS) spoofing attacks pose a serious threat to existing 5G and future 6G networks, as they rely heavily on the high-precision time synchronization information provided by GNSS. Signal feature-based spoofing detection methods effectively identify spoofing signals using early and late correlators to extract signal features. However, using multiple features as machine learning (ML) inputs can lead to the curse of dimensionality. To address this, the study proposes an optimal feature construction method. This method uses principal component analysis (PCA) to generate essential features and a logistic regression (LR)-based approach to construct these generated features into one feature, i.e. principal component regression (PCR). The study then employs a one-dimensional convolutional neural network (1D CNN) to compare and evaluate the PCR method against conventional composite features. The results show that the PCR consistently outperforms conventional features, indicating higher spoofing detection effectiveness and the potential to safeguard critical communication infrastructures against spoofing attacks.