Joint Feature Construction for Spoofing Detection Based on XGBoost and Logistic Regression

Global Navigation Satellite Systems (GNSS) have permeated various industries worldwide. However, the signals transmitted by satellites undergo path loss and obstruction, resulting in significantly weakened signals upon reaching the receiver. Furthermore, the open structure of satellite navigation signals renders them susceptible to spoofing interference. Among numerous deception detection methods, those based on signal features exhibit strong applicability, simple algorithms, and flexible structures. These methods extract signal features using a series of early and late correlators to identify spoofing signals. Nevertheless, the key signal features currently employed for deception detection rely on expert-designed manual processes, which are intricate and inefficient. Additionally, existing conventional features only cover specific aspects of spoofing signal information, thereby reducing their effectiveness. To address these challenges, this study proposes a machine learning-based joint feature construction method. Initially, the study employed Extreme Gradient Boosting (XGBoost) to rank the importance of numerous existing features, selecting high-quality features based on the ranking for subsequent joint feature input. Subsequently, the study used a logistic regression-based approach to combine the selected features, creating a more expressive feature set. The optimization of joint features crucially depends on selecting appropriate parameter vectors, enabling the joint features to integrate the advantages of various features in different aspects, thereby enhancing their expressiveness. Finally, the study employed a convolutional neural network (CNN) detection method to compare and evaluate the joint feature against conventional signal features. The results demonstrate that, under various parameter conditions, the CNN using joint features as input consistently outperforms that of conventional features, indicating a higher representative capability.