Deep Learning-Based GNSS Jamming Classification Using Simulation-Generated Data and Time-Frequency Analysis

GNSS services are essential for modern communication, navigation, and surveillance systems. However, they are highly susceptible to jamming attacks, which can significantly reduce positioning accuracy. Jamming signals originate from various sources, making the classification of GNSS jamming events crucial for identifying interference sources. In recent years, machine learning has outperformed traditional methods in solving this classification problem. However, collecting realworld jamming datasets for training remains a challenge. To address this, this paper investigates the use of simulationgenerated data to train deep learning models for classifying jamming types in GNSS signals. To evaluate the generation model, we utilize the Short-Time Fourier Transform (STFT) for feature extraction, leveraging its time-frequency representation to effectively capture jamming characteristics. Our findings show that models trained on simulated data generalize well to realworld jamming scenarios, achieving high classification accuracy.