A Study of Attention-Driven Spiking Neural Networks in Enhancing GNSS Jamming Classification

Global Navigation Satellite Systems (GNSS) are crucial for modern infrastructure but are vulnerable to jamming due to their open structure, limited security, narrow bandwidth, and weakened signal strength over long distances. Jamming attacks exploit these vulnerabilities, disrupting both civilian and military operations by degrading or blocking positioning, navigation, and timing (PNT) services. Existing jamming detection methods rely on power-hungry Deep Learning (DL) models built with Artificial Neural Networks (ANNs) and pre-trained architectures, making them unsuitable for power- and resource-constrained edge devices used in GNSS systems and are prone to errors in noisy environments. To address these challenges, we propose leveraging Convolutional Spiking Neural Networks (CSNNs) for jamming classification, taking advantage of the energy-efficient, asynchronous, and sparse nature of SNNs. We integrate attention mechanisms and multi-axis data augmentation to enhance classification accuracy and mitigate overfitting. Our approach improves radio environment awareness in wireless communication systems by providing superior jamming classification accuracy and robust physical-layer security, particularly in noisy environments. Experimental results show that CSNN-based models achieve competitive accuracies of up to 98%, comparable to CNN-CBAM and ResNet, while maintaining accuracies as high as 97% under multipath noise with superior computational efficiency, where traditional models degrade to as low as 33%.