Traditional methods for classifying Global Navigation Satellite System (GNSS) jamming signals typically require post-processing raw or spectral data, necessitating costly data transmission to cloud-based systems. In contrast, this study proposes an efficient approach that compresses GNSS data streams directly at the hardware receiver while simultaneously classifying jamming and spoofing attacks in real time. Given the growing prevalence of GNSS interference, there is a critical need for solutions suitable for power-constrained environments.
The paper introduces a novel method using generative artificial intelligence (GenAI), specifically variational autoencoders (VAEs), deployed on Google Edge tensor processing units (TPUs). The researchers evaluated various autoencoder (AE) architectures to compress and reconstruct GNSS signals, focusing on preserving interference characteristics while minimizing data size near the receiver hardware. To ensure energy-efficient deployment, the pipeline adapts large-scale AE models for Google Edge TPUs through 8-bit quantization.
Tests conducted on raw in-phase and quadrature-phase (IQ) data, Fast Fourier Transform (FFT) data, and handcrafted features demonstrate that the system achieves significant compression exceeding 42x. It maintains accurate classification of approximately 72 interference types on reconstructed signals with an F2-score of 0.915, closely matching the performance on original signals (F2-score 0.923). This hardware-centric GenAI approach substantially reduces jammer signal transmission costs, offering a practical solution for interference mitigation.
Furthermore, ablation studies on conditional and factorized VAEs (e.g., FactorVAE) explore latent feature disentanglement for data generation. These efforts enhance model interpretability and foster trust in machine learning (ML) solutions for sensitive interference applications. The study has been accepted for presentation at the IEEE/ION Position, Location and Navigation Symposium (PLANS) 2025.
