Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Sequence-to-Sequence Models for GNSS Spoofing Detection

Jan Zelinka, Oliver Kost, Marek Hrúz

TL;DR

GNSS signals are vulnerable to spoofing, and real-world spoofing data are scarce. The authors introduce a synthetic data generator to create diverse spoofing scenarios (targeted and regional) and train online sequence-to-sequence detectors using LSTM and Transformer-inspired architectures. Transformer-based models with early fusion achieve the best performance, reaching an error rate of 0.16% on a large synthetic test set, demonstrating feasibility of online spoofing detection with deep learning. The work provides a platform for rapid evaluation, though real-world validation is needed to confirm applicability.

Abstract

We present a data generation framework designed to simulate spoofing attacks and randomly place attack scenarios worldwide. We apply deep neural network-based models for spoofing detection, utilizing Long Short-Term Memory networks and Transformer-inspired architectures. These models are specifically designed for online detection and are trained using the generated dataset. Our results demonstrate that deep learning models can accurately distinguish spoofed signals from genuine ones, achieving high detection performance. The best results are achieved by Transformer-inspired architectures with early fusion of the inputs resulting in an error rate of 0.16%.

Deep Sequence-to-Sequence Models for GNSS Spoofing Detection

TL;DR

GNSS signals are vulnerable to spoofing, and real-world spoofing data are scarce. The authors introduce a synthetic data generator to create diverse spoofing scenarios (targeted and regional) and train online sequence-to-sequence detectors using LSTM and Transformer-inspired architectures. Transformer-based models with early fusion achieve the best performance, reaching an error rate of 0.16% on a large synthetic test set, demonstrating feasibility of online spoofing detection with deep learning. The work provides a platform for rapid evaluation, though real-world validation is needed to confirm applicability.

Abstract

We present a data generation framework designed to simulate spoofing attacks and randomly place attack scenarios worldwide. We apply deep neural network-based models for spoofing detection, utilizing Long Short-Term Memory networks and Transformer-inspired architectures. These models are specifically designed for online detection and are trained using the generated dataset. Our results demonstrate that deep learning models can accurately distinguish spoofed signals from genuine ones, achieving high detection performance. The best results are achieved by Transformer-inspired architectures with early fusion of the inputs resulting in an error rate of 0.16%.
Paper Structure (12 sections, 2 equations, 6 figures, 1 table)

This paper contains 12 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Nominal/Unspoofed PSR Signals
  4. Spoofing and Proposed Detection Concept
  5. Generator of Spoofed PSR Signals: Targeted and Regional
  6. Targeted spoofing
  7. Regional spoofing
  8. Input Signals Processing
  9. Approaches to Signal Fusion
  10. Sequence-to-Sequence Model
  11. Experiments and Results
  12. Conclusion

Figures (6)

  • Figure 1: Illustration of (un)spoofed GNSS signals.
  • Figure 2: Illustration of targeted spoofing trajectories. The red lines are examples of spoofing trajectories, and the blue dashed lines indicate the area where spoofed trajectories might occur.
  • Figure 3: Illustration of regional spoofing trajectories. The red lines are examples of spoofing trajectories, and the blue dashed lines indicate the area where spoofed trajectories might occur.
  • Figure 4: An example illustrating PSR and its differences.
  • Figure 5: Illustration of the proposed multi-headed attention-based module for satellite sequence data processing. Input is a sequence of satellite embeddings in every time step. Output is the confidence of spoofing in that time step.
  • ...and 1 more figures