Method for Detecting Synchronous GPS Spoofing Based On A Hybrid CNN+Bilstm Neural Network Architecture
Vìsnik Nacìonalʹnogo unìversitetu "Lʹvìvsʹka polìtehnìka". Serìâ Ìnformacìjnì sistemi ta merežì · 22.06.2026 · 19 · 371-381
DOI: 10.23939/sisn2026.19.371 ↗Kurz gefasst
Fachartikel in der Zeitschrift Vìsnik Nacìonalʹnogo unìversitetu "Lʹvìvsʹka polìtehnìka". Serìâ Ìnformacìjnì sistemi ta merežì, online erschienen am 17.05.2026. Verfasst von Dmytro Dosyn und 1 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben.
Abstract (Originalsprache)
The article develops and investigates a method for detecting intelligent GNSS signal manipulations based on a hybrid deep learning architecture. The aim of the article is to develop an intelligent defense system that integrates convolutional layers (CNNs) and bidirectional recurrent units (BiLSTMs) to detect hidden anomalies in the dynamics of GPS tracking parameters. The paper also analyzes unique aspects of synchronous spoofing, which is considered the most covert form of cyber-physical interference, relying primarily on the imitation of both code-phase and Doppler effects of authentic GNSS signals. Furthermore, the limitations of previous forms of statistical methods (i.e., C/N0 monitoring) are demonstrated, as they fail in more complex dynamic situations (where attackers can dynamically change power levels). The proposed CNN-BiLSTM architecture allows for a comprehensive consideration of signal parameter inertia and correlation links in sequential samples, which is critical for recognizing attacks with minimal power deviation. The experiment is based on the ds7 scenario from the well-known TEXBAT (Texas Spoofing Test Battery) dataset. Spatio-temporal features were derived from raw I/Q (In-phase/Quadrature) samples to capture the dynamics of correlation-peak degradation. The model architecture is based on one-dimensional convolutional blocks for noise removal and BiLSTM layers for identifying bidirectional temporal dependencies. Since the model utilizes bidirectional recurrent structures, it can account for signal context—analyzing context in both forward and backward temporal directions, which is essential for detecting smooth Doppler shift manipulation. Thanks to the implementation of the hybrid architecture, a detection accuracy of 91% was achieved on a test set of independent satellites. Features are formed from GPS receiver tracking parameters, including C/N0 statistics, residual phase error, and cross-correlation inter-satellite metrics. Moreover, the advantage of combining spatial and temporal analysis for detecting power-matched attacks was proven, as traditional amplitude detectors are the least effective. Our results show that using CNN-BiLSTM architectures in DSP receivers is an effective approach to preventing technologically advanced falsifications of critical navigation infrastructure. A key scientific contribution is the identification of a methodological error in the standard procedure for common-mode noise correction, which, when including spoofed satellites in the average calculation, completely masks the attack signal. Correcting this error provided a jump in detection quality from 50% to 91%.
Zitieren
APA
Dosyn, D., & Sydoriv, M. (2026). Method for Detecting Synchronous GPS Spoofing Based On A Hybrid CNN+Bilstm Neural Network Architecture. Vìsnik Nacìonalʹnogo unìversitetu "Lʹvìvsʹka polìtehnìka". Serìâ Ìnformacìjnì sistemi ta merežì, 19, 371-381. https://doi.org/10.23939/sisn2026.19.371
Vancouver
Dosyn D, Sydoriv M. Method for Detecting Synchronous GPS Spoofing Based On A Hybrid CNN+Bilstm Neural Network Architecture. Vìsnik Nacìonalʹnogo unìversitetu "Lʹvìvsʹka polìtehnìka". Serìâ Ìnformacìjnì sistemi ta merežì. 2026;19:371-381. doi:10.23939/sisn2026.19.371
BibTeX
@article{Dosyn2026,
author = {Dosyn, Dmytro and Sydoriv, Maksym},
title = {Method for Detecting Synchronous GPS Spoofing Based On A Hybrid CNN+Bilstm Neural Network Architecture},
journal = {Vìsnik Nacìonalʹnogo unìversitetu "Lʹvìvsʹka polìtehnìka". Serìâ Ìnformacìjnì sistemi ta merežì},
year = {2026},
volume = {19},
pages = {371--381},
doi = {10.23939/sisn2026.19.371}
}
EUVIMED