GNSS Spoofing Attack Detection

Updated 20 April 2026

GNSS spoofing attack detection is a suite of techniques that distinguishes authentic satellite signals from adversarial ones using signal-level analysis and probabilistic models.
It leverages methods such as power-distortion monitoring, carrier-phase analysis, and sensor fusion (including inertial and network data) to achieve high detection accuracy with low false positives.
The integration of machine learning, federated learning, and quantum-enhanced frameworks supports real-time, robust defenses for critical infrastructures against evolving spoofing attacks.

Global Navigation Satellite System (GNSS) spoofing attack detection refers to the suite of methodologies, algorithms, and system designs aimed at autonomously distinguishing authentic satellite navigation signals from adversarially forged signals that deceptively manipulate a receiver’s computed position, velocity, or time solution. Civil GNSS signals, especially those unprotected by cryptographic authentication, are fundamentally vulnerable to spoofing, posing risks to critical infrastructures, autonomous systems, and public safety. The increasingly pervasive integration of GNSS receivers in consumer devices, vehicles, and infrastructure has positioned spoofing detection as both a foundational research area and a high-stakes application domain.

1. GNSS Spoofing Attack Taxonomy and Threat Models

GNSS spoofing attacks are deliberate attempts to mislead receivers by transmitting counterfeit signals that, from the victim’s perspective, are indistinguishable from authentic navigation satellite signals at the waveform level. Principal categories include:

Carry-off/Position-push attacks: The attacker gradually aligns forged signals with authentic ones and then smoothly displaces the victim's position or time over several seconds or minutes, thereby evading abrupt-detection mechanisms.
Time-push attacks: Specific focus on manipulating the perceived time at the receiver, especially dangerous for synchronization-critical applications (e.g., NTP, SCADA, 5G base stations) (Enan et al., 25 Aug 2025).
Turn-by-turn, overshoot, wrong-turn, and stop attacks: Highly targeted models for vehicular navigation, where spoofed GNSS outputs mimic plausible urban maneuvers or freeze/shift positions to induce erroneous behavior in autonomous systems (Dasgupta et al., 2024).
Partial attacks and coordinated scenarios: Only a subset of satellites or receivers are spoofed, or colluding adversaries introduce more sophisticated multi-point inconsistencies (Liu et al., 2024).

The attacker’s capabilities may include precise power control, code-phase and carrier-phase alignment, multi-antenna emission, and intricate knowledge of the victim’s expected dynamics, resulting in attacks difficult to detect via traditional quality-of-service metrics alone.

2. Signal–Domain Countermeasures and Physical-Layer Testing

Signal-level GNSS spoofing detection aims to exploit features of the raw correlation process or physical-layer observables:

Power and Distortion Monitoring: The Power-Distortion (PD) detector leverages the statistical independence between received AGC power and correlation profile distortion. This technique traps stealthy attackers in a trade-off: raising power identifies blatant spoofing, while code-phase overlap yields measurable distortion. Bayesian M-ary hypothesis testing classifies each channel sample as nominal, multipath, spoofed, or jammed, with per-channel false-alarm rates below 0.6% in large-scale empirical studies (Wesson et al., 2017).
Carrier-Phase and INS Coupling: Tightly binding short-term antenna motion (measured by on-board consumer-grade or industrial IMUs) with carrier-phase residuals allows real-time discrimination between authentic geometric diversity and the degenerate response of any single-antenna spoofer. Real and spoofed signals cannot match the high-frequency, centimeter-scale carrier-phase variations arising from unshared receiver movement. Neyman–Pearson-style hypothesis testing of the residual cost achieves >90% accuracy in both static and vehicle trials, with sub-2 s detection latency even using mass-market inertial sensors (Johansson et al., 6 Feb 2025, Clements et al., 2022).
Sparse Recovery at the Correlator Level: Convex-optimization (LASSO) techniques recover the code-phase spectrum from the bank of correlator tap outputs, identifying the simultaneous presence of both authentic and spoofed peaks. A simple threshold on the ratio of secondary to primary peaks reliably detects the two-peak “spoofing fingerprint” at sub-percent error rates in both simulation and recorded spoofing events (Schmidt et al., 2020).

These methods are characterized by minimal hardware modification, ability to immediately suppress navigation during an alarm, and robust operation under multipath and non-ideal urban conditions.

3. Opportunistic, Cross-Modality, and Crowdsourced Strategies

Modern mobile and vehicular platforms are equipped with additional sensing modalities beyond GNSS, providing a rich basis for cross-validation:

Sensor Fusion and Inertial Consistency: A canonical approach fuses GNSS positions, network-based locations (Wi-Fi, cellular), and IMU readings. By fitting motion-constrained regressions and quantifying uncertainty through Gaussian processes, these methods produce a composite likelihood for the current location. Detection proceeds via Neyman–Pearson testing, outlier-ensemble methods, or direct regression (Liu et al., 2023, Liu et al., 14 Jun 2025). Detection delays are reduced to near zero for significant deviations, and mean fallback position errors are drastically lower than network-only or IMU-only baselines.
On-Board Sensor LSTM and Turn Validation: Sensor-fusion frameworks for AVs deploy stacked LSTM or random forest models to cross-check the GNSS-inferred distance traveled and turn events against IMU-, accelerometer-, steering-axis-, and speedometer-derived predictions (Dasgupta et al., 2024, Dasgupta et al., 2021). These methods detect not only abrupt spoofing attacks but are also effective against slow-drift, wrong-turn, or standstill-while-in-motion attacks with empirically observed AUC scores ≥0.99 in real-world datasets.
Crowdsourced and Peer-Based Anomaly Detection: Distributed receivers in urban or vehicular networks enable additional spoofing detection strategies. The D2SP (Double Differential Pseudorange Spatial Distribution) approach constructs the empirical variance of double-differenced pseudoranges across receiver–satellite pairs, leveraging random set theory and trilevel statistical testing to distinguish between spoof-free, fully-spoofed, and partially-spoofed regimes, achieving high spoof-detection probabilities even under severe urban multipath (Chen et al., 2024). Peer-to-peer ranging using protocols with time-of-arrival codes and trusted execution environments enables multi-user collaborative consistency checking in dense road networks, with robust meta-protocols for adversarial situations (Dutta et al., 2023).

These cross-modality approaches extend GNSS spoofing detection into signal-denied, infrastructure-poor, or highly adversarial environments (e.g., autonomous vehicles operating in cities or collaborative road-user contexts).

4. Machine Learning, Federated, and Quantum-Enhanced Methods

The application of both classical and quantum-inspired learning models deepens the sophistication and generalization capability of GNSS spoofing detectors:

Self-Supervised Federated Learning: Platforms distribute model parameters, not raw position or sensor data, via federated averaging (FedAvg): each device runs a local LSTM-based anomaly detector, generating self-supervised spoofing-deviation labels from opportunistically fused positions and their uncertainty. Local models are updated asynchronously and aggregated in the cloud. The federated approach attains an AUC of ≈87.4%, outperforming centralized and conventional position-based schemes while preserving data privacy (Liu et al., 9 May 2025).
Deep Sequence-to-Sequence Detection: Synthetic datasets simulating diverse, global, and localized GNSS spoofing are used to train LSTM- and transformer-inspired architectures. These models process second-differenced pseudoranges with learned quantized embeddings. Transformer models with early fusion deliver the lowest classification error rates (0.16%), generalizing across geographic and signal variability (Zelinka et al., 22 Oct 2025).
Quantum–Classical Hybrid Autoencoders: Hybrid Quantum-Classical Autoencoder (HQC-AE) architectures learn to reconstruct authentic GNSS tracking-stage features. Samples with high re-projection error (measured by an IQR-based threshold) are flagged as spoofed. HQC-AE achieves >98% accuracy against sophisticated zero-day time-push spoofing attacks, outperforming all classical autoencoders and unsupervised learning-based methods (Enan et al., 25 Aug 2025).
Reinforcement Learning-Based Strategies: RL agents adaptively adjust detection thresholds on distance-differential features between GNSS and on-board predictions. Bayesian Online Change Point Detection (BOCPD) on RL value-functions enables rapid (mean 2.3 s delay), accurate (98.6%) detection of drift-evasive attacks, outperforming supervised LSTM and Page-Hinkley detectors (Panda et al., 15 Jul 2025, Dasgupta et al., 2021).

These data-driven frameworks are notable for their robustness to attack variants, ability to detect unseen strategies, and minimal dependence on signal-level idiosyncrasies.

5. Probabilistic Frameworks, Multi-Subset Fusion, and Statistical Guarantees

Advanced probabilistic models generalize acoustic Receiver Autonomous Integrity Monitoring (RAIM) frameworks and likelihood-ratio approaches:

Gaussian Mixture Subset Fusion: Extended RAIM schemes generate location estimates for all combinations of GNSS, signals-of-opportunity (SOPs), and sensor measurements, subject to kinematic constraints. Gaussian mixture models integrate the results, with detection based on the density of the full-constellation GNSS solution under the mixture. If sufficient benign subsets persist, theoretical bounds guarantee low false-alarm rates and effective spoofing detection even amidst untrustworthy SOPs or coordinated attacks (Liu et al., 2024).
Analytical Bounds and Physical-Layer Adversarial Games: Using a unified signal model, adversarial attacker (Eve) and defender (Bob) strategies are cast as a zero-sum game. The optimal detection test is shown to be the generalized likelihood ratio test (GLRT), with the fundamental achievable error region given by a Kullback–Leibler divergence parameterized by the difference in satellite-channel matrices. The Nash equilibrium highlights the value of satellite diversity and authenticated channels, and provides concrete performance bounds independent of attacker details (Crosara et al., 2023).

These frameworks facilitate both receiver-side algorithm design and system-level performance evaluation, explicitly accounting for attacker adaptation and information-theoretic constraints.

6. Benchmarking, Validation, and Deployment

Spoofing detection schemes have been systematically validated across multiple real and synthetic datasets:

Controlled drive-test traces using consumer smartphones and u-blox RTK receivers (Liu et al., 9 May 2025), public driving and AV datasets (Dasgupta et al., 2021, Dasgupta et al., 2024), and large-scale IRIDIUM satellite monitoring (Oligeri et al., 2020).
Experimental metrics: Receiver Operating Characteristic (ROC) curves, Area Under Curve (AUC), false/true positive and negative rates, detection delay, Maximum Absolute Error (MAE) for location fallback, and real-time execution cost.
Comparative performance shows that cross-modality, federated, and ML-based schemes achieve true positive rates routinely ≥90% at low fixed false-positive rates (≤5%), with detection delays on the order of 1–10 s. Advanced statistical/crowdsourced detectors operate efficiently even in the presence of strong multipath, partial attacks, and colluding adversaries.

Deployment considerations include privacy (only model weights or derived statistics are shared), device heterogeneity, consumption of computational and energy resources, and seamless integration with OS-level APIs, making these approaches practical for wide-scale adoption on modern consumer and automotive platforms (Liu et al., 9 May 2025, Liu et al., 14 Jun 2025).

7. Research Challenges and Future Directions

GNSS spoofing attack detection remains an adversarially dynamic field. Open research topics include:

Extending detection to highly synchronized or multi-antenna spoofers capable of mimicking both signal-level and motion-derived statistics.
Tightening statistical bounds for GLRT-based and subset-fusion detectors under real-world complex error distributions.
Scaling federated learning frameworks for global deployments while preserving privacy and resilience against untrustworthy clients or servers.
Reducing energy and computational overhead for continuous mobile/vehicle-side execution.
Fusion of additional modalities (e.g., visual odometry, LiDAR, cooperative ranging) for challenging urban and infrastructure-free environments.
Automated online adaptation to evolving attack tactics without sacrificing generalization or false-alarm performance.

The convergence of cross-domain sensing, advanced probabilistic modeling, machine- and quantum-learning, and scalable system architectures defines the modern landscape for robust GNSS spoofing detection (Liu et al., 9 May 2025, Enan et al., 25 Aug 2025, Liu et al., 2023, Johansson et al., 6 Feb 2025, Zelinka et al., 22 Oct 2025).