Global Navigation Satellite Systems (GNSS) have become the cornerstone of Positioning, Navigation, and Timing (PNT) services, playing a pivotal role in modern maritime navigation. Beyond providing absolute and accurate positioning, GNSS serves as the primary timing reference for a large number of systems onboard a vessel, such as the Electronic Chart Display and Information Systems (ECDIS) or the Automatic Identification Systems (AIS). However, GNSS is extremely vulnerable to radio frequency interference, such as jamming and spoofing, which poses a critical threat to the integrity of positioning solutions, especially in safety-critical maritime environments. Therefore, any degradation in GNSS signals can compromise the operation of such systems which strongly depend on the provision of accurate PNT information, which consequently can jeopardize the vessel’s safety. A major challenge in detecting interference lies in the accessibility to the complete GNSS receiver chain, including the antenna, front-end and baseband blocks. While advanced digital signal processing (DSP) techniques offer high sensitivity to interference, they are typically inaccessible to users of mass-market GNSS receivers due to their high computational complexity, hardware requirements, and cost. To address this, we propose a more pragmatic approach: interference detection at the observation level using Machine Learning (ML) techniques. We use the GNSS observables (i.e., pseudorange and carrier-phase measurements) extracted from the receiver output, which are then processed using ML models trained on real-world jamming conditions. This method is particularly suitable for real-world deployment, as it operates on standard receiver outputs (i.e., RINEX) and does not require access to internal signal processing stages or highly demanding DSP techniques. ML is particularly well-suited for this jamming detection because it can identify complex, non-linear patterns in corrupted data without requiring explicit modeling of interferences. However, this approach has limitations: unlike robust DSP or adaptive antenna arrays, observable-based techniques rely on the presence of detectable anomalies after the interference has already affected the signal. Finally, the proposed method offers a highly accessible, low-cost, and scalable solution for interference detection in mass-market maritime applications. It enables early detection of interference allowing for safer navigation.

Positioning, navigation and timing services provided by Global Navigation Satellite Systems (GNSS) have become essential across various sectors and support a host of critical infrastructure and services. Yet, since received satellite navigation signals are inherently weak, they are highly vulnerable to Radio Frequency Interference (RFI).

Established and traditional methods of RFI detection and monitoring, such as the use of aircraft Automatic Dependent Surveillance–Broadcast (ADS-B) messages or dedicated ground infrastructure, have notable limitations. ADS-B relies on single-frequency receivers and provides limited insight into ground-level interference, while dedicated monitoring networks are costly to deploy and maintain.

In parallel, smartphones have become ubiquitous and are equipped with increasingly more robust GNSS receivers, often capable of tracking multiple constellations and frequencies. Moreover, devices that run Android 10 (API level 29) or higher are mandated to support access to raw GNSS measurements, presenting a scalable, low-cost opportunity for crowdsourced GNSS RFI monitoring using existing consumer hardware.

The European Commission’s Joint Research Centre (JRC) has dedicated efforts to explore the reliability of the use of the Automatic Gain Control (AGC) measurements to detect and monitor RFI in GNSS frequency bands. A proof-of-concept Android application, CrowdShield, has been developed and tested in controlled laboratory and open-sky conditions, including Jammertest 2025. The app provides both real-time detection information based on AGC and upload of AGC measurements to a cloud back-end server for data analysis and visualization.

In December 2025, a field trial was conducted across all six JRC sites in Europe. Volunteering staff was provided access to a test version of the CrowdShield app and encouraged to log data throughout their workday.

This paper details the trial design and implementation, discusses challenges in application development, and outlines the approach to data aggregation and analysis. A particular effort is dedicated to calibration of AGC measurements from different smartphone models and chipsets, in order to allow for efficient aggregation in space and time.

While previous demonstrations relied on a limited set of recent dual-frequency smartphones, the field trial enabled an in-depth evaluation of how different smartphone generations and Android versions handle RFI and facilitate access to raw GNSS measurements. As part of the trial, a 1 mW continuous wave jammer close to L1 frequency was operated at the JRC Ispra (Italy) to assess the feasibility of localising interference sources through aggregated AGC measurements in the vicinity of the jammer. This effort further emphasises the potential of mass-market technology to support resilient navigation systems and complement other RFI monitoring techniques.

Global Navigation Satellite System (GNSS) signal interference has been recognized as a growing concern for many users of these systems, for instance, civil aviation or shipping [1], and has been identified as a contributing factor to previous accidents [2]. This has sparked diverse research initiatives to understand and address the threat, such as investigations of interference detection and localization systems. Previous research has highlighted the significant utility of automatic dependent surveillance – broadcast (ADS-B) transmissions from aircraft for both wide-area GNSS interference detection as well as coarse localization of its source [3]. Evaluating and testing these GNSS interference localization systems can be challenging, as interference transmission is generally prohibited. This paper investigates an approach to use ADS-B signals as a signal of opportunity (SOOP) to test a snapshot-based time difference of arrival (TDoA) GNSS interference localization system on real-world data over a wide area without the need for radio-frequency (RF) transmission permits.
The TDoA interference localization system consists of multiple physically unconnected GNSS snapshot receivers which are synchronized ad-hoc using GNSS observables, while no interference signals are present. When an interference signal appears, the sensor network computes accurate TDoA measurements for several minutes until the sensor clocks drift apart. To test this system, a mixer and combiner unit is added before the receiver RF input, which mixes the ADS-B band into a side-band of the GNSS L1/E1 band, thus emulating the presence of an interference signal to the receiver. The TDoA of these ADS-B signals observed by multiple receivers will be consistent with the location of the transmitting aircraft, allowing TDoA localization of the aircraft as if it were an actual GNSS interference transmitter. Since some ADS-B transmissions contain the current location of the transmitting aircraft, the ground-truth corresponding to a measurement obtained from these messages can be directly obtained by demodulating the signal. The paper demonstrates the feasibility and convenience of this method on a set of real-world measurements from two receivers. Every ADS-B burst seen by both receivers yields a single TDoA measurement. Using the demodulated ground truth, the TDoA measurement error, which directly translates to localization error in a system with multiple receivers, is computed and statistically analyzed. Based on the TDoA error statistics, the resulting localization error of an example deployment of multiple receivers is computed. Finally, the main challenges observed during this research, short snapshot observation period and gain control, are discussed.

References
[1] OPSGROUP GPS Spoofing Workgroup 2024. GPS Spoofing. OPSGROUP, Sept. 6, 2024. url: https://ops.group/dashboard/wp-content/ uploads/2024/09/GPS-Spoofing-Final-Report-OPSGROUP-WG-OG24. pdf (visited on 07/23/2025).
[2] Dana A. Goward. “GNSS Interference in Ship Collision, Fires, Grounding”. In: Inside GNSS (June 25, 2025). url: https://insidegnss.com/ gnss-interference-in-ship-collision-fires-grounding/ (visited on 12/05/2025).
[3] Zixi Liu et al. “Investigation of GPS Interference Events with Refinement on the Localization Algorithm”. In: 2023 International Technical Meeting of The Institute of Navigation. Long Beach, California, Feb. 13, 2023, pp. 327– 338. doi: 10.33012/2023.18627.

Commercial GNSS receivers are vulnerable to both intentional and unintentional radio frequency
interference (RFI), which can lead to navigation errors or complete signal loss [6]. Such anomalies primarily include jamming—where high-power signals obstruct satellite transmissions—and spoofing, which involves broadcasting counterfeit signals to mislead receivers. These threats have grown due to geopolitical tensions, the availability of low-cost signal emulation platforms, and the commercialization
of jamming hardware. Modern commercial GNSS receivers employ strategies such as multi-frequency and multi-constellation tracking and advanced interference monitoring technologies to mitigate such risks e.g., AIM+ by Septentrio [12], Ublox [14] and GRIT by Novatel [10]. Recent studies show that observables from commercial receivers can be effectively utilized to develop anomaly detection algorithms
[9, 11]. Leveraging on these readily available observables, the project titled, Resilient Positioning in Eastern Finland (GNSS-ITÄ) [4], aims to develop a software toolbox namely FGI-Shield, designed for real-time interference detection and situational awareness. Thus, by exploiting data from commercial GNSS receivers, this initiative aims to develop a deployable research tool for detailed RFI event analysis,
compared to commercial receiver’s black box methods. The goal is to enable receiver agnostic and extendable anomaly detection for research use.

The objective of this study is to evaluate the Chi-Square Test (CST) anomaly detection technique [3, 13], for integration into FGI-Shield. As demonstrated by [1], CST achieved high detection accuracy when evaluated on raw digitized I/Q baseband samples from multiple front ends. Extending this work, in this study the CST-based anomaly detection is applied on the snapshot I/Q data recorded by a commercial GNSS receiver. These snapshots, comprising of successive complex baseband samples, enables the possibility for signal monitoring and spectral analysis across supported GNSS bands [2, 5]. This study also examines the receiver’s anomaly detection capability using built-in indicators, its effect on relative positioning performance, and evaluates CST’s potential to improve sensitivity towards L1-band
interference. Experiments were conducted using datasets from Jammertest 2025 (JT25) [7], which include jamming, spoofing, meaconing, and unintentional interference scenarios, each preceded and followed by sufficient clean data for baseline comparison.

The detection performance was evaluated using Accuracy (overall correctness), Recall (detecting all actual positives), impact on positioning availability, and 3D Root Mean
Square error (3DRMS). The evaluation demonstrates that the receiver generally detected interference well and maintained accurate positioning, except in certain spoofing scenarios. Thus, the impact on the position solution is relatively low largely due to receiver’s multi-frequency configuration and advance anomaly detection algorithms. In contrast, CST delivered consistently strong anomaly detection on L1,
outperforming the receiver’s L1-only detection in several cases. These findings support integrating CST into FGI-Shield as a complementary layer to receiver flags, improving sensitivity to L1-centric threats and strengthening overall interference resilience.

Global Navigation Satellite Systems (GNSS) are increasingly threatened by radio-frequency interference (RFI), including both jamming and spoofing, which degrade positioning reliability. This paper presents a hybrid detection framework that integrates statistical hypothesis testing, stochastic process modeling, and machine learning to robustly identify interference events. We derive interpretable detection statistics from generalized likelihood ratio tests (GLRT),
Hotelling’s T^2 , and covariance shape–scale indices (trace, determinant, and the arithmetic-to-geometric mean (AM/GM) ratio of eigenvalues). These multivariate tests capture structural deviations in GNSS observables indicative of interference.
To enhance temporal sensitivity, we also employ first-order autoregressive (AR(1)) modeling with median-absolute-deviation (MAD)–based Z -scores for robust residual detection, and we characterize volatility effects through Autoregressive Conditional Heteroskedasticity (ARCH), Generalized ARCH (GARCH), and stochastic volatility formulations.
Each base detector produces statistical scores that are fused within a metaclassifier together with GNSS-derived features, including east–north–up (ENU) variation metrics, signal-to-noise ratio (SNR) statistics, Doppler and carrier-phase residual indicators, power spectral density (PSD) features, and eventlog parameters. The meta-classifier integrates these heterogeneous inputs to (1) detect whether the measurements are affected by RFI and, when interference is present, (2) classify events as either jamming or spoofing. The overall framework is designed to operate using data available from commonly deployed GNSS receivers, supporting practical monitoring and operational deployment.
The methodology is validated using measurements from the 2024 Jammertest campaign held at Andøya, Norway. Results demonstrate that the proposed ensemble framework reliably detects RFI events and provides interpretable indicators of
interference type, illustrating its potential as a robust and adaptive detection solution for real-world GNSS monitoring environments.