Radio frequency interference (RFI) remains a major challenge for Global Navigation Satellite Systems (GNSS), particularly around critical infrastructure such as airports. As these infrastructures rely on stable and uninterrupted GNSS service, it becomes crucial to continuously monitor the spectrum and quickly identify any interference sources.
To address this need, we developed a low-cost network of GNSS sensor nodes capable of detecting, localizing, and characterizing interference sources under real-world conditions. Each sensor records time-synchronized baseband data and sends it to a central processing unit, where a TDOA-based localization algorithm detects the RFI, estimates its position, and characterizes its properties.
A central contribution of this work is the use of geographical constraints to improve the localization process. In many real situations—for example, when interference comes from a device carried inside a moving vehicle—the emitter is naturally limited to a certain area or a known route, such as a highway. By integrating this prior knowledge directly into the localization equations, we show that the system becomes noticeably more robust and, in some cases, can achieve reliable performance with fewer sensors.
To validate the proposed approach, we carried out an extensive measurement campaign including both static and dynamic scenarios. For static tests, when the interferer was located within the sensor network, the system achieved an average horizontal error of around 6 meters. In dynamic scenarios, the accuracy remained on the order of 50 meters. Beyond estimating position, the network was also able to track an unknown moving interferer, examine its signal in time and frequency, estimate its received power, and even infer the speed of the vehicle carrying the device.

Autonomous maritime vessels increasingly rely on Global Navigation Satellite Systems (GNSS) for Positioning, Navigation, and Timing (PNT), yet these systems remain vulnerable to jamming and spoofing attacks that compromise vessel safety. Traditional detection methods analyzing only GNSS signals struggle to distinguish attack types, particularly when sophisticated spoofing maintains valid signal properties.
This work introduces a fundamentally different approach. Rather than examining GNSS signals in isolation, we verify their integrity by checking whether reported vessel positions remain physically plausible against independent Automatic Identification System (AIS) maritime traffic data. The core insight: genuine GNSS measurements should produce trajectories consistent with realistic maritime behavior—vessel density patterns, speed distributions, course stability, and spatial relationships. This reframes interference detection as behavioral consistency verification between independent systems—essential for trustworthy autonomous navigation.
We demonstrate this using synchronized GNSS measurements from the FGI Spoofing and Jamming Dataset and AIS archives from Fintraffic. Neural networks fuse GNSS signal characteristics (magnitude statistics, spectral descriptors, phase-stability indicators) with AIS behavioral attributes (vessel density, speed variance, course stability, spatial dispersion). Rigorous feature engineering reduces 89 descriptors to 44 robust features through leakage-aware filtering.
Results are compelling. GNSS-only detection achieves 88% accuracy with substantial jamming-spoofing confusion. Adding AIS behavioral features improves accuracy to 98.4% (macro-F1: 0.981)—an 85% error reduction with only three misclassifications. The fused system achieves near-perfect jamming detection (F1: 0.995) and significantly enhanced spoofing recognition (F1: 0.969, +11.7 percentage points) by identifying kinematic impossibilities invisible to pure signal analysis.
Scenario-based cross-validation reveals opportunities for further development: performance variations across test scenarios indicate that multi-region data collection and domain adaptation will strengthen operational robustness. The work establishes technical feasibility and demonstrates substantial performance gains of behavioral cross-validation for maritime GNSS integrity monitoring. This approach offers a practical pathway toward resilient navigation systems essential for safe autonomous maritime operations and digital-twin infrastructures.

GNSS interference is an increasing threat to airborne navigation applications. For unmanned aerial vehicles this poses a special challenge, because there is fewer infrastructure for navigation than in commercial aviation. While GNSS jamming renders all or certain frequencies unavailable, spoofing is a more sophisticated type of interference, since falsified signals are harder to detect. Previous research on GNSS spoofing detection has led to the development of methods on the signal-level, authentication-based methods and cross-checking with Inertial Measurement Units (IMUs). Further detection methods are based on radio positioning other than GNSS, and methods that evaluate the navigation solution for consistency or plausibility. Our approach is a combination between cross-checking based on IMU data and consistency evaluation of the navigation solution. We propose an estimator that makes use of the Interacting Multiple Model (IMM) technique in a bank of GNSS/INS Kalman Filters (KFs).

Multiple Model (MM) methods can represent one dynamic system in different hypotheses. This makes these methods inherently applicable to GNSS interference scenarios. One such model could represent a non-spoofed nominal case and another one a spoofed GNSS position. The latter could be modeled in a simple approach by deliberately weighting down the accuracy of the GNSS observations. Implemented in a bank of KFs, the MMs are calculated in parallel, all given the same observation as an input. Subsequently, a weight for each KF is calculated, based on the likelihood of each KF’s innovation. The output of the KF bank is the weighted sum of every KF’s output.

MM methods can be subdivided into three generations, following the naming scheme of Li and Jilkov (2005). In the first generation of MM methods, called Multiple Model Adaptive Estimation (MMAE), the MMs do not interact. The second generation are those IMM methods that share information between models, but that have a fixed model structure. The third generation allows a variable structure of models that can also interact.

A preliminary study has shown that MMAE is capable of identifying the best fitting model under nominal conditions. However, if a spoofing scenario is introduced, MMAE is insufficient, because once it weights one model high, it sticks to that model, ultimately leading to the divergence of the KF bank. IMM methods of second or third generation could overcome this problem, because they are designed to switch. It is the assumption that different models can be true during different periods of time. The truth is seen as a sequence of models. Second generation IMM methods have been applied successfully to detect sensor and actuator failures (Zhang and Li, 1998). Therefore, IMM techniques offer a promising method for the detection of GNSS spoofing.

The presentation will cover the basic idea of our proposed navigation filter and how it is implemented in the open-source software framework INSTINCT (Topp et al., 2025). The potential of this approach will be shown through simulations and flight tests.

The increasing threat of intentional radio frequency interference (RFI) in the frequency bands of Global Navigation Satellite Systems (GNSS) drives the demand for establishing a reliable global monitoring system via satellites in the low earth orbit (LEO). The University of Federal Armed Forces Munich has started to build a dedicated small satellite mission with a configurable software-defined radio (SDR) payload that is used for GNSS RFI monitoring applications. The satellite serves as an experimental laboratory in space, operating in the LEO with an altitude of 505 km, following a sun-synchronous orbit with an inclination of 97.42 °. With a payload capacity of 75 kg and a takeoff weight of 200 kg, the platform is designed to support a wide range of advanced experiments. These include broadband communications and internet of things (IOT) applications, radio science experiments using GNSS signals for occultation and reflectometry measurements and monitoring applications of the GNSS RFI situation (Bachmann, et al., 2022). The launch of the satellite platform is scheduled for Q4/2026.
The RFI monitoring payload includes three key components: a zenith-facing antenna to receive GNSS signals from medium earth orbit (MEO) satellites; a nadir-facing antenna to capture signals from ground-based emitters; and a space-qualified GNSS receiver that provides precise position, velocity, and time (PVT) information of the space vehicle, that is needed for the determination of the geolocation of the RFI source. The heart of the payload is the GNSS SDR platform, that digitizes both the received GNSS signals from the zenith antenna and the nadir-facing antenna signals at a defined sampling rate and quantization resolution. Both signal streams are stored within an onboard mass memory before they are downlinked via X-band to a ground station.
Before receiving real data from the single satellite mission, a tool chain has to be developed to efficiently process the obtained signal streams. Therefore, a GNSS signal simulator is used to create realistic ground-based interference scenarios for proper testing of the developed software modules. At the moment, the tool chain has the capability to detect and geolocate basic kinds of RFI sources like continuous wave or pseudo-like waveforms. As the attack types become more advanced over time, the processing capabilities of the tool chain must increase accordingly to detect and provide reliable location information of realistic emitter sources. Therefore, the GNSS signal simulator was configured in such a way to generate spoofing-like emitter signals in a static spoofed position scenario. This was done by applying additional range and Doppler offsets to each of the authentic GNSS signals generated. The needed modifications in the processing chain are presented, as only the propagation-dependent range and Doppler information of the spoofing signal are needed for geolocation.

References:
Bachmann, J., Kinzel, A., Porcelli, F., Schmidt, A., Schwarz, R., Hofmann, C., Knopp, A. (2022). SeRANIS: In-Orbit-Demonstration von Spitzentechnologie auf einem Kleinsatelliten. Deutscher Luft- und Raumfahrtkongress, (S. 12). Dresden.