Galileo has pioneered the global introduction of a data authentication service, named Open Service Navigation Message Authentication (OSNMA), allowing end-user receivers the capability to verify the Galileo navigation messages source authenticity and data integrity.
The OSNMA is provided to the end-users over the Galileo E1 I/NAV 20bps capacity, via the Timed-Efficient Stream Loss-Tolerant Authentication (TESLA) protocol.
Following an intense set of verification and E2E validation activities, started on the 18th November 2020 with the first transmission, OSNMA has entered officially in its initial Service Phase as of the 24th July 2025.
OSNMA has been successfully penetrating the GNSS receivers’ market, with major players integrating OSNMA functionality into their products well before the official service declaration. This early adoption enabled experimentation with receiver implementations, supporting EC/EUSPA and ESA in consolidating and validating the publicly available receiver guidelines and the OSNMA ICD.
Over the last few years, several receiver models in the market have officially incorporated OSNMA and, as of today, the number of OSNMA-enabled receivers are counted by millions.
This work aims at providing a methodology to characterise and compare OSNMA performance across different receiver technologies available in the market, under challenging operational conditions. By “challenging operational environment”, we refer to scenario predominantly affected by signal fading and blockage, as typically experienced in deep urban environments.
This work will address:
• The Test Setup tools and methodologies used by the ESA receiver team for the execution of mobile testing activities in urban areas, such as in Rotterdam downtown
• The OSNMA-Enabled Mass-Market Receivers test bench available at ESA/ESTEC
Note: Receiver manufacturers will be kept anonymous throughout this abstract and the associated paper.
• The relevant Key Performance Indicators (KPIs)
• The measured OSNMA performance across the different receiver technology and derivation of relevant implemented optimisation to maximize OSNMA data extraction from SiS
Performance assessment on field will be carried out over three different levels:
1. Positioning Level:
o PVT availability: Percentage of time, within the observation window, during which the receivers produce a valid 3D position, regardless of their accuracy.
o PVT accuracy: 95% percentile of horizontal and vertical accuracy, measured only over valid PVT epochs.
2. OSNMA PVT level:
o Number of visible satellites per epoch from which the receiver is collecting OSNMA data.
o Number of Galileo SVIDs for which Clock and Ephemeris data are successfully authenticated at each MACK Epoch (currently every 30s as per current OSNMA SIS configuration).
o Number of Galileo SVIDs with timing parameters (GST to UTC and GGTO) successfully authenticated within the corresponding tag type dissemination epoch (every 60 seconds as per current OSNMA SIS configuration).
o Time between Consecutive TESLA Key Authentications.
o Time between Consecutive Authentications of at least 4 different satellites.
Additionally, receiver’s raw bits are analysed and processed offline to derive preliminary assessment of Time to First Authenticated Data, measured across different receiver OSNMA start-up assumptions and different OSNMA processing optimisations exploiting an open-source python based OSNMA library.

In dense urban environments, buildings severely obstruct GNSS signals, causing positioning errors and unstable satellite visibility. This study proposes a real-time method for modelling sky obstructions using vehicle-based GNSS trajectories and building height data. For each trajectory point, a local grid is constructed to estimate elevation angles of obstructed directions and generate a corresponding skyplot.

To overcome the high computational cost of frame-by-frame skyline generation, a template reuse mechanism is introduced. Skyline profiles are compared using cosine similarity, combined with spatial neighbourhood constraints and Area of Interest (AOI) overlap checks. If the current environment is similar to a stored template, its skyplot is instantly reused; otherwise, a new template is generated. This preserves local geometric characteristics while avoiding redundant computation.
Experimental results show that the method reduces average processing time from several seconds to under one second per epoch, without compromising obstruction accuracy or skyplot morphology. The proposed approach provides an efficient framework for real-time GNSS visibility modelling and urban obstruction assessment.

Accurate estimation of Zenith Tropospheric Delay (ZTD) from Global Navigation Satellite System (GNSS) observations is essential for high-resolution atmospheric monitoring and numerical weather prediction. While Precise Point Positioning (PPP) achieves millimeter-level ZTD accuracy under open-sky conditions, its performance degrades severely in dense urban environments due to multipath interference and non-line-of-sight (NLOS) signal propagation. These effects limit the usability of crowdsourced and low-cost GNSS observations, including smartphone and public transport GNSS data, for urban tropospheric sensing. This study presents a geometry-aware framework that integrates 3D city models and electromagnetic ray tracing into PPP processing to enable reliable urban ZTD estimation. GNSS signal propagation is explicitly classified at each satellite–receiver epoch into line-of-sight (LOS), reflection, diffraction, mixed multipath, and NLOS components using building geometry and physically based reflection (Fresnel equations) and diffraction (Uniform Theory of Diffraction) coefficients. These classifications drive adaptive observation exclusion and reweighting within the PPP filter. In parallel, a city-scale ray-tracing analysis is used to identify “healthy urban zones” with sustained LOS visibility, suitable for crowdsourced GNSS data collection. Experiments using real urban GNSS data demonstrate that conventional PPP suffers from severe instability in urban canyons, with ZTD biases exceeding 10 m and large positioning errors. The proposed ray-tracing–assisted PPP reduces code residuals to the meter level, stabilizes carrier-phase residuals, and improves ZTD accuracy by more than two orders of magnitude, achieving sub-decimeter agreement with reference stations. Healthy-zone identification further enables systematic selection of urban locations where reliable tropospheric retrieval is feasible. These results demonstrate that physics-based, geometry-aware multipath modeling is essential for extending GNSS tropospheric sensing into dense urban environments and provides a scalable pathway toward high-resolution urban atmospheric monitoring using crowdsourced GNSS infrastructure.

Global Navigation Satellite Systems (GNSS) are fundamental to safety-critical autonomous systems, yet achieving high integrity and accuracy in diverse, degraded environments (e.g., urban canyons) remains a formidable challenge. While machine learning (ML) offers the potential to capture complex environmental patterns, conventional end-to-end learning approaches often suffer from "shortcut learning", which relies on spurious correlations between signal features and positions rather than the underlying physical causality. This leads to fragility in out-of-distribution (OOD) scenarios and uncalibrated uncertainty estimates that affect downstream sensor fusion. This research aims to bridge this gap by proposing a causal-regularized GNSS estimator. Our objective is to transform the positioning task from a black-box regression into a causal inverse inference problem, ensuring the learned estimator adheres to GNSS physical axioms and geometric principles even in unseen environments.

We introduce a novel framework that learns the inverse mapping from multi-satellite observations to receiver states and calibrated uncertainties while structurally enforcing physical consistency. The architecture embeds critical inductive biases by explicitly disentangling pseudorange residuals into common-mode (clock-bias dominated) and centered (position-informative) components, structurally preventing causal confusion between timing and geometric errors. The core innovation lies in our causal regularization training strategy, which anchors the inverse estimator to the forward physical process. Instead of relying solely on label matching, we generate counterfactual data pairs by applying controlled physical interventions (e.g., a known position shift) to real samples. We then impose a differential forward consistency constraint: the network is required to predict states for both the factual and counterfactual inputs, and the difference between these predictions must geometrically explain the induced observation change. Furthermore, we enforce a geometric stiffness axiom, penalizing the model if its predicted uncertainty does not increase monotonically with degrading satellite geometry.

To rigorously validate these properties, we establish an axiom-based verification benchmark that moves beyond standard accuracy metrics. This benchmark subjects the trained estimator to controlled intervention tests: 1) verifying line-of-sight sensitivity, 2) confirming position invariance under common-mode errors, and 3) checking if uncertainty scales with geometric stiffness, thereby showing the model has internalized physical laws rather than merely overfitting to correlations. We anticipate that the causal-regularized estimator will demonstrate superior generalization performance compared to both traditional robust least squares and standard end-to-end deep learning baselines, particularly in OOD scenarios characterized by high geometric dilution of precision (GDOP) or unseen multipath distributions. Specifically, the "counterfactual difference" constraint is expected to significantly reduce positioning outliers by preventing the model from overfitting to environmental nuisances. Furthermore, the inclusion of the geometric stiffness axiom will yield uncertainty estimates that are highly correlated with geometric truth, providing a credible confidence metric essential for safety-critical integration.

This work establishes a practical "causal ML for GNSS" framework to improve robustness against environmental shifts. Furthermore, it introduces a verification-first methodology that validates estimators based on physical axioms, reducing reliance on empirical tuning and paving the way for hybrid neural-physical navigation systems.

Global Navigation Satellite System Interferometric Reflectometry (GNSS-IR) exploits signal-to-noise ratio (SNR) variations caused by multipath interference to sense environmental changes. While GPS-based IR has been widely applied to surface monitoring applications, such as soil moisture, snow depth, and water level estimation, the use of BeiDou SNR observations for atmospheric disturbance and storm detection remains largely unexplored.
A residual-based framework is applied at three permanent GNSS stations during a storm day, in which SNR observations were compared with the days without this phenomenon. Storm sensitivity is quantified using a Storm/Quiet ratio derived from residual exceedances across multiple SNR bands (SNR1–SNR5). Results from the 30-second data show moderate storm responses at ALME station in Spain and ARA2 station in Slovenia, primarily in the lower SNR bands (SNR1 and SNR2), while BeiDou-IGSO satellites at TORI station in Italy also exhibit a moderate response. In contrast, the 1-second observations significantly enhance storm signatures for GPS, with increased Storm/Quiet ratios indicating that high-rate data capture short-lived storm-induced variations more effectively than 30-second sampling. BeiDou responses, however, remain largely unchanged between the two sampling rates, suggesting limited sensitivity gain from higher than 30s sampling resolution.
These findings demonstrate that high-rate GNSS-SNR significantly enhances storm detectability for GPS and provide, to our knowledge, the first assessment of BeiDou SNR for storm detection. The results highlight both the potential and the limitations of multi-GNSS IR for monitoring fast-evolving atmospheric phenomena.