The increasing requirement for real-time Low Earth Orbit (LEO) satellite clock products to support LEO-augmented Positioning, Navigation, and Timing (PNT) services has made filter-based clock estimation combined with ultra-short-term prediction the most practical solution. While high accuracy of real-time LEO satellite clock products remains crucial, ensuring their reliability through Integrity Monitoring (IM) techniques is equally essential for operational deployment. However, the filter-based strategy relies on a relatively weak observation model, which enhances its susceptibility to biases, such as real-time GNSS Signal-In-Space Ranging Errors (SISREs). These biases can propagate within the filter and directly affect the computation of Protection Levels (PLs) of LEO satellite clocks.
To address this, an IM algorithm tailored for filter-based LEO satellite clock and orbit estimation is introduced, explicitly accounting for bias propagation. Within the proposed framework, the effects of bias propagation on the PLs of LEO satellite clocks and orbits are examined. The approach is validated using Sentinel-3B onboard GNSS data together with real-time products from the National Centre for Space Studies (CNES) in France. Key contributors to the PLs—including the observation noise, multipath and biases in the GNSS real-time products are systematically assessed. Results indicate that bias propagation is the dominant factor influencing the PLs. Under an integrity risk of 1 × 10⁻⁵, increasing the overbounding Standard Deviations (STDs) of the observation noise and GNSS real-time SISREs from 0.005 m to 0.05 m leads to only moderate growth in PLs, i.e., from nanoseconds to below 20 ns for the clock PLs, and from dm- to meter-level for the orbital PLs. In contrast, incorporating propagation and accumulation of overbounding biases of a few centimeters from real-time GNSS products can drive the clock PLs to exceed 50 ns.

The emergence of Low Earth Orbit (LEO) constellations represents a paradigm shift in Global Navigation Satellite System (GNSS) development, promising unprecedented signal power and enhanced geometry for improved Position, Navigation, and Timing (PNT) services. Xona Space Systems is pioneering this domain with its Pulsar LEO GNSS. This paper presents the first comprehensive, independent analysis of the navigation signals broadcast by the pathfinding satellite, Pulsar-0, and provides a direct performance comparison against established Medium Earth Orbit (MEO) GNSS.

Data collection is performed using a newly developed, multi-frequency GNSS receiver specifically configured to track the unique signal structures of the Pulsar system. The receiver successfully acquires and tracks both the X1 and X5 navigation signals transmitted by Pulsar-0, analyzing both their pilot and data components across multiple satellite passes.

The core of this research is a rigorous quantification of several critical Pulsar ranging and link metrics from long-term signal recordings. We first analyze the Carrier-to-Noise density ratio to objectively assess the fundamental signal strength advantage provided by the LEO altitude. Furthermore, we evaluate the tracking noise and inter-signal bias associated with the X1 and X5 signals to determine the resulting precision and stability of the ranging measurements. The reliability of data delivery is quantified through the measured Frame Error Rate (FER) of the navigation message data streams.

The findings provide crucial, independently verified data on the operational capabilities of the first dedicated LEO GNSS. By directly contrasting Pulsar’s measured performance metrics with the current state-of-the-art GNSS, this work offers essential insights into the potential, challenges, and architectural implications of integrating LEO constellations into the future global PNT landscape.

LEO constellations for navigation is a major axis of research in the GNSS domain. Many investigations are done to demonstrate the benefits of such constellations for the user. In the same time, LEO constellations are also considered as an opportunity to investigate new frequency bands. While the UHF/VHF bands are studied for indoor applications, the Ku/Ka band, denoted FR2, are also of interest for the community (Starlink, considered as a possible signal of opportunity for the navigation, is emitting in the Ku band). Even if the future signals are expected to be similar to the actual ones, to change the constellation altitude and to increase the value of the carrier frequency is not trivial for the current GNSS receivers that are designed to process MEO L-band signals.

The present paper proposes an investigation of the different challenges occurring when processing a LEO signal emitted in the FR2 band with a current generic GNSS receiver. After a discussion about the possible LEO constellation configurations between 600Km and 1200Km, the document focuses on :
- The propagation losses in the FR2 band : based on the ITU models of propagation and the power flux limitations in the band, it is demonstrated that having a C/N0 on ground high enough to allow the tracking of the signal with a passive antenna, is not possible.
- The strong Doppler dynamics : The combination of the LEO altitude and the high carrier frequency induces a strong Doppler dynamics. Compared to a MEO GNSS signal, the Doppler Rate is very high and the Doppler Acceleration is not null. These strong dynamics can damage the performances of classical GNSS receivers. Indeed, acquisition performances losses and frequency estimation biases can be observed under a high Doppler rate. For the tracking, it has been demonstrated that the Doppler rate and the Doppler acceleration are responsible for significant phase biases which can damage the receiver stability. Moreover, additional works done in this study show that it is also possible to observe correlation losses which are quantified through analytical expressions.
- Strong clock noise : The clock contributioni n the FR2 band is then very important. An empirical analysis show that the well-known expression for the clock noise is too optimistic when considering GNSS tracking loops, especially in the FR2 band. Based on the classical stability criterion , it is explained that a specific configuration is needed for the stability of the receiver.

Based on the previous analyses, the second part of the paper is dedicated to the different receiver updates needed to have a nominal tracking of FR2 LEO signals :
- To have an active antenna which ensure standard C/N0 values for the baseband processing.
- To update the acquisition step in order to deal with the Doppler Rate.
- To update the order of the loop filter and to use a quadratic phase NCO

To validate the previous theory, realistic GNSS signals are simulated and processed with a software receiver.

The present study was supported by the CNES.

Low Earth Orbit (LEO) mega-constellations such as Starlink present emerging opportunities for Signals of Opportunity (SoOP)–based Positioning, Navigation, and Timing (PNT). Compared to Medium Earth Orbit (MEO) GNSS, LEO signals offer higher received power and wider signal bandwidth, potentially enabling improved ranging precision. However, exploiting Starlink signals for opportunistic navigation using low-cost consumer grade antenna without parabolic reflectors results in significantly reduced carrier-to-noise density (C/N₀), typically in the range of 24-36 dB-Hz. In low C/N₀ conditions, noise masks the correlation peak needed for delay and Doppler estimation, severely degrading acquisition. Prior studies show that reliable Starlink acquisition and tracking typically requires C/N₀ above ~40 dB-Hz. However, no existing consumer-grade antenna with a small form factor, can provide the high directional gain required to work with Starlink SoOP signals with sufficiently good SNR. This creates a noticeable gap between the high-gain antennas needed for robust LEO PNT based on SoOP and the practical constraints of low-cost hardware components. To bridge this gap, advanced signal-processing techniques must be introduced into the receiver chain to compensate for the inherently low SNR and improve acquisition and tracking performance.

The proposed methodology, uses a denoising module as a preprocessing stage immediately before the acquisition block. This denoising module is based on a 1-D convolutional autoencoder that learns to reconstruct the deterministic synchronisation structure of the Starlink downlink waveform while suppressing additive noise components.

Following denoising, the signal is passed to the acquisition stage, which performs a two-dimensional search over delay and Doppler domains. The search proceeds in two phases: an initial coarse grid search to identify candidate signal presence, followed by a fine search to refine the estimates. A Constant False Alarm Rate detector is employed to maintain consistent detection performance across varying noise conditions. The delay-doppler estimates from acquisition are then passed to an Extended Kalman Filter (EKF)-based tracking loop, which continuously updates the code phase and Doppler frequency estimates. The tracked estimates are used for positioning performance.

To evaluate the proposed solution, synthetic waveforms are generated to simulate low-gain antenna reception conditions. These include realistic channel impairments such as carrier frequency offset, phase jitter, additive white Gaussian noise (AWGN), and six progressively noise-impaired versions of a recorded Starlink signal. Both degraded signals and their denoised counterparts are processed through identical acquisition and tracking stages. The autoencoder is trained using recorded Starlink downlink signals and synthetically generated waveforms.

Performance is evaluated through SNR improvement and Doppler and delay RMSE comparisons across noise levels. Acquisition and tracking accuracy are quantified using Doppler and delay residuals against reference trajectories derived from Two-Line Element data. Preliminary results show improved positioning performance with denoising, demonstrating 8–10 dB SNR gains and 30–40% reductions in Doppler RMSE for both acquisition and tracking, particularly at very low SNR levels representative of consumer-grade antenna configurations.

This research has been undertaken with support from Machine Learning Applied to Signals of Opportunity (MaLASO) project, funded under ESA’s NAVISP programme (NAVISP-EL1-072)

The growing need for resilient Positioning, Navigation and Timing (PNT) services has motivated the development of complementary layers to Global Navigation Satellite Systems (GNSS). LEO-PNT is one such emerging capability, with ESA’s Celeste mission preparing for the launch of its first two demonstrator satellites in early 2026. A critical requirement for LEO-PNT is accurate onboard timing with respect to GNSS time. Since smaller LEO satellites cannot carry high-stability atomic clocks, precise time synchronisation techniques are required to steer onboard oscillators towards the GNSS reference.

To do so, Celeste Pathfinder A’s payload includes an onboard receiver, receiving GNSS signals which are processed by the Orbit Determination and Time Synchronisation (ODTS) System to estimate GNSS time. A navigation unit generates an internal time in the Signal Generation Unit (SGU) based on the frequency of the oven-controlled crystal oscillators (OCXOs) onboard. These two signals are compared in the ODTS, where a0 (s) and a1 (s/s) offsets are determined and sent to the Clock Management and Control System (CMCS). A steering correction, corresponding to the offsets, is applied to the internal clock. This steered time is then fed back into the SGU for the next steering loop. Therefore, an accurate onboard time is necessary for accurate signal generation and a long term stable internal clock.

This work investigates and compares several steering techniques for achieving a well-below-microsecond timing accuracy using OCXOs. Three types of algorithms are evaluated: feed-forward linear prediction model, PID-based control loops, and Kalman filter approaches. Representative models of OCXO and GNSS time are generated to evaluate the efficiency and accuracy of the steering methods. Subsequently, real GNSS signals from Sentinel-6A are used to replace the GNSS time model and provide a more representative performance assessment.

The results show that the proposed steering methods can achieve timing phase offset in the range of 1e-8 s and stability in the range of 1e-13 ADEV, exceeding current requirements for LEO-PNT demonstrators and indicating that highly precise synchronisation is feasible even with lower-grade oscillators for future missions. The study concludes with an outlook towards hardware implementation: the selected steering solution will be deployed on an ESA testbed, using simulated GNSS signals and low SWAP (size, weight and power) OCXO hardware, to further validate its performance for future LEO-PNT satellites. This capability will contribute to delivering robust, high-performance, and resilient PNT services for Europe.