1. Introduction
Global navigation satellite systems (GNSSs) currently provide positioning, navigation and timing (PNT) services to users in different areas. In certain challenging environments, GNSS signals may suffer from limited or even no visibility—e.g., in urban canyons. The signals from mega-constellation low Earth orbit (LEO) satellites, currently in developement with satellites continuously being launched, are nowadays becoming attractive for navigation purposes. Since these LEO satellites are primarily used for communications, to enable positioning, it is proposed that either additional payloads are added to these satellites to provide GNSS-like signals or employ their signals as signals of opportunity for positioning (SOOP) [
1]. Different from the GNSS satellites that are located at the medium Earth orbits (MEOs), the LEO satellites at lower latitudes of about 300 to 1500 km [
2] are able to provide ground users with much stronger signals, which enables positioning in GNSS-challenging areas and makes the signals more resilient against jamming [
3]. A large number of LEO satellites launched or to be launched in the near future by companies such as Iridium, Globalstar, SpaceX, OneWeb, Samsung and Boeing does not only benefit the overall satellite geometry and thus the position dilution of precision (PDOP), the fast-moving LEO satellites and the rapidly changing geometry can also reduce the long convergence time in the precise point positioning (PPP) [
4].
The concept of using LEO satellites for positioning is similar to that using GNSSs, which requires knowledge of the precise LEO satellite positions. The two-line-elements (TLEs) of the LEO satellites are published and updated daily by the North American Aerospace Defense Command (NORAD) [
5]. Their provided accuracy of several kilometers, however, cannot fulfil the requirement for high-precision positioning. Similar to the precise orbit determination (POD) procedure for the GNSS satellites, one could rely on a ground network receiving the LEO signals to determine their orbits. Attempts have been made by TRANSIT, and a 3D root mean square (RMS) of the orbital errors of several meters was achieved using four ground stations [
6,
7]. The orbital accuracy improves when increasing the network density—i.e., with the root mean square error (RMSE) reduced to the sub-meter to meter-level in the radial, along-track and cross-track directions using 15 ground stations [
8]. Benefiting from the GNSS receivers equipped onboard the LEO satellites, the signals from the higher GNSS satellites can be used for positioning the LEO satellites as users, and high-accuracy orbit determination becomes possible. Combining proper dynamic models with the GNSS observations in the reduced-dynamic orbit determination scheme, the onboard real-time LEO POD can achieve an accuracy of a few decimeters even when using the inaccurate GNSS broadcast ephemeris and considering the limited computational power available onboard [
9]. Using precise GNSS products and comprehensive dynamic models in the post-processing mode, the reduced-dynamic LEO orbits can typically reach an accuracy of centimeters [
10].
For different positioning techniques, the requirements on the orbital accuracy are different. For the single point positioning (SPP) using the pseudorange observations and the real-time kinematic (RTK) positioning with the orbital errors largely reduced by forming between-satellite differences, an orbital accuracy at meter-level could be sufficient. For PPP, however, which would benefit from the rapidly varying LEO satellite geometry, high-accuracy orbit at a few centimeters is the key to successful positioning. In this study, making use of the GNSS observations collected onboard the LEO satellites, and taking the needs of different methods (SPP, RTK and PPP) into consideration, the orbital products are proposed to be generated with two different levels of accuracy. At Level A, low-accuracy orbits are broadcast from LEO satellites to users in the form of ephemeris parameters. At Level B, high-accuracy orbital corrections with a high sampling rate would be transmitted from the ground processing center to users with low-order polynomials through Internet links.
Compared to the GNSS satellites, the LEO satellites experience more complicated dynamics due to their lower altitudes and, as a result, suffer from higher influences of the Earth’s gravitational field and the air drag [
11]. As shown in [
3], using the same set of the ephemeris parameters as the Global positioning system (GPS) legacy navigation (LNAV) message, an orbital user range error (OURE) induced by orbital fitting amounts to about 7 cm for the GPS satellites within a fitting interval of 4 h. For LEO satellites, however, this is only achievable within a much shorter fitting interval—i.e., from about 10 to 20 min. Additional and transformed ephemeris parameters are reported to be helpful to improve the OUREs [
11]. The ephemeris parameters, however, need to be fit to the predicted but not the precise orbits. As discussed in [
12], even based on good dynamic models, the prediction errors dramatically grow with the age of data (AOD) to several decimeters after one hour. The prediction interval that is closely related to the data uploading rate serves as the limiting factor for the OUREs. In this contribution, for Level A orbital products, the error budgets are analyzed in detail for the LEO broadcast ephemeris using real data of a typical LEO satellite, GRACE FO-1, which are subject to relatively strong influences of the Earth’s gravity field and the air drag. The study includes the infrastructure design, the analysis of the potential data gaps between subsequent satellite contacts, the prediction errors introduced by different dynamic models and for different prediction intervals, as well as the fitting errors of the ephemeris parameters at different fitting and prediction times. The detailed error budget is given for the Level A products in terms of OUREs and the 3D RMSE.
Similar to diverse GNSS real-time correction streams, e.g., provided by the International GNSS Service (IGS) real-time service (RTS) [
13,
14], precise Level B orbital corrections are computed based on the Level A products in this contribution and are supposed to be delivered to users via Internet links. The precise orbits can be extrapolated for tens of seconds corresponding to the latency between the reference time of the applied corrections and the time of processing [
14]. The differences between the extrapolated orbits and the Level A orbits are then fitted with polynomials of different degrees. The polynomial fitting is analyzed at different prediction and ephemeris fitting times of the Level A orbits, as different orbital characteristics may apply at different phases of the Level A orbits. The detailed error budgets of the Level B products are given, and the benefits of increasing the polynomial degrees are discussed.
In general, the overall objective of this work is to design and discuss the feasibility of possible procedures to generate the LEO orbital products at two accuracy levels—i.e., less accurate Level A products as a broadcast ephemeris type, and high-accuracy Level B products as corrections to broadcast ephemeris that can be transmitted via the Internet. The paper starts with the proposed infrastructure design and the processing procedures of products A and B. This is followed by the analysis of different error types of these two orbital products using real LEO data. The results are discussed and conclusions are given at the end.
4. Conclusions
With thousands of LEO satellites launched or planned to be launched in the near future, studies have been performed to utilize the LEO signals for navigation purposes. The increased satellite number with their improved signal strength, compared to GNSSs, and the rapidly changing geometry of the LEO mega-constellations, are shown to be beneficial to the positioning service on the ground.
One main condition to realize the LEO-based positioning is the knowledge of the orbital positions of the LEO satellites. Relying on the GNSS observations collected onboard the LEO satellites, this study proposes two procedures to provide the LEO orbital products. The first is with a relatively low accuracy at meter-level, i.e., the Level A products, and the second is high-accuracy LEO orbital products at centimeters—i.e., the Level B products. The orbits are proposed to be processed on the ground with good dynamic models, extrapolated to up to several hours and tens of seconds for the Level A and B products, respectively. For the Level A products, the long-term predicted orbits are supposed to be fitted into subsequent sets of LEO-specific ephemeris parameters, uplinked to the LEO satellites and then broadcast to users. Based on the broadcast ephemeris, low-order polynomials are generated to merge the differences between the Level A orbits and the short-term predicted precise orbits. The polynomial corrections and the time of reference, as the Level B products, are then transmitted to users via Internet links to achieve high orbital accuracy of centimeters.
In this study, the real data of a typical LEO satellite, GRACE FO-1, which resembles other LEO satellites that can be used for positioning, with a relatively low altitude of about 500 km, were used for the tests. For the Level A products, it was found that the prediction errors play the dominant role in the total error budget, and the prediction interval and the corresponding dynamic models used are the key factors to keep the prediction errors at an acceptable level. The LEO orbital configuration, the latitude and the network density of the ground monitoring stations are the limiting factors for the maximum time gaps between subsequent satellite contacts, and thus the limiting factors for the required orbital prediction period. In this study, a prediction interval of up to 6 h is proposed. With different options of the estimable dynamic parameters tested, it was found that estimating the stochastic velocity pulses every 2 h in addition to the Keplerian elements at the initial condition, as well as the constant and periodic accelerations, provides the best results for mid- and long-term prediction. The OURE of the Level A orbits amount to about 0.1, 0.2 and 1 m for prediction intervals of 1, 2 and 6 h, respectively. The ephemeris fitting errors are only around several centimeters and do not have significant influences in cases of the mid- and the long-term prediction.
The Level B products are extrapolated for a much shorter period—i.e., 60 s—in this study and the OURE of the total errors amounts to several centimeters. The prediction errors still play a major role in the total error budget. At the same time, it was found that increasing the polynomial degree from one to two could reduce the fitting OUREs by a few millimeters, while further increasing the degree value to three brings almost no benefits. In general, the Level B orbits could reach an accuracy of a few centimeters, and increasing the polynomial degree from one to two or three does not significantly influence the total error budget of the Level B products.
In summary, after overcoming the challenges related to the hardware infrastructure, the LEO orbital products can theoretically be provided to users with a sub-meter to meter-level accuracy in the Level A products as broadcast ephemeris, and with an accuracy of centimeters in the Level B products as high-accuracy high-rate corrections.