A new approach to improve satellite clock estimates, removing the inter-day jumps

Time synchronization is one of the main applications of Global Navigation Satellite System (GNSS). Indeed, GNSS satellites are equipped with highly stable atomic clocks, which can be used as clock references. In particular, Galileo constellation satellites are equipped with Rubidium and Hydrogen-maser frequency standards that have a long-term (1-day) stability at the level of 10-14 for intervals of 104 s. However, in order to use these clocks as references, their time offsets must be estimated from measurements collected on ground. In this sense, the effectiveness of these references depends on the accuracy and stability of such estimates. For that reason, the International GNSS Service (IGS) is providing satellite solutions with a nominal accuracy of 75 ps (i.e. better than 3 cm). The clock accuracy is assessed by evaluating the discontinuities over consecutive days, computed independently. These discontinuities appear because of the correlations between the different parameters, which are estimated jointly with the satellite clocks (e.g. receiver clocks or carrier phase ambiguities). Therefore, satellite clocks depend on the robustness of the different models used in the estimation process. For the Galileo satellites, the inter-day discontinuities are at the level of 0.5 ns. These large jumps are due to the minor number of Galileo satellites available than in GPS, which results in a reduced number of fixed carrier phase ambiguities (other solutions from other IGS analysis centers present similar features). In this work we present a refinement of the method for estimating satellite and receiver clocks, which is able to reduce these discontinuities and, consequently, to improve the accuracy of the satellite clock estimations (according to the above-mentioned IGS metric applied). Some of the main characteristics of this new approach are 1) the use of unambiguous carrier phases, 2) processing several constellations and several frequencies and 3) considering the temperature dependency of the phase biases.Postprint (published version