GLONASS Clock Combination

Processing of GLONASS data has always been more complex than GPS due to frequency division multiple access (FDMA). As a consequence of inter-frequency code biases, satellite clock corrections from different analysis centers are often offset by a few nanoseconds. Since accounting for such offsets is not really problematic, why has the IGS never produced a GLONASS clock combination? From personal experience, I think that I know the answer to this question and I agree with the IGS decision. However, authors of a paper published this week seem to think otherwise. Who’s right?


I had always been puzzled by the lack of a GLONASS clock combination by the IGS. From a theoretical point of view, the only difference with respect to GPS is to account for inter-frequency code biases (IFCB) playing a role in the datum definition of each clock. A first GLONASS clock combination taking IFCB into account has been analyzed by Song et al. (2014). The authors investigate the impact of these biases and especially the day-boundary jumps caused by different IFCB datum between days. Their GLONASS clock combination seems to work satisfactorily and no specific problems are reported. Strange.


Let us now look at my own attempts at combining GLONASS clocks. The figure below shows a comparison between the EMX (from NRCan) and the ESA clocks on June 26 2016 (GPSW 1903, day 0). It displays the EMX clock residuals with respect to the ESA clocks (held fixed) after accounting for radial orbit differences, timing offsets and a constant shift for each satellite. Each satellite is displayed with a different color.

Fig 1 Clock residuals between the EMX and ESA GLONASS clock products on June 26 2016


The first issue we can notice is not unique to GLONASS and consists of eclipsing satellites. We can clearly see peaks for all satellites of an orbital plane, which are caused by the wind-up effect as satellites spin to maintain a proper orientation. The EMX clocks, computed using Bernese, are not modeling yaw maneuvers while the ESA clocks are, leading to these inconsistencies. However, this is not the issue I wanted to focus on.


The second issue is what manifests itself as a drift for certain satellites, sometimes exceeding 10 cm over a day. These drifts are not unique to the comparison of EMX and ESA products and can be observed for other analysis centers as well. Our orbit specialist at NRCan, Yves Mireault, had a look at this problem. When computing satellite clock corrections with Bernese using the ESA orbit products, the drifts disappeared. This finding obviously implies that the orbits are the culprit. After a few additional tune ups to his GLONASS orbit estimation strategy, Yves has been able to reduce the drifts but not completely eliminate them.


Surprisingly, in the positioning domain, both satellite clock products lead to similar performance. This can be explained by floating ambiguities which can somehow soak up this slow drift which does not exceed a couple of centimeters over a satellite pass.


In my opinion, there is no use in combining GLONASS clocks before this inconsistency is addressed. A clock combination product can only be of higher quality than the original solutions if all solutions are consistent. Nevertheless, the authors of a paper published this week (Chen et al. 2017) proposed a solution to this problem, which consists of estimating not only offsets to accommodate IFCB, but also DRIFTS for each satellite! From a practical point of view, it is true that estimating a drift would account for most of the inconsistencies illustrated in the previous figure and could allow for a better clock combination. However, I think this is a “band-aid” solution that does not address the core of the problem.


The real solution is not as simple: we need to spend more efforts into improving GLONASS orbit determination. While GLONASS ambiguity resolution could help, it is likely that other factors such as solar radiation pressure modeling also require additional investigations.



Chen K, Xu T, Yang Y (2017) Robust combination of IGS analysis center GLONASS clocks. GPS Solut doi:10.1007/s10291-017-0610-0


Song W, Yi W, Lou Y, Shi S, Yao Y, Liu Y, Mao Y, Xiang Y (2014) Impact of GLONASS pseudorange inter-channel biases on satellite clock corrections. GPS Solut doi:10.1007/s10291-014-0371-y

Write a comment

Comments: 2
  • #1

    Tim Springer (Saturday, 25 February 2017 15:50)

    Very interesting! The drift you observe does match with the fact that we see a systematic bias in our along-track orbit overlaps. This was about 60mm but with switch to ITRF14 has been reduced to 40mm. This change is most likely thanks to the changes in the satellite PCOs in the IGS14. We know we can remove the bias by adjusting the PCOs. But we are not sure if the PCOs are really the cause. May just as well be the orbit model giving rise to a erroneous orbit scale.... Earth Albedo and IR, and power thrust still have large error bars (and not only for GLONASS)

  • #2

    Simon Banville (Saturday, 25 February 2017 18:16)

    @Tim Thanks for these great insights. I will have to compare more recent GLONASS clocks computed in ITRF14 to see if these drifts are at least reduced!