USNO IGS Mean Time Scale

Data from daily USNO GIPSY processing of IGS station data produces among many outputs clock differences from the remote IGS station clock and the reference clock used in the processing. At USNO we use USNO(MC3) as the TurboRogue frequency reference. In a geodetic optimized solution unmodelled errors are accumulated in the clock differences. These errors remain fairly small (< 1 ns in time in general) between the daily solutions. Our interest lies in the fact that there about 30 hydrogen masers and about 20 cesium clocks in the IGS network. What can be done with the geodetic optimized clock data?

This page will describe the formation of a simple mean time scale. In practice clocks drop in an out of GIPSY solutions for various reasons such as data being received too late to be included in the processing, for example. Time-keepers idealy like to have a stream of continuous, well-sampled clock data in order to characterize measurement noise, clock performance, and any long-term environmental or randomly induced systematics. Work is currently in progress here at USNO to have an independent solution optimized for the clock differences. Until this work is done however, we are forced to model the clocks as best we can from the geodetic optimized solutions.

The USNO rapid GIPSY solutions are performed once per day and model a 27 hour set of data. The last 3 hours of the each days solution are overlapped by the following days solution. The purpose of the overlap period is to allow the Kalman filters to refine the initial parameter estimates.

                        DAY 1
              h          12h          h
      ...  XXX0-----------.-----------0
                                                DAY 2
                                      h          12h          h
                                   XXX0-----------.-----------0
                                                                        DAY 3
                                                              h          12h          h
                                                           XXX0-----------.-----------0

Figure 1. A simple graphic showing the filtered solution startup overlap regions as "XXX" which are removed before further clock data processing. One character width indicates one hour.

The following shows the results over an interval of about 200 days. First each clock had filtered the 3 hour overlap data from the start up regions, see figure 1 above. Next fractional frequencies were generated from the 7.5 minute sampled clock solutions. Following that simple one day frequency averages on a clock-by-clock basis were generated. The next step was to model/detrend any significant frequency trends. At the model/detrend stage clocks were rejected or accepted using standard performance criteria, i.e. try to only keep the best performing clocks! Figure 2 shows the daily ensemble frequency averages from all the available individual detrended clock frequencies. Figure 3 shows the daily ensemble average frequencies integrated into time. Figure 4 displays the number of clocks contributing to the average frequency.

Figure 2. The average detrended daily frequencies formed from the GIPSY solutions and using the "best" IGS network hydrogen masers measured againt USNO(MC3). RMS 7.0e-15.

Figure 3. Time scale integrated from the ensemble-average of detrended frequencies. RMS = 4.2 ns. Peak-to-peak variation 17.8 nanoseconds.

Figure 4. The number of hydrogen masers included in each daily ensemble frequency average.

This page is http://tycho.usno.navy.mil/gpscp_mn.html
(27 December 1998)

Back to GPS CP home page.

Back to Time Service Home Page.