[IGSMAIL-3615]: IGS time scale

[Ken Senior]

******************************************************************************
IGS Electronic Mail      28 Nov 13:36:43 PST 2001      Message Number 3615
******************************************************************************

Author: Ken Senior

Dear Colleagues,

With reference to the new clock products announced in IGS Mail #3470, we
have made further revisions to the algorithm used to generate the new
internal time scale (IGST).  This time scale has been proposed as the
basis for realigning the IGS combined clock products as a way to overcome
large instabilities in the current operational time scale tightly aligned
to broadcast GPS time.

The refinements include a number of bug fixes, as well as changes to the
steering algorithm responsible for ensuring that the scale is loosely
aligned to GPS time.  In addition, a new Rapid time scale (IGRT) has now
been generated and used to realign the IGS Rapid combined clock estimates.
Both the revised Final IGST as well as the Rapid IGRT products are
available at
  http://clockdev.usno.navy.mil/igst   and
  ftp://clockdev.usno.navy.mil/igst

The major changes that have now been implemented are:

* propagation in the filter of more realistic data weights, which are the
  RSS combination of the formal errors of the raw combined clock estimates
  (based on the agreement among the submitted observational results from
  the Analysis Centers) together with an empirical measure of the stability
  of each clock at 5-minute intervals (determined adaptively by the filter).
  This was necessary as it is often the case that the formal errors are
  unrealistically small, though they are useful to identify bad data.
* the above enhancement allowed for an improvement in outlier and rate-break
  detection using the pre-fit residuals in the filter instead of the more
  ad hoc method described in our FCS 2001 Proceedings paper describing the
  algorithm.
* the LQG steering algorithm was changed from a 2-state (phase & frequency)
  filter to a 3-state (phase, frequency, & frequency aging) filter whose
  gain is specified by G = [6e-6 ns, 2e-5 ns/day, 4e-4 ns/day^2].  In
  particular, this change enables the filter to deal with a large excursion
  of the unsteered scale during April 2000 and which the previous steering
  method could not handle.  Generally, these gain values lead to maximum
  instabilities of 2e-15 at 1 day, except on occasions (e.g., April 2000)
  when problems in the raw combined clock estimates cause large excursions
  of the unsteered scale (see more below). 
* the production of the scale is now updated automatically, searching the
  IGS product archive hourly for new Rapid or Final combined clocks.
* now included in the set of output results and plots is a cumulative plot
  of both IGST and IGRT versus UTC, which is determined using BIPM Circular T
  values for (UTC - GPS time).

Users should be aware that there have been periods during which the IGS
combined clocks have included relatively few stations equipped with H-masers.
In some cases, only 15-minute clock estimates are available instead of the
IGS standard of 5 minutes.  GPS weeks 1110, 1111, 1112, and the recent 1141
are examples; week 1141 is a particularly bad week as there were only 2
masers present on 1 day of the Rapids.  These circumstances severely limit
the internal ensemble method we have implemented.  In addition, the current
algorithm cannot utilize 15-minute clock data as this would make tedious the
treatment of the linearized quadratic term in the design matrix (equations
1 and 2 in the FCS 2001 Proceedings paper).  We believe that the new
steering gain values handle such excursions, but this is of course at the
expense of short-term stability.

We plan to fully document the updated algorithm in the near future, but
meanwhile these updated clocks products are offered for further evaluation
in a test mode.

Regards,
Ken Senior

Time Service Department--Clock Development
U.S. Naval Observatory