[IGSMAIL-2168] IGS combination changes

[Tim]

******************************************************************************
IGS Electronic Mail      Thu Feb 18  9:43:39 PST 1999      Message Number 2168
******************************************************************************

Author: Tim Springer
Subject: IGS combination changes

Subject: IGS combination changes


Dear IGS colleagues,

Based on the recommendations of the Darmstadt IGS Analysis Center
workshop some small changes have been made to the IGS combination 
software. These changes have taken effect starting with the 
IGS final products of week 994. For the IGS rapid products the 
changes took effect on GPS week 996 day 4.
The changes are mainly based on recommendation 1 of the first 
position paper of the Darmstadt workshop. 

One part of this recommendation was: 
"To avoid any distortions from missing satellites, in some of the
AC solutions, the combination, and its statistics, should be based on the
common satellites only!"

This has now been put into effect. Satellites which are excluded from
the combination and its statistics are marked by an "X" in front of the
satellite PRN number. Of course these satellites are kept for the IGS
combined orbit. They are combined like all other satellites using the
weights of the individual centers and the estimated transformation 
parameters. The only difference is that they did not contribute to the
weights nor to the transformation parameters.

It should be noted here that the down-weighting of "bad" satellites,
as also recommended, has not been implemented because it was found
that there is little point in a "weighted" L1-norm; contrary to 
weighted least squares (L2-norm). 

Additionally some smaller changes were made to help the reliability and
automation of the combination software. The most important change was
a new outlier detection method. Within the combination the outlier
level is now determined by computing the rms over all satellite and AC
residuals (only satellites common to all ACs) and multiplying this
by 4.7. This factor of 4.7 is based on Chebyshev inequality, which
states that at 95% probability  level, for the worst possible error 
distribution the sigma multiplier is 4.7.

Typically the outlier level for the final is now around the 30-50 cm 
whereas for the rapid it is around the 50-90 cm.
A similar procedure is used for automatic detection of outliers in the 
transformation parameters.


Kind regards,
         Tim Springer


 -----------------------------------------------------------------------------
 Tim Springer                      Astronomical Institute, University of Berne 
 springer at aiub.unibe.ch            Sidlerstrasse 5, CH-3012 Bern, Switzerland  
 http://www.cx.unibe.ch/aiub/      Tel: +41 31 6318592    Fax: +41 31 6313869

 Anonymous ftp access:
 ---------------------
 For CODE products:      anonymous ftp: ubeclu.unibe.ch  -or-  130.92.6.17
 (VMS system)            directory    : aiub$ftp

 For IGS Rapid products: anonymous ftp: ftp.unibe.ch     -or-  130.92.6.40
 (UNIX system)           directory    : aiub/prelim_igs
 -----------------------------------------------------------------------------


[Mailed From: Tim Springer <tim.springer at aiub.unibe.ch>]