Defects Index | Documentation Table of Contents
98/04/07 DEFECT in program page4
SYMPTOM: Abnormally large a priori residuals with no corresponding problem
appearing in the post-fit residuals for 98_087. (see note below)
PROBLEM: page4 attempts to remove any large a priori biases in the phase
ambiguities so that the estimated adjustment to the ambiguity will
be near zero. Although partials used in ambiguity estimation are
linear and no direct penalty is incurred from large estimated
adjustments, a secondary penalty is incurred because of loss of
precision in the sum-of-squares variable, bw or "basement window",
which is manipulated with the normal matrix processing. Because
the bw has become increasing important in processing, statistics
and quality control, every effort is made to keep phase ambiguity
estimates near zero. However, circumstances can arise where the
consistent (over all satellites) determination of a priori phase
biases to be impossible without backtracking in the data to
remove these biases. A conscious decision has been made and
kept to keep page4 from moving/processing backwards in the
data therefore an logic hole was created, i.e. periods can
occur when no a priori phase biases will be removed. Apparently,
as anticipated, this has been a rare event but with the advent
of completely automated orbit processing, this, albeit rare,
event can no longer be permitted. Two defenses in the program
are possible: 1) enable back-tracking through the data; 2) omit
the offending data.
CORRECTION: In obsdd.f, code was inserted so that observations, for which
consistent a priori phase bias removal is impossible, will be
omitted.
FOUND BY: Kass and Dulaney 98/03/30
FIXED BY Schenewerk 98/04/01
VERSION: 9804.06
SOURCE: /g1/HPUX.10/Src/Page4
EXECUTABLE: /g1/HPUX.10/Bin
NOTES: This is related to the defect reported on 980401.
This is an inherently dangerous strategy in that blocks of
otherwise good data up to the time span specified by the "data
gap" variable could be omitted (in the case of 98_087, two
epochs on the FORT-ASC1 baseline were deleted). Because of
the anticipated rarity and short duration of these events,
this is considered an acceptable risk.
980406.page4
June 25, 1999
Steve Hilla