In recent months, some problems have been identified with the
calibration transfer data distributed with VLBA correlator data.  All of
these problems affect only small amounts of data, and they do not affect
the actual correlation results.  Final processing results can, however,
be adversely affected through the application of flagging and
calibration.  The issues are:

1.  Occasionally the correlator job generation program creates spurious
flags for the FG table with the reason "System idle".  This is generally
at the start of a job script and relates to the bookkeeping of what
flags are set.  It can cause major portions of a good scan to be
flagged.  That could be handled by deleting such flags, except while
that flag is set, no data are being passed to the calibration tables so
the data cannot be calibrated properly.  The problem occurs sufficiently
infrequently that not all projects are affected.

The cure, if excessive data are affected, is to use the old data path
for calibration using VLOG, UVFLG, ANTAB, and PCLOD after deleting the
FG, GA, TY, and PC tables.  The amount of affected data can be
determined by looking at the FG table for the reason "System idle" or
simply by running something like VPLOT with the flags on and off
(FLAGVER=-1) and looking for suspicious data loss with them on.
Remember that flags are on when FLAGVER=0.

2.  It is possible for the calibration transfer system to mis-label data
in the pulse cal (PC) table when a station that is in other correlator
jobs for the project are not in one of the jobs.  This can happen
especially with MK and SC which cannot see sources for the full time
that others are observing.  What has been seen is that the data for the
station that is missing is not simply thrown away, but rather is
ascribed to another station.  A plot of the pulse cal data with SNPLT or
a listing of the data will show twice as many calibration points as
should be there, and alternate rows will have different patterns of
relative phases between IFs or polarizations.  Inspection of the
patterns will show that half the rows belong to the missing station.
Running the pulse cal calibration (PCCOR) on such data would create
badly calibrated data.  Again, the workaround involves using VLOG and
PCLOD.  Or the offending points could be edited from the PC table.

3.  The time associated with pulse cal data points needs some
adjustments from the values in the monitor data base.  Data in the cal
file (VLOG input) has all known adjustments applied.  But from
calibration transfer, there can be a 10s offset for truncated
integrations - typically integrations less than 2 minutes long.  Since
the pulse cal values generally change slowly within a scan, this will
have very little affect on the data in practice and likely can be
ignored.  It is possible, but very unlikely, that it will also cause a
point to appear to be after the scan in which it was measured, which
could cause it not to be used in calibration.

All of these issues potentially affect data from the start of the use of
calibration transfer around late 1998, although the flag bug may be more
prevalent since the use of Mark5 began.  It is not clear when they will
be fixed as the detailed causes are not yet fully understood and the
people involved have higher priority work to do on the EVLA.

---------------------------------------------------------------------
    R. Craig Walker            Array Operations Center
    cwalker@nrao.edu           National Radio Astronomy Observatory
    Phone  505 835 7247        P. O. Box O
    Fax    505 835 7027        Socorro NM 87801   USA
---------------------------------------------------------------------