[evlatests] AG730 data reduction and EVLA performance

[Jim Ulvestad]

Hi all,

AG730 was a very simple continuum observation, made in two
days on November 24/25 and 25/26, 10 hours each day.
It was all X band, default frequencies at 50 MHz bandwidth,
with no frequency or setup changes at any time.
The observation sequence was extremely simple, with a couple
observations of a flux calibrator, and the rest of the run spent
alternating between observations of a single calibrator and a
target source, with a total cycle time of about 8 minutes.
Six EVLA antennas participated: 13, 14, 16, 18, 24, 26.
This was purely a weak-source detection observation, with
no source in the field strong enough to self-calibrate.  Thus,
the target observations relied totally on the observations of
the calibrator, and any subtle VLA-EVLA problems cannot
be eliminated by looking for phase jumps or any other data
characteristics on the source.  For a noise-dominated source,
dropouts are not apparent, but any observations that have
extremely high amplitude for some reason are noticeable.

There were three apparent EVLA-VLA phase jumps over the
course of two days, two of about 180 deg. and one that
looks like around 110 deg.   They can be found by
looking at successive calibrator observations, but there
is no obvious way to tell whether they happened at
the end or start of a calibrator scan, end or start of
the intervening source scan, or somewhere in the middle
of the target source observation.

The six EVLA antennas all had apparent phase drifts of
about a cycle or so over the course of a 10-hr observation.
The attached postscript plot shows the plots of the six EVLA
antennas in between two VLA antennas.  Presumably, the
EVLA drifts are due to position errors.  The phase jumps
also are noticeable in the plots.

On the first day of the observation, there were occasional
scans where the EVLA antennas dropped dramatically in
the last 10-second record on the calibrator, then showed
up 10 seconds early in the succeeding record on the
target source.  See attachment listr.txt for an example.
This seemed to happen every couple hours, and it's not
at the moment obvious whether the reverse also was true
(a bad last record on the target, and an early record on the
calibrator).  From the frequency of occurrence, if I missed
a few occurrences on the target, one could almost imagine
that this was related to hiccups when the LST vs. IAT
slipped across a 10-second barrier, which ought to happen
once per hour.

And on the second day of the observation, every scan began
with a 10-second record containing only EVLA antennas,
which was followed with a 10-second record containing only
VLA antennas, which was followed by a third 10-second
record that finally had all the antennas in it.  An example of
this is also in the attached listr.txt .

And finally, the bottom line.  I did all the flagging I could
find, flagged the EVLA-VLA baselines on the target in
between the calibrator scans when the phase appeared to jump,
QUACKED the EVLA antennas on all the first scans on
the second day, flagged all the bogus calibrator records when
the EVLA antennas dropped, and flagged a few other low EVLA
points.  Then I went and imaged my source, with about 15 hours
on source over 2 days, and got an rms of 5 uJy/beam with no
detection down to 2-sigma. 

BUT, AS A TEST, I DELETED THE EVLA ANTENNAS
COMPLETELY AND RE-IMAGED.  This gave me a higher
rms of 6.75 uJy/beam as expected, AND A 4.3-SIGMA DETECTION.
THEN I DELETED ONLY THE EVLA-VLA BASELINES,
KEEPING VLA-VLA AND EVLA-EVLA.  The resulting
image had an rms of 6.55 uJy/beam and a 3.3-sigma detection.

I am pretty convinced that the VLA detection is good, since
I have divided it into the two separate days with 3-sigma
detections on each day, and into the two separate IFs with
3-sigma detections in each IF.  These nicely average together
to give me the 4.2-sigma detection with all the data. 
But the EVLA antennas make the detection go away,
although they don't make the strongest source in the field
(1.0 mJy) go away.

My present belief is that there is a subtle problem with the EVLA
antennas that cannot be seen in the noise-limited data, perhaps
phase jumps or other issues, that redistribute noise at the few-sigma
level and so wipe out the central source detection without materially
affecting the detection of the 200-sigma source.  I believe that the
standard phase calibration should be taking out the baseline error
adequately, though I haven't computed the possible effect.  And there's
likely something else that a smart astronomer can think of, but that
eludes me in this moment of holiday cheer.

I will plan to show these results at the EVLA test meeting on Jan. 2;
I won't be here on Dec. 27 if there is an EVLA meeting,
and there probably aren't many people around to
help troubleshoot anyway.

Jim
-------------- next part --------------
A non-text attachment was scrubbed...
Name: AG730-PHASE.PS
Type: application/postscript
Size: 121018 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20061226/6879685e/attachment.ai>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: listr.txt
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20061226/6879685e/attachment.txt>