[evlatests] Progress on WIDAR Imaging Woes

[Rick Perley]

    Considerable progress has been made on identifying our problem.  

    I'm using the Sunday test database, where Michael observed our 
favorite northern calibrator in five different positions:
On-Axis, and offset by 5, 10, 15 and 20 arcminutes.   The test lasted 
one hour, and was conducted at L-band, with the four sub-bands chosen to 
be relatively RFI-free. 

    The quality of the data, as viewed by (say) SPFLG, is excellent.  
All the periodic, large amplitude wobbles (of typical periods a few 
seconds) which littered our previous experiments at L-band are gone.   
(I take this as strong evidence that those wobbles were of solar origin, 
as a known sunspot group was present then, and has rotated to be out of 
view now). 

    Bandpass and gain calibration was done in standard ways.  On-axis 
phase and amplitude stability were *excellent*.    If all were operating 
as they should, we should expect excellent images of the off-axis 
observations. 

    But they are terrible. 

    I used a single channel (500 kHz) from the center of the band, to 
further diagnose the problems.  The visibility amplitudes of this 
channel, for each of the five positions, give exactly the expected 
amplitudes.  The antennas were indeed pointed where Michael told them to 
go. 

    However, the images of the 5 positions give a very different 
result:  The 'on axis' image has the correct flux at the center, but the 
image is of poor quality, with the background sources barely visible.  
The images of the four offset positions all show far too little flux 
density -- but the positions of the peaks on each are correct.  This is 
very clear evidence that *most* of the baselines which go into the image 
are correct in amplitude and phase, and that a subset are not.  

    To find which ones are incorrect, I extracted a single channel (the 
center channel) data for the observations of the 15 arcminute offset 
source -- the flux density is at half the on-beam value.  I then
ran UVSUB on this single-channel, removing the known flux density from 
the known position.  One can then play with this differential 
database.    The remarkable results are below:

    1) All visibilities amongst the 8 lowest numbered antennas are 
correct -- the residuals amplitudes are in the noise. 

    2) All visibilities connected to antennas 25 and 28 are wrong.  But 
they are not randomly wrong ... (read on ...)

    2a)  The differential amplitudes of all baselines connected to 
either 25 or 28 are either equal to the unsubtracted flux, or double the 
unsubtracted (i.e. the original and correct) flux.  There are far more 
of the latter than the former. 

    2b) Plotting the real vs. the imaginary parts yields more remarkable 
results.  The differentials fill two rings in the complex plane, 
corresponding to the two different amplitudes noted above.  Plotting 
individual baselines in the complex plane shows that each the 
indidividual data points from each scan cluster at a particular location 
in the two rings, but each scan's cluster is in a different place.  That 
is, the error either changes slowly in time, or from scan to scan.  
(There are five scans at each offset position, separated by about 12 
minutes).    For some of the shorter baselines, the 'clusters' clearly 
move around the ring in a orderly fashion.    It is interesting that for 
baseline 25 x 28, the clusters are oriented vertically --- that is, the 
real part of the error is zero, the imaginary part is found in one of 
four different places -- equal to the true amplitude, or its negated 
value, or double the true amplitude, or its negated value. 

    I believe that these characteristics are sufficient to explain all 
the imaging characteristics that we have seen, at all bands.  What has 
not yet been discovered is how this problem is generated.  George has 
confirmed that CASA sees the same effects.  He has already shown that by 
removing all data from these two antennas, the images are vastly 
improved, and background sources are properly showing up.