[evlatests] Hi-Fidelity Polarimetry, the Band Switches, and 'Delay Clunking'.

[Rick Perley]

     Polarimetric imaging of strongly polarized calibrators, such as 
3C286, have long shown problems in the dynamic range of the Stokes 'Q' 
and 'U' images.  This has been particularly evident at the higher 
frequency bands, but until recently, I've not looked deeply into the 
problem.  Analysis of the images suggested that, for some scans, the 
amplitudes of the Stokes visibilities Q and U were incorrect by small 
amounts (~10%).

     More careful investigation showed that the issue was not in the 
amplitudes of the (RL) and (LR) visibilities, but in their phases. Most 
interestingly, if the (RL) phase was erroneously high by (say) 10 
degrees, the (LR) phase was low by the same amount.  When the Q and U 
visibilities were formed, the Q and U visibilities had the correct 
phases, but incorrect amplitudes.

     Eric Greisen wrote a little AIPS task (RLCAL) which greatly helped 
the diagnosis -- essentially a polarization  'self-cal', which solved 
for R-L phase error in the data, given a polarization model.  This 
clearly showed that there was an R-L phase error affecting the data 
within some scans.  Typical errors are not large -- 10 degrees -- but 
more than enough to ruin a high-fidelity polarization image.

     The (R-L) phase solutions showed that all antennas were affected 
identically, which tells us right away that it's a property of the phase 
reference antenna.  For the database (at Ku-band) that I was using, the 
reference antenna was ea10 -- chosen because it was the closest to the 
array center.  So what's special about this antenna?

     The answer came in reviewing the (PDif) switched power.   Every 
scan for which we found an (R-L) phase error, also shows the LCP side 
(usually) with switched power low by ~ 25%.  In this experiment, I was 
changing bands every ~25 seconds.  Upon return to Ku-band, some switch 
was not at 100%, and this is apparently accompanied by a small change in 
phase to the affected polarization.  The amplitude loss is corrected by 
application of the switched power (or by self-calibration).  But we have 
(as yet) no regular procedure for correcting the phase change *of the 
reference antenna*.  (A phase change on a non-reference antenna will be 
corrected by self-calibration).

     So what's a user to do to ensure the best possible polarization 
calibration?  Choose a reference antenna that doesn't have this 
problem!  I think the best way to do this is to review the switched 
power.  Antennas with temporally smooth switched power (PDif) appear to 
not have this phase problem. I tested this hypothesis by calibrating my 
data with ea12 as reference.  This antenna has a completely smooth PDif 
plot in both polarizations.    And sure enough, the polarization 
solutions were stable, with no serious issues with the (R-L) phase.

     The switches are not the only source of R-L phase changes -- the 
'delay clunking' provides a similar effect of similar magnitude -- but 
it is normally not so evident since it operates on a much faster 
timescale (seconds, or shorter) and quickly averages out.

     I will show the evidence at the upcoming Thursday tests meeting 
(10AM, Rm 317).  We will also discuss the origins and fixes -- including 
the annoying 'delay clunking'.