[evlatests] R-L Phase Stability

[Rick Perley]

    The '4M' project -- observations of 3C273 and L, S, C, and X bands, 
for the purpose of testing super-high dynamic range imaging algorithms, 
has revealed a significant problem with the R-L phase stability. 

    Taking X-band as the example case, mostly because it's clear of RFI 
and the nucleus is clean and well-separated, I get quite easily a few 
hundred thousand:1 dynamic range.  The polarization solutions looked 
fine, but when I made Q and U images, the results were very 
disappointing, with much scattered flux about the strong nuclear 
source.  This behavior has not been seen in old VLA data, so I looked a 
little more closely. 

    The problem is easily traced to poor R-L phase -- that is, the 
phases in the opposite hands are not appropriately coupled.  To do 
effective polarimetry, this differential phase must be *very* stable on 
*at least* one antenna.  A phase change between hands is equivalent to a 
real rotation on the sky!  A differential stability of better than 
1degree is certainly needed. 

    To ensure I'm not being fooled by errors in procedure or software, I 
extracted from the X-band data a 10 MHz-wide 'chunk', without any 
calibration (no delays, no bandpass, no phase referencing).  The R-L 
phases were then generated for the three sources (3C286, 3C273, and 
J1150-0023), which were then plotted.  It's not a pretty picture:

    A) Many antennas have multiple phase difference states between 
polarizations.  For antenna 1, the two states are 6 degrees apart. 
    B) Many antennas show separate scans within which the R-L phase 
differs from the others by 3 or 4 degrees, without any hint of pattern. 
    C) Some antennas show a slow drift, of a few degrees, over a few 
hours. 
    D) Some antennas have no preferred difference at all, but show 
values -- usually constant within a scan -- which are randomly 
distributed over a range of up to 15 degrees.  Notably bad are antennas 
22 and 23. 

    It is easy to show that a reasonable fraction of the problems noted 
above are likely related to our 'switch' issue.  The observations were 
taken by cycling through the four lowest bands.  Plots of the PDif show 
good correlation between bad difference phase and changes in PDif. 
    I suppose it's possible that differential phase changes on the 
antennas which don't have good correspondance between PDif changes and 
R-L phase changes may still be due to the switch -- can it be that the 
switch attenuation is all right, but there is a small phase change?

    In any event, this is a serious problem for accurate polarimetry.