[evlatests] Polarization Results from 3C273/3C286 database

[Rick Perley]

    I've looked at the polarization results from the long observation of 
3C273 and 3C286 at X-band -- and there are some very curious things 
going on ...

    The observation was  about 7 hours in duration, mostly on 3C273, 
taken in polarization spectral-line mode:  BW = 12.5, Nchan = 16 per 
correlation.  After careful flagging, and generated a 'channel 0' 
single-frequency database by vector summation of channels 3 through 10.  
Time averaging was left at 3.3 seconds, bandpass solutions were applied. 

    This database was then calibrated for polarization by standard 
means.  I used 3C286 (essentially unresolved in polarization and total 
intensity), to generate the solutions, and rotated the phase frame to 
ensure the polarization position angle was at 33 degrees -- the 'known' 
absolute value.   There were 7 observations of 3C286, five before 
meridian transit, and two after. 

    An image of 3C286, in Stokes Q and U, showed good images, but much 
noisier than expected.  It was quickly found that antenna 14 showed 
higher polarization visibilities than it should.  Removing this antenna 
improved the images, but still the noise is much higher than it should 
be -- and is higher than in Stokes 'I'. 

    More careful review showed that both antennas 14 and 15 -- even 
after polarization calibration -- give large offsets in polarization 
amplitude and phase, against all other antennas.  The effects are 
particularly noticeable in phase:  In Stokes 'Q' (correct amplitude is 
235 mJy, and phase should be zero), well-behaved antennas show 
amplitudes deviating not more than 10 mJy (these are scan averages), and 
5 degrees.  Antennas 14 and 15 give amplitudes up to 350 mJy -- note 
that the error is always positive -- there are no baselines against 14 
or 15 with too low a polarized flux!  In phase, the patterns are most 
interesting -- values of + or - 20 degrees are seen on nearly all 
antennas.  There are no values in the +/- 5 degree range -- it's always 
near +/-20!  The amplitude offsets are nearly constant in time, the 
phases are not -- becoming much larger near meridian transit.  The sign 
of the phase offset is correlated perfectly with DCS number -- positive 
offsets have a lower DCS, negative one are higher.   This is caused by 
the VLA's phase sign convention -- there is no deeper physical meaning. 

    Although antennas 14 and 15 (DCS numbers 16 and 13) are the worst 
offenders by far, others should similar patterns, particularly antennas 
7, 16, 19, 22 23 and 24, and mostly notably near meridian transit. 

    These troubles are clearly seen in the observations of 3C273 as 
well, although because this source is heavily resolved in Stokes Q and U 
(and the powerful point source in stokes 'I' is reduced to a few hundred 
mJy in polarization), the Q and U visibilities are modulated 
considerably by the jet's visibility.  But despite this, it's obvious 
that baselines to 14 and 15 are not correct in the cross-hand 
visibilities. 

    I have no idea of what is wrong here -- but clearly the data and the 
model used to separate antenna from source polarization are not in 
agreement.   This is not an EVLA issue -- both VLA and EVLA antennas are 
equally in error.