[evlatests] Polarization Tests -- all is well!

[Rick Perley]

    For those wanting only an 'executive summary': 

    An important error in FILLM has been found and repaired, the result 
of which is that the problems noted recently in polarization data are 
completely removed.  At this point, all seems quite perfect in EVLA 
polarimetry.   I also find that EVLA sensitivities are exactly as 
expected. 

    Details:

    I reported a couple weeks ago that a 'polarization spectral line' 
observation taken earlier in the month at X-band had shown peculiar 
amplitude and phase offsets in the Stokes 'Q' and 'U' visibilities of 
3C286 and 3C273, the like of which I had not seen before.  The offsets 
were especially large -- up to 30 degrees in phase -- for antennas 14 
and 15, but it was clear that virtually all antennas (both VLA and EVLA) 
were affected to some degree at some times. 
    A review of my polarization data of the same sources, taken in 
October in 50MHz continuum, showed no such effect. 

    Last weekend, I observed these same two sources, plus the totally 
unpolarized source OQ208 (thanks to George M.!), in both 50 MHz 
continuum, and in the same 12.5 MHz polarization spectral line mode used 
last month.  The observations were all at X-band, with 3.3 seconds 
integration. 

    Careful calibration and imaging showed the same peculiar offsets, 
but in both line and continuum, although now the offsets were largest in 
15, and nearly absent in antenna 14.   What could be happening? 

    Yesterday, George and Steve noted an inconsistency between data 
filled into CASA and AIPS in the nominal sensitivities between the RL 
and LR correlations.  Eric quickly found that for some baselines, the 
'R' and 'L' system temperature data were reversed by AIPS upon filling.  
I thus re-filled and re-calibrated the weekend tests, using the 
corrected software -- and -- voila! -- all phase and amplitude offsets 
in the Q and U data are gone, in both line and continuum files. 

    We are still trying to understand just how these reversed weights 
can have so dramatic an effect on the linear polarization data (it's not 
elementary, as the polarization programs are involved in the process).  
It is clear that data filled as correlation coefficients, and which are 
not calibrated via the system temperature data, are not affected by this 
problem.  All data taken since the Modcomps were turned off, which were 
filled as visibilities, or as correlation coefficients which 
subsequently had the system temperature data applied, are affected.  My 
October data were filled as correlation coefficients, and hence were not 
affected by this problem, which explains the absence of the effect in 
those observations. 
The parallel-hand observations are not affected, as the reversal is 
calibrated out. 

    I made images of the sources, with outstanding results:

For OQ208, which has effectively no polarized flux density at all ( less 
than 0.15%), the naturally weighted Stokes 'Q' image had an rms noise of 
70 microJy.  With 530,000 visibilities, each of 6.2 MHz BW and 3.3 
seconds integration, the theoretical expectation is 60 microJy.  For the 
EVLA-only image, we got an rms 140 microJy, for the VLA-only, 190 
microJy.  The Stokes Q and U images show barely detectable point-sources 
of emission, with nearly no nearly 'disturbances' in the map which would 
be the sign of varying cross-polarization

For 3C286, the uniform-weighted images are almost at the theoretical 
noise limit in all four (I,Q,U,V) Stokes images.  The V image shows no 
emission at all (as expected for self-calibrated data -- in truth, there 
is a small V component, but this is not known to the calibration 
process, and cannot be detected without another source being utilized 
for calibration).

3C273 provided a 100,000:1 dynamic range image, without my using any 
'heroic' measures for editing.  The noise limits in Q and U are 
significantly (factor of 2 or so) above the expected noise -- this is 
likely an effect of ignoring 2nd order leakage from the Stokes I image.  
For this object, the I flux is 32000 mJy, while the noise in the Q and U 
images are less than 0.5 mJy.