[evlatests] L-Band Polarization Stability

[Rick Perley]

    I ran a 5-hour observation of the strong and unpolarized calibrator 
3C147 at L-band last evening, with the goal of determining any 
variability in antenna polarization. 

    Short Summary:  The polarizations are very constant, and excellent 
images were made in all Stokes parameters.  However -- the noise levels 
are twice as high in Q and U than in I and V -- a result of variable 
cross-polarization, of mis-calibration of the X-hand data.  The results 
are very good -- but are they good enough? 

    Details (for those who want to know them). 

    Observations were taken in continuum, at four different frequency 
pairs, ranging from 1193 through 1985 MHz.  All BW were equal to 50 MHz, 
except for 1193 MHz, which was observed at 12.5 MHz. 
    I took no particular care on the assignment of frequencies, with the 
result that four of the eight frequencies were affected by varying 
amounts of RFI.   Rather than struggle with editing these, I 
concentrated on the RFI-frequencies.  (The RFI was in some cases very 
unusual -- I'll report on these issues later). 

    The data were edited and calibrated in the usual way.  Closure 
corrections were determined for the parallel-hand correlators, and I got 
a 100,000:1 dynamic range image in Stokes 'I' at the RFI-clean 
frequencies.  Typical peak was 21,000 mJy, typical rms noise was 0.21 
mJy.  

    The RL and LR visibilities -- corrected for antenna gain, but not 
polarization -- are constant in both amplitude and phase.  For an 
unpolarized source, this is what we hope to see, and is interpreted as 
the unchanging antenna cross-polarization only.   As has been often 
reported, the typical cross-polarizations (best seen in VPLOT by taking 
the ratio of RL/RR, for example) are higher on the EVLA antennas.   

    For VLA-VLA baselines, typical X-polarizations are 2 to 4% (there is 
very little spread in the values amongst all the baselines), except for 
antenna 28 (which has a quadrature hybrid polarizer, rather than the VLA 
phase shifter), which is typically 9% against the other VLA antennas. 

    For EVLA-EVLA baselines, the X-polarizations are 1 to 10%, with some 
baselines as high as 15%.  OTOH, some baselines are very low -- less 
than 5% is not unusual.  What is most striking is the wide range in 
X-polarization. 

    PCAL gives good solutions, and the imaging results are very good.  
The Stokes Q and U maps have an rms noise about twice the level in I and 
V.  There is no identified source of emission (easy interpretation is 
that the source is completely unpolarized -- but read on for an 
alternate view!) at the phase center (but one of the background sources 
is clearly polarized!). 
    The higher rms in Q and U is clearly due to 'scattered' coherent 
power, as there is non-noiselike structure in the images, concentrated 
around the phase tracking center.  This is indicative of a gradually 
varying 'leakage' signal which is incoherent amongst the various 
baselines.    The level is very low (the rms in the image is only 0.005% 
of the peak in I), but still worrisome. 

    I investigated further by making images with VLA-only and EVLA-only 
subarrays.  In Stokes I, the results are similar for both subarrays -- 
the EVLA has a slightly higher noise level, probably due to the lower 
efficiency of these feeds (and perhaps due to some of the EVLA antennas 
having apparently high system temperature).  However, in Stokes Q and U 
images, there is a distinct difference -- the rms in the EVLA-only 
images is twice that of the VLA (and for both arrays, the noise in Q and 
U is twice that in I or V).   This result indicates that the variation 
in the cross-polarization is larger with the EVLA than the VLA antennas, 
approximately in the ratio of the magnitude of the cross-polarizations. 

    Plotting the fully calibrated RL and LR visibilities (i.e., after 
both gain and polarization calibration) shows the level of variation in 
the cross-polarization terms, (presuming that PCAL has done a good job 
in separating source from antenna polarization), and search for coherent 
behavior.  The typical variation in the ratio of RL/RR is about 0.2% for 
VLA-VLA baselines, with a maximum of about 0.5% (except for antenna 28 
again, which varies by up to 1%).  The variation in phase is most 
striking, with a steady rotation of about one full turn over the 
observing interval.  The phase of RL is opposite that for LR -- could 
the source be polarized, and rotated by ionospheric Faraday rotation? 

    If this were the case, one would expect a similar effect in the 
EVLA-EVLA baselines -- which we see, except that the magnitude is at 
least twice as high, with residuals around 0.5%, rising to 1% for some 
baselines.   This scaling argues for an antenna-based, rather than 
source-based, origin. 

    For a properly calibrated array, observing an unresolved polarized 
source at the phase center whose plane of polarization is being rotated, 
the net Q and U vectors will show an oscillatory amplitude, and a 
square-wave phase, alternating back and forth between 0 and 180 
degrees.  This is not what is seen in these new data -- both Q and U are 
complex, and variable at the levels indicated above.    The variations 
are clearly not random, and there is much commonality between antenna 
pairs.  The EVLA looks different from the VLA in all this only because 
the magnitudes are higher. 

Although it's not evident that a weakly polarized 3C147 combined with 
ionospheric faraday rotation is the answer, perhaps PCAL, confused by a 
rotating polarized vector, is somehow responsible.  The CDDIS 
ionospheric data should be online in a day or so, so when these are 
available, I'll attempt to account for ionospheric rotation. 

In summary -- polarization stability is very good, but not (in my view) 
good enough.  We're doing something wrong, but I've yet to understand 
what it is.