[evlatests] EVLA Polarization stability

[Rick Perley]

    Seven hours of excellent data were obtained last Friday at L, C, X, 
and K bands, in a test to try better understand the apparent change in 
antenna cross-polarization.  '

    The data were taken in continuum, with BW = 50 MHz, at standard 
frequencies, with 0.417 sec integration.  Two sources were observed, 
3C286 (strongly polarized, at about 10%), and 3C287 (weakly polarized at 
all bands, and almost completely unpolarized at 20cm).  About 90 seconds 
of dwell on each source, at each frequency, were obtained.    The two 
sources transited about 2.5 hours into the run, giving a very large 
range in both parallactic angle and elevation. 

    Variation in antenna polarization over time is seen at all bands, 
but is very much greater at lower frequencies.  Indeed, at K-band, it 
seems there is essentially no variation at all, while at L-band, the 
variation is quite stunning. 

    Concentrating on L-band, I see the following:

    a) Both VLA and EVLA antennas show variable cross-polarization over 
the time.  However, the variation, and the amplitude of the 
cross-polarization, is much larger on EVLA antennas. 

    b) Referenced to a good VLA antenna (28), VLA X-polarization is 
(according to PCAL) typically 2 to 5%, while EVLA X-polarizations are 10 
to 15% (notable exception is antenna 14, at about 4%).  Referenced to a 
good EVLA antenna (17), the X-polarizations are typically 8 to 10% for 
VLA antennas, and quite variable (2 to 12%) for EVLA antennas. 

    c) It makes no essential difference to the solutions if I use 3C286, 
3C287, or both, as the calibrator sources. 

    d) For the EVLA antennas, there is some evidence for an elevation 
(or parallactic angle -- can't tell the difference here) dependence for 
the variable cross-polarization.  This is best seen by looking at the 
amplitude of the RL correlator for 3C287, which is completely 
unpolarized.  However, the dependency is not so obvious in phase.  For 
all baselines,  the variations are 'slow and deliberate' -- they are not 
random, and vary on timescale of hours, rather than minutes or less. 

    e)  The variations repeat from day to day!  In comparing the 
cross-polarized responses to data taken 3 days earlier, the amplitudes, 
phases, and changes in both all repeat (nearly exactly!) for most 
baselines. 

    A note on calibration:  It is important to remove all effects of 
parallel-hand variations before looking at X-polarization.  3C287 is 
completely unresolved at L-band, while 3C286 is very slightly resolved, 
and has an excellent model available.  Great care has been taken in the 
editing and calibration to ensure none of the variations seen in X-pol 
have an origin in the parallel-hand gains. 

    A note on PCAL:  This program is a black box to me.  I use it in the 
standard way, and there are likely approximations contained within that 
may modify some of the details seen.  On the other hand, perusal of the 
3C287 data (with no polarization corrections) is sufficient to give the 
conclusions noted above.  PCAL is certainly working as expected, in 
using the parallactic rotation of the antenna to separate source from 
antenna polarization.   The values provided (and the residuals seen 
after the antenna polarizations are applied to the data) are 
reasonable.  However, what is *not* reasonable are the program's 
estimates of the errors in the polarization parameters.  Despite the 
manifestly poor fits, the program informs me that all D-term amplitudes 
are accurate to typically 0.05%!  (This is about one part in 100!), and 
the phases accurate to 0.4 degrees).