[evlatests] PDif Correction Failures

[Rick Perley]

    The previous note gave the fraction of antennas which are badly 
behaved in their PDif values.  (Note:  Be aware of my definition of 
failure in that previous note -- the 'failure fraction' is not the 
fraction of attempts which gave a different or variable value, but 
rather is the fraction of antennas which -- over time -- give unsteady 
values of PDif). 

    Here I tally up the failures for PDif values to correct the 
visibilities.  That is -- following application of PDif to the 
visibility data, which antennas were *not* adequately corrected for the 
changes in visibility amplitude caused by poor switches? 

     The list is quite short -- meaning that application of PDif nearly 
always corrects the gain variability associated with the bad switches. 

    Band      Notes
--------------------------------
L             ea28 in LCP.  PDif here is wildly variable most of the 
time.  There are periods of good behavior.
S             ea6C -- ~7% changes in PDif which are *not* seen in the 
visibilities.
C             ea25 in A and C:  Five of the 37 scans showed a 2% change 
in PDif not seen in visibilities. 
X             ea12 in IF 'D' -- PDif is simply large values of random 
noise. 
Ku           ea18 in all IFs:  ~2% variations in PDif not seen in 
visibilities
               ea25 in all IFs:  ~10% variations in PDif not seen in 
visibilities
K             All short-scale changes in visibilities are corrected by 
PDif, to ~2% accuracy. 
Ka           ea01, in RCP:  PDif provides large random values most of 
the time (this is like ea28 at L-band).
               Failures at the <5% level cannot be easily distinguished 
from pointing errors. 
Q            Same as K-band, but at the 10% level, set by the pointing 
repeatability. 
--------------------------------------

Bottom Line:  Except for the few failures in PDif noted above, 
application of PDif does an excellent job in correcting for the 
switch-induced gain losses. 

Unfortunately, we cannot yet recommend automatic application of PDif to 
our visibility data because of the clear evidence that PDif is reacting 
to changes other than in the gain.