[evlatests] More on compression and switched power

[Rick Perley]

    Another observation of Cyg A and two local calibrators at L, S, C, 
and X bands was made yesterday to improve our knowledge of what works, 
and what doesn't work... 

    I've reduced only the L and C band data so far.  But these are 
enough to make some things a little clearer.  Unfortunately, the 
situation is complicated, and there's more than one problem ...

    1) At both L and C bands, observing Cyg A causes the PDif values to 
decline *in general* compared to the local calibrators -- by a 
significant fraction at L-band (generally 10 to 30 %), and much less at 
C-band.  Application of these PDif values to the data does partially 
restore the flux density of Cygnus A towards the correct values -- but 
not completely.   The best example of an incorrect correction comes from 
Cband, where (as has long been reported), ea24, in IF 'A' has a PDif 
compression of about 40% when on Cyg A.  Application of this value to 
the visibilities hugely overcorrects the visibilities -- and 
self-calibration promptly removes the entire correction! 
   
    So it would appear that the statement made by me a few months ago 
that PDif compression does not affect the visibilities, and the 
statement made by me yesteday that the PDif compression does affect the 
visibilities are both correct! 

    2) Perhaps more serious, and possibly unrelated, is the non-closing 
nature of the Cygnus A data at L-band.   Calibration of the Cyg A L-band 
data (either with, or without the PDif applied) gives a result where it 
is clear there are large (tens of percent, and sometimes larger) closing 
errors.  This is most easily seen by plotting the Stokes 'V' 
visibilities after calibration -- these have amplitudes of tens to 
hundreds of Janskys!!!  This cannot possibly be due to the beam squint 
(Cyg A is only 1/15 of a L-band beam wide). 
       On the other hand, the C-band Cyg A data self-calibrate very 
beautifully -- the 'V' stokes visibilities are now at the level of 1% or 
less ... quite explainable by the squint. 

    So what can cause the closure failure?  I note that Cyg A quadruples 
the system temperature -- and that the T304s are running 'hot' -- the 
output attenuators are typically at 25 to 31 dB.   Are we overflowing 
something in the digital system?  Overflowing the requantizer seems 
unlikely to cause a closure problem, but what if we are overflowing the 
cross-multiplier accumulators.  Note that Cyg A has a correlation 
coefficient of over 75% at L-band ... 

    We plan some test tomorrow to see if reducing the input power can 
restore the closure relations...