[evlatests] Something Odd in applying RQ and T304 gain corrections

[Rick Perley]

     I reported last week on the effects of 'PDif' issues on gain 
corrections.  Within that report was a note on some odd problems with 
some antennas.  This note expands on this.

     The observations spanned about 16 hours, last October.  All bands 
(from P through Q) were observed.  There were typically 200 observations 
at each of the nine bands -- that's a lot of receiver and correlator 
reconfigurations!  Troubles were expected -- and did indeed occur.

     The 8-bit samplers were employed for all nine frequency bands. 
Unique to this observation was resetting the requantizers for every 
observation at P, L, and S bands.  The modifications to the script were 
done either by Ken or Vivek.    Note that we normally employ the RQ 
resetting only with the 3-bit samplers -- and in the subsequent 
analysis, *only* the RQ changes are applied to the data. (This because 
we have yet to face up to the issues of understanding, and correcting 
for, the dependence of the digital PDif values as a function of power).

     In calibrating these 8-bit/RQ adjusted data, a peculiar effect was 
noted.  For all three bands, large gain changes occurred for a few 
antennas.  The characteristics are:

     1) Both polarization affected equally.
     2) All spectral windows affected equally.
     3) The same antennas affected in the same way for all three bands.  
Worst was ea11, followed by ea04.  Most antennas were not affected.
     4) The gain change was always the same way -- the visibilities 
dropped in amplitude.
     5) The magnitude of the drop was typically 10 -- 50%.
     6) The change of gain lasts only for

     More careful review showed that, in many, but not all, cases, the 
change of gain was accompanied by a small change in delay (< 1 ns).

     So, on the face of this, it looks like a clear case of the T304 
attenuators changing (when they shouldn't, since 'set-and-remember' was 
invoked).  If this is what is happening, the result of this should be 
clearly evident in the switched power data.

     There are *no* changes in PDif accompanying these observed large 
changes in fringe power.    The large drops in visibility flux are 
(almost) exactly the same whether the RQ-only changes, or the PDif 
changes are applied.

     We're missing something here.  Our understanding is that all the 
needed information for gain correction is in the PDif values.  That is 
-- whether the RQ was changed or not, the PDif alone is sufficient for 
complete gain correction.

     Is this correct?