[evlatests] PDif compression and expansion -- a few 'final' notes

[Rick Perley]

    I spent the day trolling through the 3-bit test data, looking for 
more clues to the mysterious apparent (and false) gain changes implied 
by the synchronous calibration system. 
    We have two good consecutive days of data, taken 16 and 17 Nov., of 
the same sources, using the same script, at the same LST, in nearly 
identical weather.  Comparison of these lends a few clues to the 
stability of the phenomenon.  I also smoothed the switched power data to 
20 seconds, to help increase sensitivity. 
    A few conclusions emerge:

    1) The 'PDif' compression problem is not restricted to the 3-bit 
data.  It is also in the 8-bit data, but at a much lower level.  The 
antennas on which Pdif compression on the 8-bit path are seen are:  
ea07L-- both days at the 1% level, and ea23R and L, at the 0.5% level, 
but on the first day only for RCP, and the second day for LCP.  Other 
antennas *may* show the effect at levels of ~0.25% or less. 

    2) The 3-bit PDif compression is time variable!  I've already noted 
that ea06R had compression on the first day, but expansion on the second 
day.  (This alone proves that the problem is not due to power alone -- 
the power levels, and the variation of these levels in time were 
identical on both days).  Even more notable is ea14, where the PDif went 
from compression to expansion within a 3 hour period.  It did this on 
both polarizations on both days -- but not in identical ways. 

    3) We can be fairly sure, on physical grounds, that the Tsys values 
should be the same on the two days for any one antenna-IF.  This is 
certainly the case for the 8-bit side, where the agreement is always 
within 0.5 K.  But (unsurprisingly), this is not the case on the 3-bit 
side, where some antennas show up to a 10K apparent (and clearly false) 
change in Tsys between the two observing runs. 

    I went to some trouble to confirm that AIPS is correctly utilizing 
the Pon and Poff data, and that the CD (cal device) table, which 
contains the Tcal values, are correct.  Everything checks out. 
   
    I've noted before that the visibilities appear to be unaffected by 
this 'compression' problem.  The degree to which they are unaffected is 
still not known with high precision.  I'll endeavor to try put tighter 
limits on this over the next few days.