[evlatests] Major PDif problem on 3-bit paths

[Rick Perley]

    The test run Friday night with the 3-bit system has revealed a 
serious issue with the switched power system.  The problem is the same 
as we've seen with 8-bit, but much, much more severe. 

    In reviewing the data, I looked at the values of PDif, PSum, and the 
Tsys values generated by these first two.  Right away a major issue was 
seen:  For all of the 3-bit equipped antennas, the PDif values are 
strongly modified by the total power.  The effect is not small.  3C84 
adds about 3K to the system temperature (10% of cold sky).  When we move 
the antennas from 3C84 to cold sky (a 10% drop in total power) , the 
values of PDif usually *increase* -- by up to 5%.  This is huge, and we 
know it's wrong, since the derived value of Tant from observations of 
3C84 lead to an efficiency exceeding 75%.  An impossibility. 
   
    Each antenna and polarization behaves differently.  (I haven't yet 
checked if the subbands behave differently too).  And -- significantly 
-- the behavior is quite different between A1 and A2, and between C1 and 
C2.  Since these pairs come from the same receivers, and go through the 
same IF electronics *until the T304* -- the differences noted above must 
be generated either within the T304 or the digital system which follows. 
    Not all antennas show PDif compression when going onto 3C84.  ea14 
shows expansion -- the PDif values get larger when on the strong source. 
    On the 8-bit side, all is good, except for ea10 on LCP, where a 
faint PDif expansion (by 0.5% only) is visible. 

    The data taken will not allow us to determine whether the 
visibilities are affected by the problem -- i.e., is the problem one 
which affects the actual data, or is it solely in the monitor data.  
Experience with this problem on the 8-bit side (as shown at C-band) 
leads me to predict that we will not see an issue in the visibility 
data.  But a test is needed to show if this prediction holds. 

    Since we were not successful in the Friday evening test to show any 
temperature-dependence effect for polarization (because the night was 
cloudy and the temperatures hardly changed), I propose repeating the 
Friday test, and adding in a third source, with flux density ~ 1% of 
3C84 (easy to do).  We'll pick a good-weather night, with maximum 
temperature swing. 

    Never a dull moment here ... :)