[evlatests] 3-bit PDif Compression -- a clue?

[Rick Perley]

    At the ECSV meeting yesterday, Barry opined that the cause of the 
'PDif' compression (switched power value decreases when we observe a 
source strong enough to raise the system temperature) may be related to 
saturation of the 3-bit samplers -- if a significant fraction of the 
voltages lie outside the 8-level ladder, (caused, for example, by a 
significant offset in the voltage distribution) some non-linear effects 
can be expected.

    We have some data in hand which can help test this idea.  For the 
Friday night test of 3-bit stability, we observed 3C84, and adjacent 
cold sky, at Ku-band, starting at a very low elevation.  Due to ground 
pickup and atmospheric emission, the system temperatures varied a lot 
during this time.  As the attenuator/requantizer levels were set on cold 
sky (near the vertical), any marginal voltage distributions would be 
exacerbated when we move the antennas to the horizon.  In fact, the 
system temperatures over the four-hour run dropped from ~54K to 33K -- a 
65% decrease.  So .... what happened to the PDif values during this 
time.  Given that the temperatures barely changed at all, we have an 
expectation that the amplifier gains should not change significantly 
(due to temperature) during this time. 

    The answer is that the PDif values changed -- a lot!  The PDif on 
cold sky *rose* by 40%  while the system temperatures were declining by 
65%.  So good an (inverse) power meter are the PDifs that they easily 
saw (in reverse) the passage of a cloud which increased the system 
temperature by about 3K. 

    The PDif change is not linear with the power, however.  While the 
PDifs were increasing by ~ 40%, it appears that the *change* in PDif due 
to going on and off 3C84 'only' rose by 30%. 

    I checked the PDifs for the 8-bit path during this same 
experiment.    There is also a rise in PDif during the period, but it is 
my much less -- a 'mere' 6% -- we might possibly be able to ascribe this 
a temperature change?  Or perhaps this is an elevation dependence of 
system gain? 

    I suggest our next experiment be done by deliberately reducing the 
power level to the samplers by a 2 dB or so -- probably also change the 
requantizer levels? -- to ensure that we don't get even close to 
overpopulating the end ladder levels of the sampler or requantizer...