[evlatests] PDif, PSum and Tsys for 3 and 8 bit paths

[Rick Perley]

    The test from yesterday allows some trolling through the switched 
power monitor data, looking for clues ...  There are lots of clues, but 
they don't lead me in any useful direction.  Perhaps one of you can see 
something in this.

    As noted earlier, the test was taken with four different 3-bit 
setups, in parallel with an 8-bit path, all at the same frequency.  For 
each setup, the system first adjusts the T304 attenuators (and, for the 
3-bit path, selects the appropriate slope filter), after which it 
adjusts the requantizers to give the desired state count distribution. 

    I looked at the reported PSum values to see how repeatable this 
power setting procedure is.  For the 8-bit path, it is very repeatable, 
with power levels the same for each of the four setups the same within a 
few percent.  The measured power level, as measured in 'counts' is 
between 14.0 and 14.6.  Only rarely is it found outside this range.  The 
maximum deviation seen for all antennas/setups was only 7%.  For all 
setups, the reported system temperature did not change for any 
antenna-IF by more than a couple tenths of a Kelvin -- indicating that 
when a different power level was set, the PDif values accurately 
recorded this different gain, so that an expected system constant (the 
system temperature) was indeed constant. 

    None of the above is true in the 3-bit paths.  Every antenna-IF 
showed quite different PSum values for the four individual setups.  The 
maximum range seen was 10.8 to 15.1 -- the distribution of powers makes 
it very clear that nearly all deviations are 'down' -- too little 
power.  And they are common -- the median power level is around 13.0. 
    This may not be a worry if the PDif values tracked these power 
changes -- but they do not do this accurately.  The error is most easily 
seen in the Tsys values (=Tcal*PSum/PDif).  We know that this value 
should not change over the short range of time and elevation that the 
test occupied -- and this is seen on the 8-bit side.  But on the 3-bit 
sides, very large changes in Tsys are reported, with some antenna-IFs 
claiming to have more than double the Tsys on some bit setups than on 
others.  All the largest erroneous Tsys values are seen with the 234 and 
345 bit levels.  But the 012 and 123 are hardly immune from this 
problem.  It's just 'not as bad'.  Of the four 3-bit level setups, the 
012 setup gave the best match to the 8-bit values.  Unsurprisingly, the 
calibrator 'antenna temperature' (change in system temperature between 
cold sky and on-calibrator) is quite variable for the 3-bit paths, 
depending on setup and antenna-IF.  All 3-bit values for this are too 
large, usual about double what they should be.  The 8-bit values are all 
the same (between antennas and IFs) at about 0.6 K (which is about the 
right value, given the source strength and antenna efficiency).  The 
3-bit values range all the way up to 2K (which would lead to an antenna 
efficiency of over 100%!).  There is no correspondence between 
deviations in the A1C1 path with those in the A2C2, other than 
statistically -- both are worst at the 345 setup, and best on the lower 
two setups. 

    There is good evidence that the 3-bit PSum values are accurately 
responding to short-term variations.  ea07R conveniently shows rather 
fast power variations on the 8-bit side -- and these are seen 
identically in all 3-bit paths.  Furthermore, the change in PSum when we 
go from cold sky to 3C345 appears to be identical (to the rather rough 
accuracy that these plots show) between the 8-bit and 3-bit paths. 

    All 3-bit path combinations show 'PDif compression' when we go from 
cold sky to calibrator -- this is the reason we get ridiculous antenna 
efficiencies.  This decrement is not seen at all on the 8-bit path.  Not 
all 3-bit combinations lead to the same compression values (which is why 
the claimed Tsys varies from setup to setup).   The compression appears 
least for the 012 combination -- probably explaining why the Tsys values 
for this setup are the closest to those given by the 8-bit path. 

    All the above comes from looking at a single subband (#9) in the 
3-bit sides, and comparing it to the coresponding subband (#5) for the 
8-bit path.  I'll now review how these levels vary within subbands ...