[evlatests] PDif Compression -- where it doesn't come from ...

[Rick Perley]

     We ran two test scripts last week, observing Cygnus A and some 
nearby standard calibrators at P and L bands.  The primary purpose was 
to check on system linearity, in preparation for upcoming 'flux density' 
runs.    The regimen utilized was to set the power levels (in the T304) 
on  Cygnus A, and also to set the requantizer power levels for each 
subband -- something not yet done for regular observing at these 
bands.    These levels were then 'remembered' for the rest of the 
observing run.

     A benefit of this test was a check on the long-standing 'PDif 
Compression' problem.  To remind readers, this is the observation that 
the synchronous power decreases, for many antennas and IFs, when the 
antennas observe a strong source such as Cygnus A, or the moon.  There 
is much evidence that this decrease in switched power is *not* due to an 
actual reduction in gain (so application of the switched power 
calibration regimen would be incorrect) -- but we have not had to date a 
clear measure of whether the visibilities are affected.   We have now a 
clear answer.  (Read on ...)

     The essential results are:

     1) There is almost no PDif Compression at P-band.  Cygnus A 
multiplies the system power by a factor of about four.

     2) There is significant PDif compression at L-band, where Cygnus A 
multiplies the system power by a factor of about 5.  Significant 
compression (more than 10%) is seen in only a few antennas/IFs -- ten 
out of 52 signal paths.

     3) By utilizing self-calibration, I can compare the required gain 
corrections for the visibility data against the PDif compression values 
to judge whether the presence of PDif compression causes visibility gain 
changes.  The answer is *yes*, but the effect is not large.  For those 
antenna-IFs which 'compress' more than 5%, there is a loss of fringe 
power (not amplitude) by 2 to 10%.  For a 15% PDif compression, the 
visibility power drop is about 5%.  All antenna-IFs with significant 
compression show visibility decline. This suggests the presence of a 
small non-linear response in the analog system.  PDif measures the slope 
of the gain curve -- for a non-linear response, this not the same as the 
loss of overall gain. For a system going into compression, the slope 
will decrease as the power increases.  This qualitatively matches the 
observations.

     4)  This measurement encouraged us to try a simple test yesterday, 
with the goal of isolating where in the system the compression might be 
occurring.  For two antennas with the largest compression in both 
polarizations (ea01 and ea04 -- all four paths were compressed by 10 to 
15%), we had 6dB attenuators inserted into the RF path, immediately 
after the receivers.  This lowers the power to the T302 (LSC Converter), 
and to the input amplifiers of the T304.    Note that the input 
attenuators of the T304 will put the power levels to all downstream 
components to the same level as in the first experiment.  Hence, the 
only portion of the signal change for which the power levels changed 
should be the T302 and the input amplifier of the T304.
     Observations of Cygnus A and 3C380 were then taken, with exactly 
the same setup procedures as described above.

     *** The result of this was that there was no significant difference 
in the PDif compression in these two antennas!! ***  (Nor was there any 
significant difference in any other antenna).



     5) This null result strongly suggests the problem must lie in one 
of two places:

     a) In the front end.
     b) In the amplifiers in the D30x sampler modules, or the samplers 
themselves.

     A strong argument against (b) is this:  If the D30x amplifiers were 
at fault, we should see the same compression in all bands for the same 
IFs, since both the power and the spectral power presented to the 
sampler is the same at each band.   But we don't.  Each band is 
different.   Furthermore, many tests done a couple years ago where we 
stepped the T304 attenuators over wide ranges exonerated the T304s and 
the subsequent digital path.  (Or, to be honest, we think we exonerated 
these ...)

     A strong argument against (a) is this:  The FE amplifiers are 
designed with very high headroom, since we know that incidental RFI 
could add more than 20 dB of power.  Furthermore, the Y-factor (hot/cold 
load) tests done by Bob Hayward and I, at all bands, regularly added up 
to 10 dB of wide-band power, with no sign of any non-linearity.

     So we (or at least I) are again at a loss to explain this 
well-observed problem.  We could separate these two proposed origins by 
swapping either a D30x sampler module, or a FE, between two antennas 
(one with clear compression with one without).  I suppose the first is 
much easier ...

     Comments, suggestions (or, better, solutions!) are welcome.