[evlatests] More mysterious results from stepped power 3-bit tests

[Ken Sowinski]

Rick's message was mangled by my mail reader.  In case others
have the same problem, here is a reformatted version which,
with any luck, has not been re-mangled.

     Late yesterday afternoon, Ken repeated the 'stepped power' test with 
the 3-bit samplers.  The idea is to step the power levels to the sampler 
in 3-db steps, for each of which he observed the calibrator 3C286, and 
an adjacent noise field.

     The power levels chosen were -38 dBm through -26 dBm -- the output 
analog power meter in the T304 modules was used to measure the power, 
from which the appropriate output attenuator level was chosen.  The 
input attenuator in the T304 was kept fixed.  (This is my understanding 
of the experiment -- Ken should correct me if I am wrong).   All 
antenna/IFs were treated identically -- we should then expect (or at 
leat hope) that 'downstream' measure of the power should be quite similar.

     I give below some curious results from this test.

     A) Power increments.

     The output attenuators were adjusted in 3 dB steps.  If all is 
linear, I should see 3 dB steps in the output cross-correlation power, 
and in the the 'autocorrelation' spectra.  But this doesn't happen -- 
it's not even close.  I give below a table of the power increments, in 
dB, as judged from the CALIB solutions (essentially, the step in the 
fringe power observed, factored out by antenna/IF).  There are five 
power levels, so we have four steps.

Power levels        12R     12L    15R   15L    22R   22L    28R   28L
-------------------------------------------------------------------
-38 -> -35 dBm      1.7 dB  2.7    3.2   4.4    2.4   3.4    4.0   2.8
-35 -> -32          2.7     2.5    2.4   2.2    2.6   2.6    3.1   2.8
-32 -> -29          2.4     1.8    1.8   1.9    2.3   1.7    1.7   2.0
-29 -> -26          0.5     0.9    1.3   1.6    1.3   1.0    1.9   1.3
----------------------------------------------------------------------

     It's clear that as the power rises by steps of 3dB (as judged by the 
analog power meter), the increment in fringe power becomes less, and in 
some cases much less, than the expected step.  All antenna/IFs share this.
     Ken explains that this is likely due to the 4-bit requantizer gain 
following the station board -- there is no attempt to adjust its levels 
for the change in input power level.  He also notes that at some point 
the output attenuator reaches 0 dB, and can go no lower, so we will 
(eventually) hit a ceiling.  I personally think the first explanation 
better explains what we see in the table above.
     I checked the power increments as judged by the autocorrelation 
spectra.  They agree with the general trends noted above -- power 
increments as judged by the correlator roll off as the input power rises.

     B) Power Levels.

     If the T304 power meters are good and reflect real power, then we 
should expect to see about the same spectral power levels as judged by 
the autocorrelation spectra, and see the same CALIB gains needed to 
change the cross-power amplitudes to units of real Jy.  (This argument 
assumes that the spectral window we are actually using reflects the 
overall spectrum -- i.e. there are no major slopes in the 2 GHz-wide 
spectral power, and (for the CALIB gains) that all antennas have about 
the same basic sensitivity).

     The range in spectral power seen in the autocorrelation spectra 
varies over quite a wide range:

             12A   12C    15A   15C    22A   22C    28A   28C
Spec. Pwr  0.75   1.8    1.8   1.4    1.5   2.6    1.3   2.0

     The maximum range is for 22C/12A -- about 5.4 dB.   This is 
conceivably a spectral slope issue -- I'll check this later.

     But more confusion comes from analysis of the AIPS gains needed to 
convert the cross-power to actual Jy.  If all antennas have about the 
same SNR, then the higher power antennas listed above should have the 
smallest corrections needed to return the cross-power to real 
amplitudes.  But this is certainly not the case ...
              12A   12C    15A   15C    22A   22C    28A   28C
Correction   6.2   1.7    2.3   2.6   10.8   2.5    7.0   3.5

     There should be an inverse relation between the correction values 
listed above and the power levels listed higher up.  I see no relation 
at all!  The weakest antennas as judged by corrrelation power (22A and 
28A) have average spectral autocorrelation powers.  The strongest 
antenna as judged by correlation power (12C) also has a normal 
autocorrelation power...

     C) Noise changes as a function of power input.

     I noted yesterday a small improvement in rms noise as the input 
power rose.   I give below the values for all 12 correlators for three 
of the 5 power levels -- the lowest, the middle, and the highest.

Baseline/Pol       -38 dBm         -32 dBm         -26 dBm
----------------------------------------------------------
12 x 15 RCP        .28  Jy          .23             .22
12 x 22            .54              .54             .44
12 x 28            .40              .34             .30
15 x 22            .54              .42             .36
15 x 28            .37              .28             .25
22 x 28            .77              .62             .48
12 x 15 LCP        .23              .17             .17
12 x 22            .27              .23             .23
12 x 28            .31              .26             .25
15 x 22            .32              .24             .23
15 x 28            .33              .26             .24
22 x 28            .43              .35             .30
----------------------------------------------------------

Despite the evident saturation of the requantizer, there is notable 
improvement in nearly all noise levels.  Only one correlator -- 12 x 15 
in LCP, is even close to the expected level of 0.15 Jy.

D)  Autocorrelation Curiosities

     The 'parallel' autocorrelations (RR and LL) in general look 
reasonable, and change as expected as power levels change (albeit not 
with the 3 dB step we expected ...).  But there is one notable effect 
seen in early all of the spectra -- when the input power reached a 
certain level, all spectra became negative!  The threshold is the same 
for all antenna/IFs.  The shape of the spectrum remains more or less 
correct as this threshold is passed.  Evidently an overflow problem.
     The CW 'tones' are visible in two of these 12 antenna-IF spectra: 
15R and 28L.  (This doesn't mean these are absent in the other 
antenna-IFs -- we have only one spectral window in these tests).

     One other item which is less hard for me to understand is the 
'cross-power' autocorrelations (RL and LR).  For three of the four 
antennas, the 'cross auto' spectra are pure sinusoids of very high 
amplitude and zero mean.  Clearly a single lag has a very high 
amplitude.  The spectral frequency of the sinusoid is the same for RL 
and LR, and different for the three antennas which show this.  The 
antenna which does *not* show this is 15.  The 'cross-power' spectram 
displayed is enormously negative and shows no imprint of the bandpass 
(i.e. power goes up at the edges, rather than down ...).  Barry, Ken, 
and Michael all claim to sort-of-understand all this.  I'm not too 
embarrassed to admit that I do not ...

     This is about all I can squeeze out of this ...