[evlatests] 3bit vs 8bit flux transfer

[rperley]

A short test run was take a few days ago, which compared flux transfer 
accuracy between 3 and 8 bit quantizers, and which assessed the accuracy 
of 'absolute' flux density calibration with the 8-bit samplers.

Observations were made of 3C286 and J2007+4029 (the calibrator source 
for Cygnus A) when the two sources were at the same elevation.  Bands 
utilized were X, Ku, K, Ka, and Q.  For each band, observations were 
made in both wideband (3-bit) and narrowband (8-bit) modes.  For the 
latter, the two IFs were chosen to lie near the opposite ends of the 
bands.

Test1:  Compare flux transfer between the sources.

    For the 8-bit mode the full switched power values were applied to the 
data.
    For the 3-bit mode, only the requantizer (RQ) gains were applied, as 
it is known that the 3-bit digital switched power ('PDif') is not 
linearly related to the analog power.

   Result:  The determined flux density for J2007, based on 3C286, were 
the same, within 1%, for both sampler modes.
    This surprised me a bit -- I had expected a few percent differences.  
Evidently, the actual gain variations of the amplifiers/IFs between the 
two sources (located far apart on the sky) is very small.  So small that 
in fact, no gain correction at all is required!  (At least, not for this 
day).

Test 2:  How good is the gain calibration when based on knowledge of the 
system gain constants (Tcal and antenna efficiency).

    For this, the full switched power values were applied to the 8-bit 
data, and 'CALIB' was run on the visibilities from 3C286.  If everything 
is perfect, the resulting gains will all be 1.0.

    They weren't.

    But for most antennas, they were quite close.

    I plotted histograms of the voltage gains for the central SPW for the 
two IFs in the 8-bit data, each polarization separately.   Note that the 
conversion from correlation to calibrated visibilitites uses the Tcal 
and efficiency values which are externally determined (i.e., they are 
tabled quantities).  The observed gain values will be greater than 1.0 
when the tabled Tcal is too low, or the tabled efficiency is too high.  
In other words, if the corrected visibility is too low, then the actual 
Tcal is higher than that listed, and/or the actual efficiency is lower 
than that listed.

    a) X-band.  For both frequencies (8.3 and 11.3 GHz), the mean gain 
values are between .99 and 1.01.  The spread is quite small (about 
0.12).  One antenna is seriously discrepant -- ea06, whose gain value is 
about 1.3.
    For the higher frequency, the gain voltage spread is much wider, with 
the histogram very skewed to the high side.  Antennas 6, 21, 28, and 1 
all have abnormally high required gains.  Due to the skew, the median 
gains are 0.96 and 0.98 for RCP and LCP, respectively.

    b) Ku-band.  The means and medians are between .98 and 1.02.  ea06 is 
again discrepant.  The spread in voltage gains is 0.1 to 0.2, with the 
higher spread at the higher frequency.

    c) K-band.  The means and medians are again .98 to 1.02, for both 
frequencies.  The spread is very small at 19 GHz (0.1) slightly higher 
at 25 GHz.  ea06 is better behaved here, but still too high.

    d) Ka-band.  At 32 GHz, the medians are about 1.01, but the means are 
higher, as there are some antennas with high corrections required:  
ea25, ea01, ea19, and ea21.  All are above 1.2.
    At 37 GHz, the situation worsens.  The medians are about .98, but the 
same antennas as listed at 32 GHz are quite discrepant.

    e) Q-band.  The trends noted above continue.  Means and medians are 
both above 1.0 -- and antennas 1, 6, 19, and 25 are truly bad.  (They 
also have very bad vertical beam profiles).  Excluding the obvious 
deviants, the mean corrections at 42 GHz will be very close to 1.0, but 
at 48 GHz, it is about 1.1 -- a 20% error in flux.

   By adjusting the Tcal or efficiency values, the spread in the gain 
voltages can be greatly reduced, the medians (and probably the means) 
brought to within a couple percent of 1.0.  The trick will be to 
determine these adjusted values as a function of frequency.  There are a 
lot of frequencies!