[evlatests] SysPower calibration of amplitudes and weights

[Vivek Dhawan]

Time for a status report after playing with some datasets. Things are pretty
close now, and should be widely tested.

Amplitudes are good, once the 18% inflation is corrected (email yesterday).
The average flux may still be a few % high, but this is getting into details.
C,X and K bands were tested. C & K were also compared to the VLBA total-power
flux densities (3C84 = 16 and 21.8 Jy at 5 and 22.5 GHz), and agree within a
few % with the traditional E/VLA route of SETJY and CALIB.

Tsys also look good, barring the usual small handful of oddities - a bad
receiver, a problem with switched power (e.g. ea12B), or the antenna happens
to mis-set its power level in the set&remember scan.

Data weights are also truly (1/sigma^2) in Jy^-2, within a few % of hand
calculation (based on the antenna efficiency table that TYAPL is using).
This is true when there is no time or frequency averaging done. But there
is an odditity in the order of operations - TYAPL applied before SPLAT
gives the correct weights, scaled by (N_chan * integ_time). When applied
after, the weights are too high by a factor N_chan. This could be something
I am doing wrong, but watch for it. The amplitudes are identical either way.
The weights are the same on a blank field or on a modestly strong source.

Here is the outline of processing.

BDF2AIPS - Load data including all tables. (CASA route in testing?)

SNEDT - apply UV data flags from FG table to SY (SysPower) table. This does
         not work perfectly, I'm not sure why, it could be the time stamps on
         the syspower data, or something else. Further editing of the SY table
         is usually needed.

FRING, BPASS, SPLAT - delay, bandpass (no normalization), time&freq averaging.
                       If using SPLAT before TYAPL, see note above.

UVCOP - to fork into 2 paths - one to calibrate the traditional way, one with
         TYAPL

Amplitude calibration a). TYAPL.   b). SETJY, CALIB, (getjy).

DIFUV - this lets you divide 2 datasets, and check the calibration factor
         on an individual baseline/subband/baseband. It is clear that the
         current scheme of using a single average antenna gain (Jy/K) is
         only good to about 10% on an individual baseline. To get better,
         we would need to measure the SEFD on an antenna/IF/subband using
         the same Tcal that we calibrate with. This makes for much book-
         keeping so its not something I expect we'll do soon. It is what
         we do on the VLBA (we keep a matched set of Tcal, Tsys and SEFD,
         but not a whole spectrum of each.)

Vivek.