[evlatests] Flux Density Accuracy via Switched Power Alone

[George Moellenbrock]

Rick-

On Wed, Jul 10, 2013 at 2:27 PM, Rick Perley <rperley at nrao.edu> wrote:

>
>    5) The very high dispersion in the values, particularly at S band, and
> in the upper half of L-band, is a worry.  This is very unlikely to be due
> to efficiency errors, so likely reflects errors in the Tcals.  These are
> large errors -- tens of percent!  In general (except at Q-band), I'd have
> expected the dispersion in these values to be at the 5% level.  The best we
> have at present is about 10%, and 20% is more typical.  This is far too
> large for reliable gain calibration.
>

You've said precious little about the freq-dep Tcal catalog.  When he was
here, Andreas took some data with many subbands with different bandwidths
and LOs (now, far more common, I think), with some narrow bands falling
within
the wider ones.   This lead to distinctly different f.d. scales in these
overlapping
subbands, presumably due only to the excessive (and spurious) structure in
the Tcal spectra.   The upshot is that is gets particularly hard to even
quote the
likely systematic error in the f.d. scale, e.g., if you are applying
switched power
_and_ transferring residual gain calibration among subbands (wide to
narrow, eg).
Merely shifting an ordinary 128 MHz subband by a few MHz (read:  Doppler
setting
differences over months, or for different sources) is likely to tickle
the f.d. scale systematics excessively.  Also, I think the Tcals are
calculated according to
the subband edge freq (or equivalently, the Tcal catalog is so recorded)
such
that the correct Tcal is some function of what your chosen bandwidth is.

Got to fix up the Tcal spectra before you can say anything reliably
quantitative
about the blind f.d. scale, seems to me.

The peculiar gain table is just the normalization (to astronomer-friendly
units)
of your gain curves, and a version is already effectively in use,
hard-coded in
TYAPL  ("Rick Perley efficiencies"), and now also available via CASA's
gencal
(when generating gaincurve corrections).   Granted, it isn't antenna-based
there,
but the antenna-basedness isn't really the hard problem.  The problem is
the poor
(antenna-based) spectral Tcal info that makes it practically impossible to
generate any sort of maintainable peculiar gain table that can also be
called
reliable.

So, fix the Tcals so the switched power calibration delivers reliable K,
then
its easy, if you are clear (for programmers and users) about it:  decide
what factors
the "efficiencies" (Jy/K) account for (e.g.,  including which
mode-dependent correlator
efficiency factors and such, and w.r.t. the chosen gaincurve normalization
convention),
and make those  efficiencies available in a manner similar to gaincurves
and antenna position corrections (where they can be antenna-based and as
freq-dep as necessary, and where they will be updated on an appropriate
timescale and at decisive h/w events, etc.).   And _then_  you can
advertise a
X% blind f.d. scale where X isn't blush-worthy.

-George
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20130710/e2b034e7/attachment.html>