[evla-sw-discuss] Synchronization of switched power & visibility data

[Michael Rupen]

The switched power calibration discussion left a couple outstanding
implementation questions.  One of those was the synchronization of the
switched power and the visibility data.

We agreed that we should at least initially write out the switched power
data Pdiff & Psum with the same time resolution as the visibility data,
with the proviso that the minimum switched power dump time must 
sample both the Tcal-on and the Tcal-off states. With the current noise
tube cycle this yields a minimum switched power dump time of ~50 msec. For
OSRO the visibilities are dumped at most once per second, so that's the
fastest rate for Pdiff & Psum as well.  This requires that the Station
Boards know the final dump time out of the CBE, average Pdiff & Psum
over that interval, and send those numbers to MCAF.

The next question is whether we have to dump Pdiff & Psum and the
visibilities at the same time: i.e., must they be synchronized?
This would require that the StB know when integrations start and stop --
i.e., they would have to know t (say, the start of a subscan) & dt
(the dump time) rather than just dt.  That's rather more painful.
Ken has suggested that we adopt a zero of time (midnight UT) and align
everything with that, in part to make this easier; Bryan feels this would
crawl up and bite us in the future, and would like to avoid the
synchronization question entirely if we can.

We use Pdiff & Psum to scale the visibilities and the weights, and as
diagnostics (e.g., to flag the data without poring through all the
visibilities).   I think we *do* want to label Pdiff & Psum with the
correct times, but we do *not* need to synchronize those times with
visibility dumps.

- For scaling purposes this requires that the scale factors do not change
   much from one integration to the next.
   (1) Generally this should be true -- if it's not, you should be using
     a shorter integration time!
   (2) The worst case is when the scale factors are half for source A and
     half for source B.  I think those we just flag and move on, as we do for
     VLBI currently.
- For diagnostic purposes we mostly want to know whether the data are
   any good (e.g., do they match our assumptions that nothing changes within
   an integration).  If we don't synchronize the signals we may over-flag
   the data a bit but that seems a minor problem.
- Obvious worries include RFI and the Sun, which can do almost anything;
   but RFI leads to bigger problems than this, and we're putting off the
   Sun until we have someone around who understands it.

Note too that, if we decide we're very worried about any of this, we can
always dump Pdiff & Psum more often than the visibilities, and minimize
the disagreement in their dump times in post-processing (hi George,
Eric! ;).

The SDM does NOT require that calibration data and visibilities be on
the same time grid.  I believe CASA & AIPS do not require this either, but
maybe GEORGE and ERIC can comment.

I see two flies in this brand of ointment:

1- We do want proper time stamps on Pdiff & Psum.  "Proper" here
   means, in the same time system as the visibilities.  Is this easy
   to figure out on the StB?

2- The measurements on the individual StB may not be made at the same time.
   Even measurements at various filters on a single StB may not be made
   simultaneously.  The SDM can handle this but it may be a pain to
   deal with this approach in post-processing.  As suggested above we
   could over-sample Pdiff & Psum and interpolate to a common grid, maybe in
   the filler (or even in MCAF???), but that's pretty yucky.  Alternatively
   we can ignore the disagreement.  Others (Ken?) will have to comment
   on how big these time offsets might be.

Comments? Counter-proposals?  Rotten eggs?

                 Michael