A Virtual Observatory Data Model

[Frank Valdes]

Ed,

I am going to agree with you in part, define some areas as calibration
issues, and throw out some ideas in response.

I like the idea of a WCS object.  The FITS WCS effort has led the way.
But their goals have been focused on the zeroth order aspect of WCS,
namely a point transformation from an index to a coordinate.  The WCS
data model needs to be expanded to higher order information.  The next
order would be to define pixel/bin extents.  One can do it, sort of,
with the scheme of saying the boundaries are half values of the index
but this is only an answer for certain types of sampling detectors that
have no gaps between the sample bins.

You identify another important aspect, resolution.  This is different
than coordinate uncertainties and bin sizes.   FITS WCS has tried to
define uncertainties somewhat.  Coordinate uncertainites tell you how
well you know where your sample bins are while resolution is, as you
say, how well you know photon information after the binning has lost
the precise identity of the photon.  I agree that this has to be a
higher order part of a data model for sampled photon count data.  This
is not, IMO, a calibration issue but an intrinsic property of the
sampling process.

There is more than one kind of crosstalk.  One is what you describe and
has the interpretation that you gave, that a photon which actually fell
in one bin "leaks" into another bin.  The other is one that I have to
deal with in mosaic data where a photon in one bin is counted but it
generates a small fraction of a spurious count in another bin which is
spatially disjoint (in another detector in fact).  Maybe the two could
be unified by some crosstalk coefficient but I'm not sure at the
moment.  Describing crosstalk seems to me to be a difficult thing to
do.  Note that I would argue that this is a instrumental problem that
should not be present (other than identified in an associated data
quality/uncertainty component) in calibrated VO data.  For our CCD
mosaic data we attempt to remove crosstalk effects in our calibrated
data.  I'm not sure how photon diffusion in a digital detector
could be calibrated and might have to be treated as a resolution effect.

While I was trying to develop the data model ideas that I presented I
was considering the idea that one should provide a "filter" function
for each bin.  There seems to be two filter functions, one based on
position in the 4D bin and the other in the 4D space.  In the first
category would be things like variable pixel sensitivity.  Maybe this
type of filter should be something that must be removed in VO
calibrated data?  In the second category is the usual filter concept
for a single wideband bin; ie. an image.  It might also include DQE in
energy and time windows in chopping or paused integrations. (Note I
think generally the DQE question goes away as part of the calibration
process.)  The resolution description could also be a part of this
second type of filter.

To try and summarize the minimum WCS-like functions to describe sampling bins
for a simple class of data:

bin center coordinates as a function of bin index
bin width or edge coordinates as a function of bin index
bin resolution, or filter function in 4D space, along each dimension
   as a function of bin index
    
So I agree with you, resolution is another important component of even a
simple data model.

Tracking a detector on a moving object is a very interesting thought.
Indeed, in the 4D binning model that means that over the course of the
collecting (the integration) the binning is changing.  If we wanted
this type of data to fall into the same VOCLASS as celestially tracked
data then it would require some time dependence in the bin geometry.
Describing this indirectly through the orbit of an object or by some
ephemeris would need to be decided.  In my view of data model classes I
might advocate that integrations along non-siderial tracks should be a
different class or subclass of data with specialized metadata.

Frank


> From edward.j.shaya.1 at gsfc.nasa.gov Mon May 12 08:52:29 2003
> 
> Frank,
>     I think this is the right direction.  I just want to add a few items 
> that the scientist, even the non-instrument scientist, needs to have but 
> I think are missed here.
> 1. resolution - Although photons fall into bins or pixels, the 
> resolution tells one the probability that a photon in a particular bin 
> could have fallen into a nearby bin.
> 2. crosstalk - There is some tendency for the liberated electrons to 
> leak to the adjacent pixel.  So, even when pixel size is much larger the 
> spatial resolution the photon has a distinct probability of registering 
> in the wrong pixel.
> 
> I believe that time-series data is also covered by this model.
> VOCLASS = 4DBIN.TIMESERIES
> 
> Finally, what do we do about observations on a moving target (asteroid, 
> planet, comet, perhaps even high proper motion star)?   I think there 
> should be an option to substitute an object name for the two spatial 
> coordinates (unless you want to define a path on the celestial sphere).
> 
> Ed