Chaps,
this is an interesting thread, I hadn't given it much thought before.
As Peter at least knows, something similar, but with apparently
opposite conclusions, has been going on over on the IVOA DAL list.
The message below contains some of Doug Tody's reasoning for
encouraging BINTABLE rather than 1D FITS as a spectral representation
format:
http://www.ivoa.net/forum/dal/0906/1379.htm
The debate seems to have stalled with Doug getting the last word
as usual. Do you lot think this is badly wrong? Though I don't
relish the idea of locking horns with Doug once he's made up his mind,
if 1D FITS is in some or most respects better than, rather than just
an alternative to, BINTABLE FITS for this purpose, it may be worth
making this argument on the DAL list, as I don't think it's been
made before now.
There is already (assuming Petr Skoda knows what he's talking about,
which I imagine he does) some wish from the SSA providers to be able
to deal with 1D FITS rather than have to covert to BINTABLE for SSA
purposes.
Since Peter has been watching the DAL debate and has a better idea
of the underlying ideas than I do, he may be able to correct or
improve on my analysis above - go ahead if so.
To go back to a couple of David's specific points:
On Wed, 15 Jul 2009, David Berry wrote:
> Well I have done a few other things along the way. And paper III is
> not that hard really (except for the bit describing the tabular format
> :-) ). A lot of it is to do with describing what the axis values
> represent once you've got them. Which is all stuff that needs to be
> done whether you look values up in a table or calculate them from a
> known expression. Which brings me back to my twice stated question -
> how do you know what the values in the table represent? Is there some
> associated STC in there?
the answer to that, if there is one, must be in the Spectral Data Model
somewhere (http://www.ivoa.net/Documents/latest/SpectrumDM.html).
> > Using WCS presumably saves you a factor of ~2 in file size - does it
> > have other major benefits?
>
> For large spectra it's a lot more than a factor of 2. It could be hundreds.
Can you explain this? I thought a BINTABLE just had two values
(wavelength and flux, or similar) for each data point while a 1D
array has one (flux, with wavelength implicit from the WCS and pixel
index). Clearly I'm missing something.
Mark
--
Mark Taylor Astronomical Programmer Physics, Bristol University, UK
[log in to unmask] +44-117-928-8776 http://www.star.bris.ac.uk/~mbt/
|