[fitsbits] FITS for DSN spectral line observations

[Thierry Forveille]

Tom Kuiper writes:
 > As far as I am aware, the software that we currently use -- Velu's data
 > conversion program, COMB, and CLASS -- understand only simple FITS files.
 > - Am I correct in that?
No: I don't know about COMB, but CLASS does definitely support the SD-FITS
convention for binary table use. Actually, that convention was first 
validated through an interchange between CLASS and an NRAO package (UNIPOPS
I think). Jeff Mangum was my NRAO counterpart during that process and
may have additional useful advice. That capability of CLASS has not had
a widespread use (as far as I can tell, i.e. from the lack of bug reports),
but is defintely there. It may on the other hand have a few remaining 
minor bugs (I am almost sure that some ancillary keywords are ignored), 
that I'll be happy to fix if/when you find some.

 > The proposed Single Dish Format (SDF) is based on the BINTABLE extension.
 > One might think of this extension as a spreadsheet, with the additional
 > feature that a column can be a data array.  A typical structure consists
 > of a series of rows, and the fields in each row contain some of the variable
 > names and values that would appear in a simple FITS header, and the last
 > "field" would contain the data array.  This type of structure is more
 > efficient (for example, the blank space in standard headers is avoided), and
 > many scans can be included in a file. 
 >
My advice would be to use it rather than the IMAGE extension, on grounds of 
both elegance and efficiency.

 > There would have to be some trade-off 
 > between parameters specified in the main header (those that seldom change)
 > and those that are specified in the table (like scan number and time).  Any
 > time you change a main header parameter, you'd have to start a new file.
 > 
If you use this format for (near) real-time application this is correct. If
you use it as an interchange format only (as we do from CLASS), or know what
your real-time system will do next, you can figure out which of the parameters
can be factorized in the main header and which need a column.

 > The drawback to the SDF format is that I doubt that any package except
 > AIPS++ or UNIPOPS could process SDF files. 
As mentioned above CLASS can process them too.

 > It's also worth pointing out 
 > that the standard was drafted in 1995 and is still a draft, so I don't 
 > think that there is a real ground swell here. 
I agree that this process has lost some momentum, but there is nonetheless a 
significant base of programs that understand it. Adding this capability
to others is fairly easy (a factorising SD-FITS writer is quite some work,
but the reader is not), so you should probably talk to the authors of any
package that you plan to seriously use, and see if they'd be willing to do 
that work.

Thierry