Simple spectral model revision

[Ivo Busko]

I think there is an inconsistency between the VOTable and
FITS serializations described in the IVOA SED Data Model
(v 0.93).

The VOTable serialization allows the option of writting
data for each individual segment, with its own set of
physical units (for both spectral coordinate and flux
values). 

The FITS serialization apparently does not allow for that.
Multiple rows in a table (meaning, multiple segments) are
required to share the units instance associated with each 
column.

When writing a SED assembled from multiple data sources,
I believe it is desirable to keep each segment in its
own "native" units, instead of transforming everything to
a common set of units. And what that common set would be?
It is not specified in the data model to begin with.

Note that the SED model allows for independent UCDs for
each segment/row (SPO_UCD and FLUX_UCD columns. However, 
assuming that UCDs can change from row to row, doesn't mean 
that units have to change from row to row as well?

A possible solution would be to add extra columns to the
SED model in order to hold the units. However, this might 
violate the FITS standard that requires units to be a part
of the column dscriptor itself. Or am I wrong in assuming
this?

Any ideas?

Regards,

-Ivo Busko