ranges

[Doug Tody]

Another existing example of this, similar to the CVO model is the simple
spectral bandpass model used in SIA.  This defines the following to
characterize a spectral bandpass:

    ID, Unit, RefValue, HiLImit, LoLimit

where the bandpass is characterized numerically (and reliably) by the
reference or characteristic value and the range (high and low limits).
ID is undefined by the standard but is usually the informal filter name
or bandpass name used by the institution that took the data.  Unit is only
allowed to take on a couple of values since this is not a user interface.

This provides both short, familiar names for queries and so forth, plus
a numerical range to rigorously define the bandpass.  This same approach 
could be used for any similar quantity as with CVO.

Doug


> Surely an interval defines a bandpass better than a name? Of course,
> something even more complicated with the response function et al would be
> a better characterisation but much worse to query against...
> 
> 
> Clive responded:
> > Since I have argued in favour of the spectral ranges approach in the
> > past,
> > maybe I can explain it, and defend it somewhat.
> 
> In our CVO data model, we characterise every observation by a 
> "spectral_bounds" interval. This allows one to ask something like
> "how many observations have spectral_bounds that includes 
> $my_favourite_wavelength?" etc. The great thing is you are remaining in the 
> analog world rather than attaching names to things; in analog, an interval
> [300,330] (nm) and [302,332] are almost the same and both would be returned by 
> most exact searches (both would be returned by all suitably  fuzzy searches).
> But, they are not exaclty the same, so attaching the same name is wrong and
> attaching different names leads to picking one or the other OR horrendous 
> complexity as we manage 1000s of bandpass names.
> 
> Intervals are great to use beacause they can overlap and they are easy to
> index and search against, even in large DBs...