APFIT and possible approach to fibre crosstalk

Frank Valdes wrote on Oct 15, 1998

Q:  I am using APALL to extract fibre-fed spectra from a 2D image, and the
fibres are close-packed across the whole CCD. When one fibre is extracted,
there is a considerable amount of 'cross talk' from adjacent fibres.

I'm trying to come up with methods of minimising the cross-talk, and one
method is by multiplying the selected fibre image with a template image.
The template image contains the traced fibre curve with a gaussian width
equal to that measured with APALL from the flat field images (such as
those produced by mk2dspec) - in this way I hope to minimise the
cross-talk.

Do you have any routines that can extract the fitted fibre curves for a
given aperture from the database/ directory? I have a feeling I'd be
re-inventing the wheel by writing such a routine!


A  The task APFIT fits profiles to the data using the defined apertures
and outputs the fit, difference, or ratio.  The trick here is that you
can specify a subset of apertures (in V2.11).  So you could either use
APFIT to output the fit for just the aperture you want or output the
difference specifying all apertures but the one you want.  In the first
case you will get a smooth profile for the fiber your want and in the
second you will get the actual data fiber with some model subtraction
of the remaining fibers. 

The profile fitting is NOT a functional fit.  The profile shape can be
anything.  Thus the idea of a FWHM is not really applicable.  The "lower"
and "upper" aperture limits simply define the region of data to be use;
both for the profile modeling and for extraction.  (Note that for
profile centering and tracing the width parameters need not be the same
size as the aperture limits and these parameters specify something like
the full width at the base of the profile.).  The profile modeling is
described in "help approfile".  Basically it works as follows.  A simple
extraction is done to get the 1D spectrum which then defines the profile
amplitude at each wavelength (line/column).  This 1D spectrum is divided
into each point parallel to the dispersion (across the profile) to
normalize the profile at all wavelengths to the same value.  Then smooth
low order functions are fit parallel to each line/column parallel to
the dispersion.  The low order function allows for slow changes in the
profile shape with wavelength (as well as accounting for small tilts
relative to the image lines/columns).  The final model is then formed
by multiplying the function fit at each point across the dispersion by
the 1D spectrum scale factor.  There is some iteration and cosmic ray
rejection that can go into the loop.

This is a little hard to describe without pictures.

Since there is no profile fitting (which would be a different approach to
profile weighting and extraction not used in APEXTRACT) don't assume that
some kind of deblending is going on.  The profiles you get from APFIT
are basically a smooth representation assuming some slowly varying behavior
in the profiles with wavelength of the actual data.

---

Last post on Oct 15, 1998