spectro-aras.com

The question raised by Stephane Charbonnel (binning 3X3) leads me to raise the opposite question: what about oversampling? I have had this question in mind for a long time but never found the opportunity to raise it. With my question, I do not mean the situation where the pixel size is small compared with the slit width (the converse of the situation described by Stephane). I mean the following: I make extensive use of interpolation in my processing algorithms. This may lead me submit to Bess a profile with, for instance, 5000 samples whereas my CCD Chip is 1000 pixel wide. I could of course convert (by interpolation) my 5000 sample profile so as to match the CCD characteristics but I suspect this conversion to degade the result.
Can we envisage to submit to Bess profiles involving much more samples than there is in the CCD chip?
Patrick Lailly

Patrick,

Consider a calibration line of FWHM pixels wide (neon line for example).

The best sampling (during acquisition, at the detector level - important) is between 2.5 and 3.5 time the FWHM.

For value superior to 3.5 x FWHM the oversampling can start to degrade the signal to noise ratio (slowly).
An oversambling superior to 4.5 is considered excessive.

Do not mistake with numerical oversampling (during processing phase). A quality oversambling of raw spectral
profile (spline function, sinus cardinal interpollation, ...) is not a source of quality degradation and can help in some
circonstance for a better process (minimisation of correlated noise from sample to sample).

Suppose a observampling of 5 x FWHM during processing. Submit and analyse the spectrum with these oversampling,
do not reduce (never ! - because you add possible numerical noise).

Christian Buil

I roughly agree. Thanks Christian.
I will not miss submitting reasonably oversampled spectra to Bess.
Patrick