Stange results from ISIS

[Tim Lester]

HI all,

This post is directed mainly to Christian.
I have encountered some very strange behavior from ISIS when processing my echelle spectra. I have been working on this for quite some time to make sure I wasn't doing something silly.
The problem I am about to describe occurs randomly and I can not figure out why.

The following comparison illustrates the problem

full spectrum comparison.PNG (60.76 KiB) Viewed 3260 times

This is a comparison of the spectra for HD123299 taken several nights apart. This is the star I use for calibrating response when shooting AG Dra.
The star is bright so there is only one exposure in each case. Notice the regions of high noise on the blue spectra. Here is a comparison for order 32 to show the effect in more detail.

order 32 comparison.PNG (54.14 KiB) Viewed 3260 times

And for order 51 where the order spacing is much larger.

order 51 comparison.PNG (52.25 KiB) Viewed 3260 times

No reasonable adjustment of ANY of the parameters in ISIS will change the result. The new Optimized binning flag does affect the ripple on my flat traces but has no effect on the regions of very high noise.
If you look at the #1_32.fit image that ISIS produces the noise is clearly not present (this can be confirmed by generating a profile using Maxim DL)
Further if you use the command L_OPT (or L_ADD) to process #1_32 you get a clean profile over a wide range of binning values (12 and 24 in this case) as shown in the next comparison.

order 32 L_OPT processing.PNG (51.93 KiB) Viewed 3260 times

As I said at the beginning this problem occurs randomly and I can not see anything in the data that would explain it.
I have discovered a workaround that allows me to process spectra. What I do is divide all the images by 10 (including the dark and bias images). These are saved in floating point format so no information is lost. It is simply a scaling. When these scaled images are processed in ISIS everything seams OK. Here is the comparison for order 32 processed in this way. The div10 profile has been multiplied by 10 for this comparison.

order 32 div10 comparison.PNG (49.74 KiB) Viewed 3260 times

This is the worst kind of random problem to troubleshoot so I have uploaded my working directories for the normal and strange HD123299 spectra to the following link: https://1drv.ms/f/s!AvXqhKI5kEGRlhwmjuu5rQPxM00r
If the downloaded directory "isis_problem" is place directly in C: then the xml files should work correctly. The download is rather large 130 Mbyte.
I hope this information is helpful.

Tim

[Christian Buil]

I can generate the problem from your data - the first step to the solution...

But I have a preliminary question. Your spectra traces are very thin. I measure a FWHM wide less then two pixels (along the spatial axis). It is a potential critical point in regard to the ISIS optimal extraction processing (see Nyqvist criteria - your seem undersampled). My question : the camera is used in binning 2x2 or 1 X 1 (can you precise the camera and detector type).
I recommend very well use of a 1X1 for all the situation of course.

Christian

[Tim Lester]

Christian,

You are right about the sampling. I am slightly under sampled for some of the orders. To maximize S/N I designed for a sampling just over 2 for 2x2 binning and a 23 um slit but the actual line widths seam to be a bit better than the design goal. I should probably bin 1x1 for brighter objects. When seeing is very good the FWHM of the traces in the cross dispersion direction can become less than 2 (for 2x2 binning).
The camera is a QSI583.

Tim

[Robin Leadbeater]

I know this is about undersampling in the non dispersion direction but there is a very interesting paper just published on the effects of bin size sampling on errors in spectroscopy.

Detector sampling of optical/IR spectra: how many pixels per FWHM?
J. Gordon Robertson
https://arxiv.org/abs/1707.06455

(Thanks to Christian Netzel on the VdS forum for highlighting this)

Cheers
Robin