Improvement for eShel user’s
1. More options for optimal extraction
ISIS implement now full K. Horne algorithm (K. Horne, 1986, PASP, 98, 603) in an adapted form for extract the optimal spectrum profile from the 2D spectrum image.
For this, from the eShel tab (Instruments section), select the Optimal binning option. ISIS can « clean » profile by rejection of comics and deviant pixels (hot pixels, trap, …). Look the eShel Setup dialog box for the tuning…
For a normal work, adopt a 3-sigma to 10-sigma rejection coefficient for exclude bad pixels. Also, if the (reciprocal ) gain of your electronic camera is know (in electrons per ADU), enter the value (ISIS use these information to estimate the photon noise is your 2D image). If the gain is unknown, enter the value 0 for the gain. Example:
The electronic gain of ATIK460EX camera, often used with eShel spectrograph, is 0.27 e-/ADU. The ASI1600MM camera gain, also a good option with eShel, is 0.48 e- / ADU (20 dB setting).
For deselect deviant pixels rejection enter a large value for the Rejection coefficient, for example 1000, and a null value for the Camera gain .
You can also deselect Optimal binning option. In this situation, ISIS calculates the spectral profile by a simply addition of the the pixel intensities (not weighted) along a spatial axis range (see parameter Binning width zone in the Setup dialog box).
For the example below, the Optimal binning option is deselected (a simple arithmetic addition is used along the binning zone for construct the spectral profile) :
Next, the Optimal binning zone option is selected (reduction of the noise), but the deviant pixel rejection function is not concerned (1000-sigma rejection + gain = 0) :
Finally, a 10-sigma rejection is activated (the bad pixels are erased) :
Detail of the R Aqr spectrum, before and after optimal extraction (note the reduction of the noise and the cleaned bad pixels) :
2. Consideration about the spectral sampling
For optimal binning, it is mandatory that the spectrum be correctly sampled. In the previous examples, the spectrum is taken with CMOS detector with small pixels (3.8 microns, camera ASI1600MM), in binning 1x1 and by using a 135-mm focal length camera lens. Under these conditions, the sampling is about 10 pixels per FWHM (i.e. 10 samples per resolution element). A generous sampling but a fine choice for application of profile optimal extraction methods.
Problems occur when the spectrum is poorly sampled. In the example to follow, the width of the order traces is less than 2 pixels and the spectral sampling is 2 pixels per FWHM, the critical limit of Shannon theorem. The echelle spectrograph is here directly attached at the telescope and the stellar image is very sharp. In addition, the 2x2 binning detector mode is used, which is really not recommended in association of an echelle spectrograph, in which the data must undergo geometric transformations. In the 2D image, the orders limit appear immediately irregular because of the pixelization effect, which is not a good sign.
Here, a detail of the corresponding spectral profile :
The Optimal binning option is unselected, the only reasonable solution to work here safely.
The Optimal binning option is selected, but the deviant pixel rejection capacity is not used. You have to be careful, even if the result seems correct here. Note that the gain in SNR is not significant because this SNR is already high (the optimization only works on low SNR spectra).
An attempt to correct cosmetic defects produces a catastrophic result here, with the appearance of a considerable structural noise associated with each order. This is the limit of sub-sampling decision.
The problem concern also eShel equipped with an Atik460EX in bin. 2x2 mode (9.08 microns equivalent pixels size). Once again, I recommend working in binning 1x1 if you want to exploit all the possibilities of ISIS (rejection of bad pixels if necessary). Contrary to popular opinion, the difference in detectability between 1x1 bin mode and 2 x 2 bin mode is quite negligible when using an echelle spectrograph. The 1x1 bin is even better thanks to the reduction of sampling noise in the general case and give a better spectral profile restitution (reduction of pseudo noise). I propose to test by yourself the various options that arise.
3. Median filtering
A median filtering of 2D images is implemented for remove small scale noise (hot pixels residual, telegraph noise if you use CMOS sensor, …). The option is only valuable if the spectral sampling is up to 5 pixels/FWHM (for example is you use AS1600MM 3.8 microns pixels CMOS camera in association with eShel spectrograph).
Aspect of ThAr image when using an ASI1600MM camera + 135 mm chamber lens. The core image of the 50 microns fiber is easy to see. Note the presence of the geometric distortions (part of « image_sum1 », see section 5). In this particular situation, median filtering can be used.
4. Order merging method
The association of orders to produce a complete spectrum is always a delicate problem (note: for a precise exploitation it would be better to exploit the individual orders, not the merged spectrum).
Normally, ISIS uses the common parts of successive orders to harmonize their intensity. When the signal is near to zero value, or equal to zero (the case for example for cold stars in the blue part of the spectrum), the noise that dominates and the procedure fails. The result is of very poor spectrophotometric quality.
For resolve these difficult situations, a new ISIS option is implemented to connect the orders without using the common parts. There is no calculation made: The quality of the merging operation is based only on the signal from the tungsten image and the blaze function correction of each order. The result is less precise for high SNR spectra, but more efficient for quantify the spectral information contain on very low SNR data. To use this new method, simply unselect the Order merging adjustment option.
Here, the result with successively Order merging adjustment selected, and Order merging adjustment not selected, for the situation of a symbiotic star (R Aqr), with a nearly zero continuum signal detected at 4000 A, except emission lines :
The relative signal is artificially increased in the blue part of the spectrum in the first case because the presence of a high noise in the common part of the orders (the opposite situation can be also produced, with an attenuation of the blue part compared to the rest of the spectrum). The relative intensity of the details along the spectrum is much better respected in the second case.
Look for example, the relative intensity of the Hgamma (4340A) and Hdelta (4102A) emission lines: the ratio is much more correct if ISIS do not adjust the inter order signal.
5. Faster processing
The normal way for ISIS is to extract the spectra of the individual images from a sequence (one profile per image), and at the end, add those profiles.
ISIS offers a new possibility, the Sum before option, which consists of adding first the 2D images (after preprocessing) and calculating the final profile on this sum image (ISIS then produces in the working directory the 2D image « image_sum1 », that you can view).
The first method is in principle better if we consider the final noise, but the difference is small in general (do comparative tests). The second method is significantly faster.
6. More control
If you select Zone traces option into Setup dialog box, ISIS add some information on the intermediate 2D images produced (calib2_xxx, flat2_xxx, #1_xxx, #2_xxx, …) for a better control during tests phases (unselect the option for a normal work) :
Example, a part of « calib2_31 » rectified image of Thorium-Agron spectrum at ordre #31:
The width of the binning zone is an important parameter to be determined, not too narrow to accommodate the maximum signal of the object, nor too wide not to add too much noise (see the Binning zone width parameter in the Setup tab ).
Here the binning zone is too narrow…
and here is too wide…
Remember: To access at the control images of the object itself, set the Erase automaticalliy intermediate files option to No in the general Settings dialog box. Here the example of the geometrically corrected 2D image of the order 34 for R Aqr spectrum, i.e. image file « #1_34 » (camera ASI1600MM, chamber lens Samyang 135 mm f / 2, Sum before option + Median filtering) - the Halpha line is near the center: