Trouble processing 2x2 binned images in ISIS.

[Keith Graham]

Hello.

In taking spectra of dim targets (such as Nova Cyg 2014) with my Alpy600, I saw that the snr was low when comparing my results with others. At the advice of Francois Tessyier, I tried binning my images 2x2, something I had never before tried. Sure enough, the images are coming out much cleaner, however, when I process those binned images in ISIS, I notice that the value for the X coordinate of the 5852.49 line in my neon spectrum shifts from 407 (as determined by clicking on that line) to 470 when using the calibration assistant. This results in a high calibration error (around 22). I cannot see how to prevent this "jump" from happening.

I am applying 2x2 binned flats, darks, and cosme files to the target images, so those should all be good. This issue does not occur when using 1x1 binned images as my rms values are then generally around .2-to .3. Is there something I need to do in ISIS when binning images that will keep this from happening? The only change made is that I am now trying to process 2x2 instead of 1x1 images.

Cheers,

Keith Graham

[Francois Teyssier]

Hello Keith,
Have you changed the value of the pixel size (x 2) in main page ?
François

[Keith Graham]

Hi Francois,

Well now, this is a great big DUH on me. I knew I was missing something, and increasing the pixel size was it. Actually, I did need to tweak the size a little after doubling the pixel size from the 1x binned size, but it now works just fine. Thanks a bunch.

Keith

[Miguel Rodriguez]

Regarding to binned spectra images, ¿is it worth the gain in sensitivity when taking spectra images in binning 2x2 mode, despite the loss in resolution?
Tonight I tried nova cyg in 2x2 and I could confirm a much better recording of spectrum, but with a lesser R, =183.

[Robin Leadbeater]

Hi Miguel

It depends on the size of your pixels relative to your slit width. For example with a 23um slit and my 6.45um pixels I can 2x2 bin and the resolution hardly suffers because I still have almost 2 binned pixels in the width of the slit (You need at least 2 pixels/slit width to get the maximum resolution available.) You should not go much below 2 binned pixels/slit width though as undersampling can cause problems.

The increase in signal/noise you see with binning is partly an illusion though. For example you would see almost the same improvement if you took a spectrum unbinned and then combined the bin values in pairs afterwards. (For this reason SNR should really be calculated and quoted per resolution interval not per bin) There is a small real advantage to binning in the camera though as you get sqrt(1/4) = 1/2 the amount of camera noise (one dose of camera noise for every 4 pixels instead of 4 doses) but with modern cameras that can be quite low anyway.

Cheers
Robin

[Miguel Rodriguez]

Thanks very much for your answer Robin.

[Robin Leadbeater]

Actually I see I made a small mathematical error (now corrected) in the original post. The read noise is reduced to sqrt (1/4) =1/2 when binning 2x2

Cheers
Robin