Jump to content

4xPropeller, 2CCD, 2CMOS, Ha


vdb

Recommended Posts

So I did some testing of different camera's an telescope setups, 

ODK 10 inch SX H36 (3.5nmHa baader) 47x15min versus

FSQ 106 ASI 183 (5nm Ha Astrodon) 30x5min

+- same Arcsec/p but huge difference in Aperture

 

GSO RC SX814 (3.5nmHa baader) 27x15min (less moon the other 3) versus

GSO RC 8 QHY 163 (3.5nmHa baader)40x5min, same telescope and about same Arcsec/p

 

The ccd's have 2 to 4 times more data and are taken with 15min subs, the CMOS data 5 min subs, the FSQ was at a true dark site but was taken when full moon was up but was really cut short only 30x5min... The ODK was a mix of conditions and also lots of moon, also the only one with flats taken to correct vignet.

All stacked with same settings (APP), screen stretch in PI, save.

 

42200471355_616566f1fd_c.jpgODK10-H36-47x15min 3.5nmHa by Yves, on Flickr

 

43105319421_4560b07980_c.jpgFSQ106-ASI183-30x5min 5nmHa by Yves, on Flickr

 

29309873358_2ed0ee2587_c.jpgGSORC-SX814-27x15min-3.5nmHa by Yves, on Flickr

 

42277149075_20d74d7dfe_c.jpgGSO RC8-QHY163-40x5min-G30-3.5nm by Yves, on Flickr

Link to comment
Share on other sites

Interesting comparison, although I have tough time assessing the differences in images.

Could you take 4 stack, align them to a certain base frame (select one that you see fit) and do frame equalization prior to doing the stretch? That should make difference in resolution and SNR more obvious.

One more thing, I really think you should do flat calibration on CMOS sensors regardless of vignetting / dust. There seems to be pixel to pixel variation in QE or per pixel AMP due to manufacturing / CMOS technology. I've seen this in few CMOS sensors and this can improve SNR quite a bit.

I just checked one of my master flat files for ASI1600. It is stack of 256 subs (simple average stacking). I've selected small uniform piece (so no vignetting or dust shadow) and did measurement on it.

Average value is ~2465e, while Stdev is ~24. For this signal level, after stacking 256 subs, shot noise std value should be around 3 (2465^0.5 / 16). This shows that not all pixels are uniform in either QE or amp and if not calibrated by flat, image will contain this additional noise.

Link to comment
Share on other sites

Stacking is less effective for CMOS cameras as the noise is not Gaussian. It tends to be dominated by 1/f - telegraph noise.

I am not sure the standard advice to use more shorter exposures with CMOS is valid.

There is, as you say, also more pixel to pixel variation.

Although I have noticed this seems to have increased in CCDs too as the dark current has been lowered. Not that I have any real data to back this up. It just seems there are more warm pixels these days. However they may just have been uncovered by the lower dark current.

Regards Andrew

Link to comment
Share on other sites

5 minutes ago, andrew s said:

Stacking is less effective for CMOS cameras as the noise is not Gaussian. It tends to be dominated by 1/f - telegraph noise.

I am not sure the standard advice to use more shorter exposures with CMOS is valid.

There is, as you say, also more pixel to pixel variation.

Although I have noticed this seems to have increased in CCDs too as the dark current has been lowered. Not that I have any real data to back this up. It just seems there are more warm pixels these days. However they may just have been uncovered by the lower dark current.

Regards Andrew

Do you mind elaborating on telegraph noise? I've seen this mentioned, but not even sure what it means in context of image data?

Is it the case where there is variation in pixel "hotness" over subsequent subs (pixel is hot in one sub, then goes cold for few subs, then "turns on" back again, and switches like that)? If that is the case, I would say:

a) it is by no means dominant thing

b) sigma clip?

 

Link to comment
Share on other sites

1 minute ago, andrew s said:

I meant to ask compred to what?

Regards Andrew

Disregard that comment, since I had no clear understanding of what telegraph noise is.

If I understand it correctly (I had a brief look at first link / pdf you posted), characterization of telegraph noise is done by examining dark subs on pixel level (not each pixel is affected by this in the same manner) and looking at the histogram of single pixel over range of subs (darks? can bias be used or this effect depends on integration time?). In ideal case plotted distribution should be sort of Gaussian (read gaussian + thermal poisson). If there are more than one peaks present - we have telegraph noise, and pixel value "alternates" between different Gaussian distributions (sort of jumps in offset randomly)?

 

Link to comment
Share on other sites

3 hours ago, vlaiv said:

Interesting comparison, although I have tough time assessing the differences in images.

Could you take 4 stack, align them to a certain base frame (select one that you see fit) and do frame equalization prior to doing the stretch? That should make difference in resolution and SNR more obvious.

One more thing, I really think you should do flat calibration on CMOS sensors regardless of vignetting / dust. There seems to be pixel to pixel variation in QE or per pixel AMP due to manufacturing / CMOS technology. I've seen this in few CMOS sensors and this can improve SNR quite a bit.

I just checked one of my master flat files for ASI1600. It is stack of 256 subs (simple average stacking). I've selected small uniform piece (so no vignetting or dust shadow) and did measurement on it.

Average value is ~2465e, while Stdev is ~24. For this signal level, after stacking 256 subs, shot noise std value should be around 3 (2465^0.5 / 16). This shows that not all pixels are uniform in either QE or amp and if not calibrated by flat, image will contain this additional noise.

Interesting remark on the flats indeed it could play a role!

To what do I need to align the 4 stacks to? I'll take one of the 4 as a reference and do a histogram fit of the 3 stacks against the reference stack?

 

Link to comment
Share on other sites

18 minutes ago, vdb said:

Interesting remark on the flats indeed it could play a role!

To what do I need to align the 4 stacks to? I'll take one of the 4 as a reference and do a histogram fit of the 3 stacks against the reference stack?

 

Yes, exactly, choose a single one to be reference frame and do align (preferably choose interpolation that does not affect SNR too much - like Lanczos or Spline, avoid bilinear / bicubic) and then do normalization of frames - not sure what is the proper term for this but offset and scale needs to be the same (like it is normally done with single subs in preparation for stacking - to equalize background and signal level - I guess histogram fitting is that sort of thing).

2 hours ago, andrew s said:

Hi Vlaiv have a look at this http://harvestimaging.com/pubdocs/103_2006_dec_IEDM_random_telegraph_noise.pdf as a starter for 10. The is also the comments on it here http://www.astrosurf.com/buil/CMOSvsCCD/index.html where Christian Buil is comparing CCD and CMOS cameras available to us.

Regards Andrew

Indeed there is quite a bit of telegraph noise on some of the pixels. Here are four histograms for four different pixels (64 samples each). One has very pronounced telegraph noise pattern:

image.thumb.png.fb2216b2c69d6a3b910aceaa4867b11a.png

It is good to know, I just need to do a sort of large scale characterization of this phenomena, and see what is best solution to deal with this. Of the top of my head, one could examine dark subs and do normality test of sorts on each pixel thus creating "abnormal" pixel map. These could be treated the same way (or something similar to) as hot/cold pixels in calibration depending on "abnormal" pixel density. It also shows how important dithering is for imaging with CMOS sensors (provided that "abnormal" pixels are sparse enough).

Link to comment
Share on other sites

Hm, this is very interesting, here is Stdev stack of 64 darks from ASI1600 - just trying to find out if there is simple way to mark pixels as "abnormal" - it occurred to me that having multiple peaks is bound to produce higher stdev of samples than a single peak.

Enlarged and linearly stretched crop of stack.

image.png.68f6371470086c0e4c3e23c803983627.png

It looks like telegraphic noise predominantly appears in diagonally linked pixels? Defect must be located somewhere between pixels so it is affecting both adjacent pixels (maybe in altering manner?).

Link to comment
Share on other sites

valiav - nice work. I took a very pragmatic approach. As my ASI 1600 over-samples my spectra ( 10 pixels per resolution unit) I did a plot of  number of "hot" pixels v cut off level and picked a value for the cut off that removed the majority without going mad (not very scientific I know). I used 10 minute Darks for his as this is my normal exposre length per sub.

I then use the "clean" process in IRAF to replace them by interpolation from nearest neighbors before extracting the spectra. (In IRAF you can specify blocks of bad pixels as well as individuals.) If during the processing I notice a "hot" pixel I add it to the list.

If you find a better method I will try it out.

Regards Andrew

Link to comment
Share on other sites

Hmmm doing it the way you propose seems to be harsh, there is an enormous difference is pixel values between the ccd and CMOS camera's, it would inevitably favour the reference frame,

I'll take the one with the lowest value as a reference and one with the highest to see what the impact is

Link to comment
Share on other sites

I have not been able to draw any meaningful conclusion from this apart that ODK has the most bloated stars, GSO RC is second, and FSQ106 is in first place - very small difference between first and second place.

I've also noticed that images are quite "polluted". I don't know what is the cause of this, it might be due to not using flat calibration, LP levels (but since this is narrow band - there must be really heavy LP to show like this) or possibly high altitude clouds in each session that caused problems.

Here is what I mean by pollution, I binned down (to make it obvious) aligned images x8 and created animated gif out of your sequence. Notice how each frame differs in sort of "illumination gradient" - no two frames are the same.

 

Stack-3.gif

Link to comment
Share on other sites

heavy moon and difference in sky conditions, flats where used with the ODK and the FSQ with such a small chip is "quite" perfect but was taken at full moon.

I will see if flats make a difference with the GSO RC, but I doubt as the SX and QHY where taken with it and showed a difference as well, the baader 3.5nm has maybe some off band leakage as well. 

Anyway what I find quite interesting is that while on paper these camera's are day an night in real life usage there hardly is any difference, the illumination difference do not stem from the camera,

flat fielding in PI is not really an option here as most of the image contains signal, I'll try but I doubt it's possible ...

 

 

Link to comment
Share on other sites

8 hours ago, andrew s said:

Not withstanding your analysis, it seems to me there are so many uncontrolled variables (e.g. focus, seeing, tracking etc.) it is hard to draw any firm conclusions.

Regards Andrew 

The conclusion for me is that camera technology is irrelevant and that spending al that money on an old ccd technology is just not worth it ... as you mentioned other factors are far more important for the end result, the camera is just not one of them, as even a cheap camera can obtain the same SNR (probably even better as the CMOS tests had way less total integration time) 

By the way the GSO RC test had identical focus and tracking, the big difference was moon phase, but those 2 results are pretty close to me ...

And for this test it was more about how much signal would there be to work with, not so much to get identical images as external factors will always have an impact. Should have made that more clear in the opening post. Anyway my decision to go for a Moravian 16200 will be put on ice for the moment, hoping for a mono cmos with the same dimensions.

Link to comment
Share on other sites

31 minutes ago, vdb said:

The conclusion for me is that camera technology is irrelevant and that spending al that money on an old ccd technology is just not worth it ... as you mentioned other factors are far more important for the end result, the camera is just not one of them, as even a cheap camera can obtain the same SNR (probably even better as the CMOS tests had way less total integration time) 

By the way the GSO RC test had identical focus and tracking, the big difference was moon phase, but those 2 results are pretty close to me ...

And for this test it was more about how much signal would there be to work with, not so much to get identical images as external factors will always have an impact. Should have made that more clear in the opening post. Anyway my decision to go for a Moravian 16200 will be put on ice for the moment, hoping for a mono cmos with the same dimensions.

Can't argue with that conclusion.  Measurements by C Bull, that I linked to in a different thread, led to the same conclusion.

I think the only issue with CMOS was increased amp glow compared to CCD. The addition of RAM to mitigate this did not help! 

Regards Andrew 

 

Link to comment
Share on other sites

Amp glow can be a problem if not dealt with correctly, and that can be sometimes a problem, and is very camera depended, the 183 chip behaves completely different than the 1600 and even using darks and turning off dark scaling can sometimes lead to incorrect results, but software developers are fine tuning their softwares as they get new test data from these new camera's ...

 

Link to comment
Share on other sites

Comparison of QHY and SX data, I took also some flat frames so I could make a better comparison. Even with the relative small chips the GSO RC 8 with reducer gives serious vignet. So at a lower cost you get more FOV and similar SNR ...

/Yves

 

 

28424294937_339b9f2dee_o.gif

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue. By using this site, you agree to our Terms of Use.