Am I expecting too much from CMOS

[centroid]

QUOTE:- Martin B

"Could the issue be that you are comparing OSC with mono Dave?"

Perhaps I should rephrase my question to read "Am I expecting too much from a  OSC CMOS astro camera" Martin.

There has never been any doubt that mono is more sensitive than OSC. Back in the day, I  moved over from OSC to Mono, because I wanted to image in NB, and LRGB. For this purpose I used a 7 position motorised, and usb controllable filter wheel, with L,R,G,B, Ha, Olll, and S11 filters installed. I am not that serious about astro imaging this time around, and see it very much as a sideline to my photography.

After reading user comments on various astro forums that CMOS was so much more sensitive than CCD, I thought I could get acceptable results with a CMOS OSC camera, but I have yet to see this. The image quality I  got with my SX CCD OSC asto cam (H9C), I have yet to see repeated with this 294c camera.

I chose the 294c because of the 4/3rds sensor size, and 4.63x4.63um  pixels, as opposed to the smaller sensor 183, with its very small 2.4x2.4um less sensitive pixel. My SXVR H16 had 7.4x7.4um pixels, and very good sensitivity, but of course that was mono.

I accept that I am now using a relatively budget imaging setup, comprising a £1200 APO Refractor, with a £900 294c based camera (I paid £2000 for my H16 back in 2011), but its not at the real budget end, and as such would expect better than I am getting with a supposedly more sensitive CMOS camera.

However, that is what I have, and I will persist with it, and maybe as I get to better understand its quirks, I will get better results. Time will tell.

I do understand that the IMX294 is old CMOS technology now, and am told that the later generation of CMOS cameras are much improved, and more ccd like, such as the 2600 I believe. However, I am not prepared to invest any more money in what what is very much a secondary interest now.

I really enjoyed the 12 years (2002-2014) that I spent astro imaging, before a medical intervention prompted me to give it up, and move house. This time around it has been one frustration after another, and as yet, little to show for it, and I have come close to walking away from it.

I have nor ordered a 2" Optolong Enhance filter, which will reduce the sensitivity even further. 😅

 

Edited January 31, 2022 by centroid

[vlaiv]

4 minutes ago, centroid said:

After reading user comments on various astro forums that CMOS was so much more sensitive than CCD

Well, let's compare the two with that regard.

Your old CCD:

Peak QE seems to be around 55%

Your new CMOS

this being relative QE graph and peak being estimated at 75%.

(75 - 55) / 55 = 36.4% peak performance improvement.

Ha line is about 30% absolute QE for CCD while it is at (0.9 * 0.75 = 0.675) 67.5% with CMOS

(67.5 - 30) / 30 = 125% Ha performance improvement in sensitivity

I'd say that above statement is correct - it is much more sensitive camera as far as QE goes (mono version probably more).

 

[Stuart1971]

50 minutes ago, centroid said:

QUOTE:- Martin B

"Could the issue be that you are comparing OSC with mono Dave?"

Perhaps I should rephrase my question to read "Am I expecting too much from a  OSC CMOS astro camera" Martin.

There has never been any doubt that mono is more sensitive than OSC. Back in the day, I  moved over from OSC to Mono, because I wanted to image in NB, and LRGB. For this purpose I used a 7 position motorised, and usb controllable filter wheel, with L,R,G,B, Ha, Olll, and S11 filters installed. I am not that serious about astro imaging this time around, and see it very much as a sideline to my photography.

After reading user comments on various astro forums that CMOS was so much more sensitive than CCD, I thought I could get acceptable results with a CMOS OSC camera, but I have yet to see this. The image quality I  got with my SX CCD OSC asto cam (H9C), I have yet to see repeated with this 294c camera.

I chose the 294c because of the 4/3rds sensor size, and 4.63x4.63um  pixels, as opposed to the smaller sensor 183, with its very small 2.4x2.4um less sensitive pixel. My SXVR H16 had 7.4x7.4um pixels, and very good sensitivity, but of course that was mono.

I accept that I am now using a relatively budget imaging setup, comprising a £1200 APO Refractor, with a £900 294c based camera (I paid £2000 for my H16 back in 2011), but its not at the real budget end, and as such would expect better than I am getting with a supposedly more sensitive CMOS camera.

However, that is what I have, and I will persist with it, and maybe as I get to better understand its quirks, I will get better results. Time will tell.

I do understand that the IMX294 is old CMOS technology now, and am told that the later generation of CMOS cameras are much improved, and more ccd like, such as the 2600 I believe. However, I am not prepared to invest any more money in what what is very much a secondary interest now.

I really enjoyed the 12 years (2002-2014) that I spent astro imaging, before a medical intervention prompted me to give it up, and move house. This time around it has been one frustration after another, and as yet, little to show for it, and I have come close to walking away from it.

I have nor ordered a 2" Optolong Enhance filter, which will reduce the sensitivity even further. 😅

 

I agree with you, and I think the issue is, all the superb images posted, taken with these cameras, gives you high expectations, and when it does not work out as expected it’s disappointing, I have the newer QHY268c with the IMX571 chip, and I too expected this to be better than my older SXVR M25c, but alas, is not the case, or at least no the case with me, 40mx 5 min subs with this CMOS camera, and the image leaves a hell of a lot to be desired….so my question, is where and how are all the fantastic images shown on here and other forums, produced, and what am I doing so wrong….

BTW, there is no calibration frames in this, just a RAW stack and straight out of APP, this was just to check what I had in the data…..not much was the answer….☹️

 

[vlaiv]

4 minutes ago, Stuart1971 said:

is not the case, or at least no the case with me, 40mx 5 min subs with this CMOS camera, and the image leaves a hell of a lot to be desired….

Would you be happy with 10 x 5 minute sub from your old camera if it produce image like above?

Using x2 smaller pixels - 3.75µm vs 7.8µm (actually by surface that is x4.32 so a bit more than x4) on same scope is like effectively shooting for 1/4 of the time.

If you want to match the performance of CCD camera (or surpass it) - then use the same resolution you did before - bin your pixels to match the size of old ones!

[Stuart1971]

2 minutes ago, vlaiv said:

Would you be happy with 10 x 5 minute sub from your old camera if it produce image like above?

Using x2 smaller pixels - 3.75µm vs 7.8µm (actually by surface that is x4.32 so a bit more than x4) on same scope is like effectively shooting for 1/4 of the time.

If you want to match the performance of CCD camera (or surpass it) - then use the same resolution you did before - bin your pixels to match the size of old ones!

Hmmmm, I may just try that….👍🏼 So 2x2 bin or 4x4…??

[vlaiv]

2 minutes ago, Stuart1971 said:

Hmmmm, I may just try that….👍🏼 So 2x2 bin or 4x4…??

That is a tough one for me to answer

Ideally you want to bin x2 - but with OSC consideration. Not sure what software does that, so you can do following which will be sort of equivalent: bin x4 your linear stack and then enlarge it by upsampling x2. That is closest operation to binning non debayered data in a special way and then debayering to get the color.

[Stuart1971]

I’m always confused by binning especially on CMOS, the pixels are there can can collect so much light, so by adding 4 together I get it collects four times the light but resolution is much smaller, but the pixels are still the same whether 4 individual ones collecting X amount of photons, or 4 together collecting the same amount….I don’t understand how it makes them more sensitive, how does it physically work, and can be so much better….

Not sure I have explained my question very well, apologies if not….🤔🤔

[Stuart1971]

10 minutes ago, vlaiv said:

Would you be happy with 10 x 5 minute sub from your old camera if it produce image like above?

Using x2 smaller pixels - 3.75µm vs 7.8µm (actually by surface that is x4.32 so a bit more than x4) on same scope is like effectively shooting for 1/4 of the time.

If you want to match the performance of CCD camera (or surpass it) - then use the same resolution you did before - bin your pixels to match the size of old ones!

I totally get what you say here about pixel size, but many people moved from CCD to CMOS which have always had smaller pixels, yet one of the main things I read from people was “you only need short exposures with these camera 1 or 2 mins, but lots of them” well I get nothing in a 2 min sub….and need to use 5 mins at least, 🤔🤔🤔🤔

so we’re all those people referring to mono CMOS cameras, I suspect many were, but not all, and when they say “You need a lot of subs” how many is a lot….100, 200, 500…..?? 
I see regular images with my make and model of camera that are superb, with say 2 hours of data….this is when I get disheartened….

But am wondering how much my Idas P2 LP filter is hindering me, so I have bought the Optolong L-Extreme, to try…..🤞🤞

 

[vlaiv]

2 minutes ago, Stuart1971 said:

I’m always confused by binning especially on CMOS, the pixels are there can can collect so much light, so by adding 4 together I get it collects four times the light but resolution is much smaller, but the pixels are still the same whether 4 individual ones collecting X amount of photons, or 4 together collecting the same amount….I don’t understand how it makes them more sensitive, how does it physically work, and can be so much better….

Not sure I have explained my question very well, apologies if not….🤔🤔

I do understand the question and here is rather simple answer:

If you take 4 images and stack them - it improves SNR, right? It does so regardless if you add or average them. Difference in adding and averaging is just division by for as adding them is a+b+c+d and averaging them is (a+b+c+d)/4. SNR does not change when you divide whole thing with a constant - so if we have S/N then it is the same as (S/4) / (N/4) - as those fours cancel each other.

Adding them is exactly the same thing as taking one exposure that is x4 as long (think pouring water into bucket - it does not matter if you pour in one bucket for 4 minutes or you take 4 buckets and pour for 1 minute and then add all that water up).

Same thing happens with adjacent pixels (they have to be adjacent so that signal is almost the same). Binning is just adding pixel values up - and adding 4 pixel values is like either integrating for x4 longer with same pixel size - or having pixel that is x4 larger and integrating the same. In either case - SNR for that pixel goes up same way as with stacking, and SNR per unit time really represents "sensitivity".

Here is another way to look at it - more signal - better SNR.

Adding 4 pixel values puts all that signal into "single bucket" - more signal - better SNR (or simply - pixels are more sensitive as they gathered more signal in same amount of time).

[Stuart1971]

1 minute ago, vlaiv said:

I do understand the question and here is rather simple answer:

If you take 4 images and stack them - it improves SNR, right? It does so regardless if you add or average them. Difference in adding and averaging is just division by for as adding them is a+b+c+d and averaging them is (a+b+c+d)/4. SNR does not change when you divide whole thing with a constant - so if we have S/N then it is the same as (S/4) / (N/4) - as those fours cancel each other.

Adding them is exactly the same thing as taking one exposure that is x4 as long (think pouring water into bucket - it does not matter if you pour in one bucket for 4 minutes or you take 4 buckets and pour for 1 minute and then add all that water up).

Same thing happens with adjacent pixels (they have to be adjacent so that signal is almost the same). Binning is just adding pixel values up - and adding 4 pixel values is like either integrating for x4 longer with same pixel size - or having pixel that is x4 larger and integrating the same. In either case - SNR for that pixel goes up same way as with stacking, and SNR per unit time really represents "sensitivity".

Here is another way to look at it - more signal - better SNR.

Adding 4 pixel values puts all that signal into "single bucket" - more signal - better SNR (or simply - pixels are more sensitive as they gathered more signal in same amount of time).

Yes you did understand me, and I get it, thanks….👍🏼
DO CMOS camera BIN well…? As I know it’s different to binning CCD…?

[vlaiv]

1 minute ago, Stuart1971 said:

I totally get what you say here about pixel size, but many people moved from CCD to CMOS which have always had smaller pixels, yet one of the main things I read from people was “you only need short exposures with these camera 1 or 2 mins, but lots of them” well I get nothing in a 2 min sub….and need to use 5 mins at least, 🤔🤔🤔

Ok, so here is the thing with shorter exposures and all of that.

Ideal sub length depends on ratio of read noise to some other noise source in the system. Usually that is LP noise, since we have cooled cameras and thermal noise is very small. Targets are faint and target shot noise is also small. LP is by far biggest source of noise with most setups.

For a given setup - difference between CCD and CMOS is that read noise - and that is smaller by factor of about x4-x5 between the two. CCDs were usually around 8e mark for read noise (some more, some less - like Sony sensors that could go down to 4-5e of read noise), while CMOS sensors often have 1.5-2e of read noise.

This really means that CMOS sensors can utilize x4-x5 shorter exposures than CCD (all other things being equal). People used to image with 5-10minute exposures with CCDs - so direct comparison would be 1-2 minutes. That part is right.

However - not all things are equal, and setups are different.

- If you image in darker skies - you'll be better of with longer exposures since your LP levels are lower and LP noise is lower

- If you image at higher resolution - you'll be better of with longer exposures - since that LP is spread over more pixels and each pixel gets less signal so associated noise is also lower

This holds for both CCD and CMOS equally and has nothing to do with fact that CCDs have x4-x5 times higher read noise than CMOS.

If you need to image for 5 minutes with your setup and CMOS (with say 1.5e of read noise) - then you'd need to image for 25 minutes per sub for CCD given same circumstances (resolution and LP levels).

 

[vlaiv]

4 minutes ago, Stuart1971 said:

Yes you did understand me, and I get it, thanks….👍🏼
DO CMOS camera BIN well…? As I know it’s different to binning CCD…?

They bin as well as CCD.

Difference is that with CCDs it happens before readout and with CMOS it happens after readout in software.

Only difference between those two is when read noise happens and how much of it there will be (which you should consider in the light of above argument about sub duration).

With CCDs - when you bin - you still get the same read noise - one "dose" of it per binned pixel.

With CMOS - when you bin - each pixel already received one "dose" of read noise before binning and when you bin them that noise adds up and that is like having bigger pixel but also higher read noise camera.

There is simple "law" to calculate resulting read noise.

Say if you have CCD with 8e of read noise and 7µm pixel size and you bin that camera x2 - it will be like having 14µm pixel size camera with 8e of read noise

If you have CMOS with say 2e of read noise and 4µm pixel size and you bin that x2 - it will be like having 8µm pixel size but with 4e of read noise this time.

With CMOS sensors - equivalent read noise increases by bin factor - so if you bin 2x2 - you'll get double read noise, and if you bin 3x3 - you'll get triple the read noise.

What does that mean in terms of sub duration - well nothing. If you set sub duration given read noise on unbinned data - then it will be sufficient for binned data as well, so you don't have to change anything and you can just bin your data in software.

[centroid]

2 hours ago, vlaiv said:

Well, let's compare the two with that regard.

Your old CCD:

Peak QE seems to be around 55%

Your new CMOS

this being relative QE graph and peak being estimated at 75%.

(75 - 55) / 55 = 36.4% peak performance improvement.

Ha line is about 30% absolute QE for CCD while it is at (0.9 * 0.75 = 0.675) 67.5% with CMOS

(67.5 - 30) / 30 = 125% Ha performance improvement in sensitivity

I'd say that above statement is correct - it is much more sensitive camera as far as QE goes (mono version probably more).

 

Yep, that is the theoretical result,/prediction,  probably taken under Lab conditions, but to quote an old English saying, "the proof of the pudding is in the eating", and theory doesn't always play out in practice.

I had a 40 year technical career, the last 20 of which were in R&D  (Radio Frequency Engineering, and Electromagnetic Compatibility), and If I had a £ for very time practice didn't align with theory in the real world, I would be a very rich  man.

The only theory that was never wrong, was good old Ohm's Law. 🙂

BTW Vlaiv, your command of the English language never ceases to amaze me 👏.

I can get by with some German and French when away on holiday, but fluent, absolutely not.!!

[vlaiv]

4 minutes ago, centroid said:

BTW Vlaiv, your command of the English language never ceases to amaze me

Thank you.

4 minutes ago, centroid said:

Yep, that is the theoretical result,/prediction,  probably taken under Lab conditions, but to quote an old English saying, "the proof of the pudding is in the eating", and theory doesn't always play out in practice.

I had a 40 year technical career, the last 20 of which were in R&D  (Radio Frequency Engineering, and Electromagnetic Compatibility), and If I had a £ for very time practice didn't align with theory in the real world, I would be a very rich  man.

I have slightly different view on this: If in practice you are not seeing results predicted by theory, then:

- you are applying theory in the wrong way

- you are applying theory outside its domain of validity

- the least likely scenario - theory is wrong and should be corrected or replaced (how often does that happen?)

[centroid]

43 minutes ago, vlaiv said:

Thank you.

I have slightly different view on this: If in practice you are not seeing results predicted by theory, then:

- you are applying theory in the wrong way

- you are applying theory outside its domain of validity

- the least likely scenario - theory is wrong and should be corrected or replaced (how often does that happen?)

Theory is devised under controlled lab conditions,  and can only estimate the multitude of variables that could impact on it under "Real World" in use conditions.

Where laboratory testing is involved, unless the test equipment used is calibrated to traceable national standards (UKAS in the UK), the results carry very little standing. Then of course there is Measurement Uncertainty, which usually follows a mathematical distribution, The uncertainties can either be systematic or operator induced, but they are real, and must be taken into account . 

Hey, I been retired for some 19 years, and didn't  think I would be discussing theory v practice, and measurement uncertainties, revolving around an amateur level Chinese CMOS based astro camera, for what is a hobby pass time 🙄

We have probably bored other members to tears. ☹️

If I find out why this 294 camera is giving me poor results, I will let you all know.

Edited January 31, 2022 by centroid
typo

[MartinB]

Does that mean that in the real world a Reliant Robin might out accelerate a Porche 911? 🙄

[centroid]

2 hours ago, MartinB said:

Does that mean that in the real world a Reliant Robin might out accelerate a Porche 911? 🙄

If that is what some folk are foolish enough to believe,  then that is up to them.

If we going from logical discussion to frivolity, then  I'll leave you to it, but in terms of ridulous  exaggeration, I would say that there is a "big difference between scratching your backside, and tearing the skin off". 

For a man whose intelligence, and knowledge, I have much respect for Martin, you surprise, me.

[ONIKKINEN]

9 hours ago, Stuart1971 said:

I totally get what you say here about pixel size, but many people moved from CCD to CMOS which have always had smaller pixels, yet one of the main things I read from people was “you only need short exposures with these camera 1 or 2 mins, but lots of them” well I get nothing in a 2 min sub….and need to use 5 mins at least, 🤔🤔🤔🤔

so we’re all those people referring to mono CMOS cameras, I suspect many were, but not all, and when they say “You need a lot of subs” how many is a lot….100, 200, 500…..?? 
I see regular images with my make and model of camera that are superb, with say 2 hours of data….this is when I get disheartened….

Thought i would mention that its not necessary for the single sub to look good, so you might be chasing something that doesn't need to be. Of course the data must be there somewhere and you have to have some sort of statistics to tell you the sub was decent (basic stuff, FWHM, number of stars, etc), but it doesn't have to be visible in a single exposure. Not a bad thing if the single sub looks good of course.

Below are 2 example pictures, one single calibrated and stretched/colourbalanced 30s sub and an image consisting of several hours of subs exactly like this one or slightly worse (picked one at random from hundreds). I am not sure how many hours is in the stacked/processed shot but i think it is either 5 or 6 hours of only 30s exposures, its a work in progress for at least triple this time to wrestle the IFN cleanly out of the background...

There is not even a hint of many of the smaller galaxies in the single sub, even the biggest ones are just a couple of pixels above noise. Never mind the IFN which only showed up in the negative version of a 2h stack and i could never guess that it existed in the single exposure. I am actually amazed myself that stacking somehow makes these structures come out from what looks like a mess of nothing at all, but i trust the technology in this case.

Point is: Its not important for the single sub to look good if the data is there. Photons go in the camera, picture comes out in the end.

[MartinB]

39 minutes ago, centroid said:

If that is what some folk are foolish enough to believe,  then that is up to them.

If we going from logical discussion to frivolity, then  I'll leave you to it, but in terms of ridulous  exaggeration, I would say that there is a "big difference between scratching your backside, and tearing the skin off". 

For a man whose intelligence, and knowledge, I have much respect for Martin, you surprise, me.

Sorry Dave, only a bit of flippancy😑

[centroid]

16 minutes ago, MartinB said:

Sorry Dave, only a bit of flippancy😑

Water under the bridge now Martin, but it came across as a bit of  P*** Take, and that didn't go down well at all.

My apologies if I misread it 🙂.

Onwards and upwards 👍

Edited January 31, 2022 by centroid
Typo

[tomato]

+1 on the expectations of individual subs. 2 minute exposures on the 268c are very uninspiring, especially when I compare them to a 5 min mono sub from the KAF 8300 CCD. It’s the stacked image that counts.