12 bit vs 16 bit

[Thommy]

I'm planning an upgrade from my current mono CCD. Apparently a lot of people go the CMOS way these days, so I consider whether I should follow the trend or stick with the CCD. I'm primarily interested in imaging faint far-away galaxies. I noted the new CMOS cameras from SX, but one thing worries me: They are 12 bit whereas CCD cameras are 16 bit. I know that you gain contrast when processing the images, so does it matter?

 

[alexbb]

The significant difference comes from the low read noise and the recommended typical usage. With a CMOS sensor camera you'd probably want to take more shorter exposures.

The mono Panasonic sensors found in the ASI1600/Atik Horizon/Qhy163 has a full well of ~20000 electrons which is converted by the ADU to a value represented on 12 bits (0-4095). Using a gain for the highest dynamic range, the ADU will indeed limit the resolution of read data. But for the highest dynamic range, the read noise is highest and this is not the best way to use such a camera.

As I said in the beginning, with these CMOS sensors, for optimal results you should increase the gain - in order to benefit from the lower read noise. This also decreases the dynamic range.

For an ASI1600, above a certain gain the dynamic range decreases enough so that the recorded data can be represented on a 12bit integer value so the ADC is no longer the limiting factor. What you need to do is to record as many exposures as you can.

Be aware that compared to the typical usage of a camera with a CCD sensor with long exposures, a sub taken with a CMOS sensor will be much darker and a stack of subs as well. In order not to lose relevant data, make sure that the stack is represented on at least 32 bits. Some applications (ie. Registar) doesn't work with 32bit float fits files, but most of them do (PixInsight, APP, GIMP 2.10). If you want a comparison, is like having a cup filled with water (stack of long exposure CCD subs) or the same amount of water in a bucket (same amount of hours stack of short exposure CMOS subs) - the water will have different heights.

[Adam J]

3 hours ago, Thommy said:

I'm planning an upgrade from my current mono CCD. Apparently a lot of people go the CMOS way these days, so I consider whether I should follow the trend or stick with the CCD. I'm primarily interested in imaging faint far-away galaxies. I noted the new CMOS cameras from SX, but one thing worries me: They are 12 bit whereas CCD cameras are 16 bit. I know that you gain contrast when processing the images, so does it matter?

 

No it won't make a difference so long as you use it like a CMOS and don't try to use it like your CCD. 

[Thommy]

Thanks a lot for the detailed explanation.

I can see that I need to study the CMOS technology to understand this better.

If I understand you correctly, getting the most out of a CMOS based camera nails down to finding the optimal gain, where read noise is low. This also reduces dynamic range, so using short exposures (not to fill the well) but take lots of subs would be the way forward. This would result in a similar S/N and dynamic range as a CCD with similar total integration time but using longer subs. Is that about right?

So the advantage of the CMOS would be lower price of course, but also that the shorter subs are more forgiving to accuracy of mounts but the downside would be generating more GB data. So there's no particular advantage in choosing a CCD when only imaging faint stuff and not like M31 or M42? 

[Alien 13]

20 minutes ago, Thommy said:

Thanks a lot for the detailed explanation.

I can see that I need to study the CMOS technology to understand this better.

If I understand you correctly, getting the most out of a CMOS based camera nails down to finding the optimal gain, where read noise is low. This also reduces dynamic range, so using short exposures (not to fill the well) but take lots of subs would be the way forward. This would result in a similar S/N and dynamic range as a CCD with similar total integration time but using longer subs. Is that about right?

So the advantage of the CMOS would be lower price of course, but also that the shorter subs are more forgiving to accuracy of mounts but the downside would be generating more GB data. So there's no particular advantage in choosing a CCD when only imaging faint stuff and not like M31 or M42? 

That seems about right, CMOS cameras might also have smaller pixels which are good for shorter focal lengths and the ability to shoot high frame rates so can be also used for Lunar and Planetary imaging too.

Alan

[JamesF]

50 minutes ago, Thommy said:

Thanks a lot for the detailed explanation.

I can see that I need to study the CMOS technology to understand this better.

If I understand you correctly, getting the most out of a CMOS based camera nails down to finding the optimal gain, where read noise is low. This also reduces dynamic range, so using short exposures (not to fill the well) but take lots of subs would be the way forward. This would result in a similar S/N and dynamic range as a CCD with similar total integration time but using longer subs. Is that about right?

So the advantage of the CMOS would be lower price of course, but also that the shorter subs are more forgiving to accuracy of mounts but the downside would be generating more GB data. So there's no particular advantage in choosing a CCD when only imaging faint stuff and not like M31 or M42? 

This video may be worth a watch, too:

 

One question it leaves me with is this:

If your sky is sufficiently dark that you can still appreciably increase the image SNR by using longer exposures, but those exposures would result in clipping of the data due to the more limited ADC resolution of a CMOS camera, is there a point at which the CCD with its larger dynamic range becomes a more compelling choice?

James

[Thommy]

That's a good question, James! Perhaps we will have some opinions on this. Glover does a good presentation - just a shame that he had to skip the gain part. Nonetheless a difficult choice. With my suburban sky, I would probably be better off with a cmos, but I would have to add a new laptop to cope with the extensive number of subs. The low read noise of the new SX trius pro series may be a good compromise 😏

[JamesF]

5 minutes ago, Thommy said:

That's a good question, James! Perhaps we will have some opinions on this. Glover does a good presentation - just a shame that he had to skip the gain part. Nonetheless a difficult choice. With my suburban sky, I would probably be better off with a cmos, but I would have to add a new laptop to cope with the extensive number of subs. The low read noise of the new SX trius pro series may be a good compromise 😏

The remainder of his presentation is on his youtube channel.  I'll see if I can find the link again.

James

[JamesF]

This should be it:

Can't check at the moment, as the children have just got back from school and eaten all my bandwidth

James

[MarsG76]

Very interesting presentation.... I think that the mathematics is all good and a good guide, but in the real world there are too many factors at play and sop there is still a big trial and error component involved.

 

[Thommy]

16 hours ago, JamesF said:

Can't check at the moment, as the children have just got back from school and eaten all my bandwidth

James

Yeah, children can be like that! For the last few years I had to reduce my imaging to a minimum due to those beasts.

15 hours ago, MarsG76 said:

Very interesting presentation.... I think that the mathematics is all good and a good guide, but in the real world there are too many factors at play and sop there is still a big trial and error component involved.

Indeed, a very interesting presentation. I think the right way to look at it is to use the theory as a starting point for all the tweaking. The span of trial and error component is very reduced when you have an idea where to start.

[MarsG76]

6 hours ago, Thommy said:

Indeed, a very interesting presentation. I think the right way to look at it is to use the theory as a starting point for all the tweaking. The span of trial and error component is very reduced when you have an idea where to start.

Yeah also true

[cabfl]

In general, the dynamic range is confused with the ADC.
CCD cameras do not have 16bits of dynamic range.
They usually have a range between 11 and 13 bits, but also a lot of noise and that's why they need a 16bit ADC.

11 bits = 66db
12 bits = 72db 
13 bits = 78db

Sony ICX694 - CCD Atik460EX / QSI660 = 75db = 12,5 bits = 12.5 DRStops.
http://www.qsimaging.com/660-overview.html
Kaf-8300 = 70db = 11,66 bits  = 11.66 DRStops..
https://www.onsemi.com/pub/Collateral/KAF-8300-D.PDF
Kaf-16200 = 69db = 11,5 bits  = 11.5 DRStops..
https://www.onsemi.com/pub/Collateral/KAF-16200-D.PDF
KAF-1603 = 74db = 12,4 bits  = 12.66 DRStops..
https://www.onsemi.com/pub/Collateral/KAF-1603-D.PDF

CCD cameras have a lot of noise, and to improve the signal / noise they need very long shots.
CMOS cameras have a very low noise, and do not need long shots to get good signal / noise.
As you make short shots, then you need to stack more shots to add more signal.
But if it's a very weak object, you can make long shots with the CMOS, but watch the fullwell to avoid fat stars.

Edited May 30, 2019 by cabfl

[newbie alert]

On 30/05/2019 at 23:22, cabfl said:

CCD cameras have a lot of noise, and to improve the signal / noise they need very long shots.
CMOS cameras have a very low noise, and do not need long shots to get good signal / noise.
As you make short shots, then you need to stack more shots to add more signal.
But if it's a very weak object, you can make long shots with the CMOS, but watch the fullwell to avoid fat stars.

Hang on a minute... you're saying ccd's have alot of noise and cmos don't?

Why do cmos require darkframe calibration  and ccd don't.. they both are cooled..

[Adam J]

13 hours ago, newbie alert said:

Hang on a minute... you're saying ccd's have alot of noise and cmos don't?

Why do cmos require darkframe calibration  and ccd don't.. they both are cooled..

You can get away without dark frames for some of the most modern Sony CCD's but people often still choose to use them. Some of the older sensors like the KAF8300 do very much require Dark frames.

The thing to remember about dark frames is that they dont reduce random noise (actually they can increase it...) they are used to take out bad pixels and fixed pattern noise such as banding and amp glow.

As such the need to remove such artifacts and the amount of random noise are not always related.

Adam

[cabfl]

16 hours ago, newbie alert said:

Hang on a minute... you're saying ccd's have alot of noise and cmos don't?

Why do cmos require darkframe calibration  and ccd don't.. they both are cooled..

Dark noise or Read noise?

Read Noise:

Atik 490EX - 5e
Atik 460EX - 5e
Atik 383L+ - 7e
Atik 16200 - 10e
Atik 11000 - 11e

ASI 183 - 1e
ASI 1600 - 1.2e
ASI 294 - 1.2e
QHY 247 - 1e
ASI 071 - 2.3e
ASI 094 - 2.1e
ASI 128 - 2.5e

 

The new CMOS sensors:

QHY268C - 1e (ADC 16 bits)

QHY600M/C / ASI6200Pro - 1e to 2.3e (ADC 16 bits - 90db = 14,5 bits)

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=94&id=56&cut=1

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=94&id=55&cut=1

CCD Sensor:

Kaf-16200 = 10e (ADC 16 bits - 69db = 11,5 bits)

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=95&id=17&cut=1

 

KAF8300 vs ASI1600:

https://www.cloudynights.com/topic/549417-asi1600-experiments-with-high-gain-nb-imaging/

Edited June 9, 2019 by cabfl

[newbie alert]

7 hours ago, cabfl said:

Dark noise or Read noise?

Read Noise:

Atik 490EX - 5e
Atik 460EX - 5e
Atik 383L+ - 7e
Atik 16200 - 10e
Atik 11000 - 11e

ASI 183 - 1e
ASI 1600 - 1.2e
ASI 294 - 1.2e
QHY 247 - 1e
ASI 071 - 2.3e
ASI 094 - 2.1e
ASI 128 - 2.5e

 

The new CMOS sensors:

QHY268C - 1e (ADC 16 bits)

QHY600M/C / ASI6200Pro - 1e to 2.3e (ADC 16 bits - 90db = 14,5 bits)

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=94&id=56&cut=1

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=94&id=55&cut=1

CCD Sensor:

Kaf-16200 = 10e (ADC 16 bits - 69db = 11,5 bits)

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=95&id=17&cut=1

 

KAF8300 vs ASI1600:

https://www.cloudynights.com/topic/549417-asi1600-experiments-with-high-gain-nb-imaging/

Hang on..thought the comparrison was with a 460 with a Sony chip not a Panasonic? The kaf 8300 chip will do better with a longer sub, the cmos will do better with shorter subs..

Problem is with a variable gain with a cmos you will get a variable electron reading.. cmos will have a lower stat as it works in a different way...do a 15min exposure with a cmos and compare...

Let's compare apples with apples..

.. Everyone I know that uses the 1600 in whatever format uses darkframe calibration....read noise is bias?

 

[newbie alert]

10 hours ago, Adam J said:

On 08/06/2019 at 09:21, newbie alert said:

Why do cmos require darkframe calibration  and ccd don't.. they both are cooled..

You can get away without dark frames for some of the most modern Sony CCD's but people often still choose to use them. Some of the older sensors like the KAF8300 do very much require Dark frames.

Yes,with the older Panasonic chip, the newer Sony chip don't...agree that adding darkframe subtraction can add noise..

You don't get ampglow with ccd, with cmos you do

[Whirlwind]

34 minutes ago, newbie alert said:

Hang on..thought the comparrison was with a 460 with a Sony chip not a Panasonic? The kaf 8300 chip will do better with a longer sub, the cmos will do better with shorter subs..

Problem is with a variable gain with a cmos you will get a variable electron reading.. cmos will have a lower stat as it works in a different way...do a 15min exposure with a cmos and compare...

Let's compare apples with apples..

.. Everyone I know that uses the 1600 in whatever format uses darkframe calibration....read noise is bias?

 

Yeah this test is probably not ideal in the way it is being shown.  The KAF8300 CCD has higher read noise and the image appears to be narrowband.  To truly get the best out of the CCD you need exposures so that the other sources of noise will overwhelm the read noise.  With a low read noise in the CMOS this doesn't take very long and 60s will do this.  In CCDs for narrowband you need much longer exposures of 20-30 minutes to overwhelm the read noise.  A fairer test to demonstrate achievable results would be to compare to 4 x 1200s exposures vs 80 x 60s.  If you compared a 20minute CMOS exposure to the a 20minute CCD exposure the results would probably favour the CCD.

What this really demonstrates is that the CMOS allows us to image with less accurate mounts as any lost time because of guiding/worm irregularities can easily be discarded.  In this case the CCDs need better mounts so that they can expose for longer (especially in narrowband).

They both have their pros and cons.  CMOS also tends to require much more aggressive dithering from what I have read to combat fixed pattern noise.  In addition around bright stars the micro-lens can give strange diffraction patterns.  But they can both give excellent results if used in a way that utilises their strengths and mitigates their weaknesses.

Edited June 9, 2019 by Whirlwind

[newbie alert]

38 minutes ago, Whirlwind said:

Yeah this test is probably not ideal in the way it is being shown.  The KAF8300 CCD has higher read noise and the image appears to be narrowband.  To truly get the best out of the CCD you need exposures so that the other sources of noise will overwhelm the read noise.  With a low read noise in the CMOS this doesn't take very long and 60s will do this.  In CCDs for narrowband you need much longer exposures of 20-30 minutes to overwhelm the read noise.  A fairer test to demonstrate achievable results would be to compare to 4 x 1200s exposures vs 80 x 60s.  If you compared a 20minute CMOS exposure to the a 20minute CCD exposure the results would probably favour the CCD.

What this really demonstrates is that the CMOS allows us to image with less accurate mounts as any lost time because of guiding/worm irregularities can easily be discarded.  In this case the CCDs need better mounts so that they can expose for longer (especially in narrowband).

They both have their pros and cons.  CMOS also tends to require much more aggressive dithering from what I have read to combat fixed pattern noise.  In addition around bright stars the micro-lens can give strange diffraction patterns.  But they can both give excellent results if used in a way that utilises their strengths and mitigates their weaknesses.

Yes it seems like the cmos brigade like you use the  8300 chip as the basis of the whole ccd verses cmos thing..(nothing wrong with a Panasonic 8300 chip, it's a classic) Cmos has it's strengths and weakness,and the same for ccd's ...

Once the cmos sensors camera's have overcome the ampglow,  Micro lensing, storage space, what gain and offset to use problems then I'm sure I take another look, but for the time being I stick to ccd, purely for the simplicity of use.. more or less plug and play

[Adam J]

39 minutes ago, newbie alert said:

Yes it seems like the cmos brigade like you use the  8300 chip as the basis of the whole ccd verses cmos thing..(nothing wrong with a Panasonic 8300 chip, it's a classic) Cmos has it's strengths and weakness,and the same for ccd's ...

Once the cmos sensors camera's have overcome the ampglow,  Micro lensing, storage space, what gain and offset to use problems then I'm sure I take another look, but for the time being I stick to ccd, purely for the simplicity of use.. more or less plug and play

The only reason the KAF8300 (Kodak not Panasonic) is compared to the ASI1600mm pro is that they are similar size the Sony chip is half the size of either one of them. Size of chip does matter allot in imaging half the sensor size twice the integration to take the same FOV image on the same scope.

[newbie alert]

1 hour ago, Adam J said:

The only reason the KAF8300 (Kodak not Panasonic) is compared to the ASI1600mm pro is that they are similar size the Sony chip is half the size of either one of them. Size of chip does matter allot in imaging half the sensor size twice the integration to take the same FOV image on the same scope.

Kodak!! Sorry was thinking of the 183!

 The 1600 has a 4/3 sensor... yes has a wide fov but a 460 Sony chip frames up m45 nicely

Edited June 9, 2019 by newbie alert
Added info