CMOS image sensors

[Gan]

I wonder if some one could explain the difference in the image quality between CMOS sensors of e.g. Canon DSLR and a dedicated Astroimaging CMOS camera e.g ZWO. If we are able to match the size of the pixel and say size of the sensor  ( APS-C), why should there be significant benefit from an Astro-camera. I would be grateful if some one could explain to me. thanks guys

Gan

 

[vlaiv]

Astro cameras use same sensors as DSLR cameras - either CMOS or CCD (of course actual sensors will differ a bit depending on camera model, but are in principle the same).

It is other features that distinguish astro cameras and DSLR-s.

Lack of filters - DSLR has IR/UV cut filter that needs to be removed/replaced to get the most out of DSLR (so called astro modding of DSLR).

Astro cameras also don't have anti aliasing filter on them - some DSLRs do.

Most significant feature is set point cooling (not all astro cameras have that) which enables precise calibration to be carried out on your data. Cooling as such is not as important for calibration - it does help with thermal noise, but ability to always have sensor at certain temperature is the key for good calibration, so main difference is that.

[JamesF]

It might not also be unreasonable to suggest that DSLRs tend to be designed for situations where there is an abundance of light with relatively short exposure times whereas astro cameras are often going to be working in situations where the absolute signal level is very low and exposure times can be relatively long.  Each can in turn influence the design and layout of the electronics to suit the priorities of the manufacturer and target market.  For example, whilst some Canon cameras have relatively low levels of noise, others are significantly worse.  In most situations that's not an issue because the noise is swamped by the signal from amount of light available.  For astro-imaging, it's not so great.

James

[Physopto]

Yes my Canon 5D is awful for Astro imaging due to thermal noise. The newer versions are considerably better so I am told.

I now use  a QSI 683wsg . Very low noise in comparison and the set point allows easy noise elimination to a great extent.

Derek 

[Gan]

On 05/12/2019 at 19:17, vlaiv said:

Astro cameras use same sensors as DSLR cameras - either CMOS or CCD (of course actual sensors will differ a bit depending on camera model, but are in principle the same).

It is other features that distinguish astro cameras and DSLR-s.

Lack of filters - DSLR has IR/UV cut filter that needs to be removed/replaced to get the most out of DSLR (so called astro modding of DSLR).

Astro cameras also don't have anti aliasing filter on them - some DSLRs do.

Most significant feature is set point cooling (not all astro cameras have that) which enables precise calibration to be carried out on your data. Cooling as such is not as important for calibration - it does help with thermal noise, but ability to always have sensor at certain temperature is the key for good calibration, so main difference is that.

That's very helpful note. I can understand the thermal noise and the positive effect of cooling. In that case a mod DSLR should be at least as good as an astro CMOS camera that is not cooled.  My DSLR is giving me a resolution of 0.38 arc sec / pixel ( 1x 1 binning) on a 8inch SCT, can I expect a significantly better performance with an astrocamera ( CMOS) like ZWO ASI294 without cooling? Kindly advise 

[vlaiv]

10 hours ago, Gan said:

That's very helpful note. I can understand the thermal noise and the positive effect of cooling. In that case a mod DSLR should be at least as good as an astro CMOS camera that is not cooled.  My DSLR is giving me a resolution of 0.38 arc sec / pixel ( 1x 1 binning) on a 8inch SCT, can I expect a significantly better performance with an astrocamera ( CMOS) like ZWO ASI294 without cooling? Kindly advise 

That really depends on comparison of two sensors and their characteristics.

What DSLR do you currently have?

Specifications that you want to look at are:

- read noise in both sensors (lower is better)

- dark current levels (again lower is better)

- amp glow (absence is better - this one is particularly nasty if you don't have set point cooling, there is sort of a way to deal with it but it might or might not work for particular sensor - it is called dark frame optimization)

- QE of each sensor (higher is better)

In the end you have to see which one is going to be better matched in terms of resolution once you bin (you will need to bin on 8" SCT since it has quite a bit of focal length). You would ideally want to target sampling rate in range of 1.2-1.5"/px.

With current DSLR and resolution of 0.38"/px that will be - super pixel mode (OSC sensors have twice lower sampling rate than mono sensors due to bayer matrix) + x2 binning, which will give you 1.52"/px.

With ASI294 you will have 0.48"/px, so after super pixel mode that will be 0.96"/px - that is ok only if you have rather good guiding and good skies (like 0.5" RMS guiding and good seeing) and your optics are sharp (EdgeHD). If you go for ASI294 you will want to use reducer for SCT.

My personal opinion is that ASI294 without cooling would not justify upgrade unless you are planning to use it for EEVA or similar along side imaging. If you want true upgrade then consider extending budget to cooled version.

There are few additional benefits of astro camera vs DSLR - weight being one, external powering (USB connection in case of non cooled model) rather than internal battery (may help with thermal issues. Drawback is of course need for laptop to use astro camera along with cabling (at least USB lead in case of non cooled model, and power cord with cooled one).

[Gan]

5 hours ago, vlaiv said:

That really depends on comparison of two sensors and their characteristics.

What DSLR do you currently have?

Specifications that you want to look at are:

- read noise in both sensors (lower is better)

- dark current levels (again lower is better)

- amp glow (absence is better - this one is particularly nasty if you don't have set point cooling, there is sort of a way to deal with it but it might or might not work for particular sensor - it is called dark frame optimization)

- QE of each sensor (higher is better)

In the end you have to see which one is going to be better matched in terms of resolution once you bin (you will need to bin on 8" SCT since it has quite a bit of focal length). You would ideally want to target sampling rate in range of 1.2-1.5"/px.

With current DSLR and resolution of 0.38"/px that will be - super pixel mode (OSC sensors have twice lower sampling rate than mono sensors due to bayer matrix) + x2 binning, which will give you 1.52"/px.

With ASI294 you will have 0.48"/px, so after super pixel mode that will be 0.96"/px - that is ok only if you have rather good guiding and good skies (like 0.5" RMS guiding and good seeing) and your optics are sharp (EdgeHD). If you go for ASI294 you will want to use reducer for SCT.

My personal opinion is that ASI294 without cooling would not justify upgrade unless you are planning to use it for EEVA or similar along side imaging. If you want true upgrade then consider extending budget to cooled version.

There are few additional benefits of astro camera vs DSLR - weight being one, external powering (USB connection in case of non cooled model) rather than internal battery (may help with thermal issues. Drawback is of course need for laptop to use astro camera along with cabling (at least USB lead in case of non cooled model, and power cord with cooled one).

Again very informative and helpful discussion. Thank you.

Sorry I should have mentioned full information: my DSLR is Canon 80 D.

It seems very clear now that a dedicated astro-camera preferably with cooling is the way forward. Obviously this is a bit expensive venture. In the meantime , what can I do to get the most out of my DSLR?

How do I get sampling rates of 1.2" to 1.5" using my Canon 80D?

I didn't quite understand this super-pixel. Please could you explain and how to use it? 

 

[vlaiv]

4 hours ago, Gan said:

Again very informative and helpful discussion. Thank you.

Sorry I should have mentioned full information: my DSLR is Canon 80 D.

It seems very clear now that a dedicated astro-camera preferably with cooling is the way forward. Obviously this is a bit expensive venture. In the meantime , what can I do to get the most out of my DSLR?

How do I get sampling rates of 1.2" to 1.5" using my Canon 80D?

I didn't quite understand this super-pixel. Please could you explain and how to use it? 

 

Unfortunately I can't seem to find read noise value for 80D expressed in electrons at different ISO settings (another advantage of astro cameras - you get those specs, or you can measure them), but regardless of that I would use longer exposures since you are vastly oversampling with 8" SCT. There is a hint of it being "iso-less" online so we can assume that read noise is pretty low. It also means that you can use something like ISO200-ISO400 range to give you more full well capacity.

So first recommendation is go as long as you can in exposure length - at least couple of minutes.

Second recommendation would be to try proper calibration regardless of the fact that you don't have set point cooling. For your next session  - gather all calibration subs to try it out (you can still skip some steps and do different calibration by just omitting certain files from your work flow).

- Darks at temperature close to those you worked with during the night - maybe take 10-20 darks before you start your lights and then another 10-20 darks after you finish, or maybe do it on a cloudy night when temperature is close to that you shot your light subs at. Try to get as much dark subs as possible (at least 20-30, but if you can - make it more)

- Do set of bias subs - again gather as much as you can

- Do set of flats (you need to do it on night of imaging if you don't have obsy and need to disassemble your rig at the end) and

- do a set of matching flat darks

Don't know what software are you using, but do regular average for bias, flats and flat darks and use sigma reject stacking for darks. Also use Dark optimization (there is a checkbox in DSS if you are using that for stacking). In case you find any artifacts like vertical / horizontal streaks or similar in the background in your final image - that means that dark optimization failed for your sensor - then try what most people do with DSLRs - using bias instead of darks (and flat darks).

Next thing to do is use superpixel mode when debayering. Again that is not the best way to do things, but best way would be very complicated in terms of software support, so we will settle for second best.

Super pixel mode just means that R, G and B channel images are made in such way that 4 adjacent pixels in bayer matrix result in a single pixels in each channel. It just uses one R pixel from bayer 2x2 block for R channel image, one B pixel for B channel image and it averages 2 green pixels for G channel image.

Resulting R, G and B images will have twice lower resolution than your sensor, and in this case it will be 3144 x 2028 instead of 6288 x 4056. It also means that these R, G and B images are no longer sampled at 0.38"/px but at 0.76"/px (that is actual sampling rate of color sensor).

In DSS again there is an option for that in RAW settings

Now stack your image and save result as fits file.

Next step would be to bin that image x2 to get your sampling rate to 1.52"/px. For that you will need ImageJ software (it is free and written in java so runs on almost any OS).

You open your fits file (for each channel, or if it is multi channel image it will open as stack) and run Image/Transform/Bin menu command. Select 2x2 and average method. Do this on each channel or once on whole stack.

After that you can save resulting image as fits again (or in case if it was opened as stack  - use save as -> image sequence, select fits format and other options and it will write individual channel images that you can combine back in photo editing app of your choice).

In case you are using Pixinsight, all above options are also available to you (at least I'm sure they are, I don't use it personally).

Btw resulting image will be halved in height and width once more after bin, so final resolution will be 1572 x 1014 (or rather close to 1500 x 1000 if you account for slight cropping due to dither between frames).

Yes, almost forgot - do dither between each sub, that will improve your final SNR quite a bit.

[Gan]

Thats great. Very clear step by step process. I will try this in my next session. Very grateful for your input. Much obliged 

BTW I found this info via a link, although its beyond my understanding

gan

[vlaiv]

10 hours ago, Gan said:

BTW I found this info via a link, although its beyond my understanding

Found similar graphs myself, but have no idea how to read them, or rather what is the meaning of DN units (or for that matter log2(DN), although I suspect it is number of bits needed for DN unit - just a log base 2 of the number).

[old_eyes]

I think @vlaiv has given you all the information you need, but just to add some practical experience.

 

I started using an unmodified Canon 1000, and got quite a lot of nice pictures with it, particularly of bright nebulae.

I then switched to a one-shot-colour cooled CMOS astro-camera (QHY168C). So a similar chip (both OSC, I think the QHY uses the same chips as the Nikon D3), similar resolution, but different electronics and set point cooling.

The difference is chalk and cheese. Much less noise, particularly in the background sky. No amp glow. Cooled detector means one set of darks lasts months.

You will get a lot more useful data, faster with an astro CMOS, but you can take beautiful shots with an unmodded, uncooled DSLR. I did for several years before I traded up.

Edited December 20, 2019 by old_eyes

[Gan]

On 20/12/2019 at 19:37, old_eyes said:

I think @vlaiv has given you all the information you need, but just to add some practical experience.

 

I started using an unmodified Canon 1000, and got quite a lot of nice pictures with it, particularly of bright nebulae.

I then switched to a one-shot-colour cooled CMOS astro-camera (QHY168C). So a similar chip (both OSC, I think the QHY uses the same chips as the Nikon D3), similar resolution, but different electronics and set point cooling.

The difference is chalk and cheese. Much less noise, particularly in the background sky. No amp glow. Cooled detector means one set of darks lasts months.

You will get a lot more useful data, faster with an astro CMOS, but you can take beautiful shots with an unmodded, uncooled DSLR. I did for several years before I traded up.

Thank you all, guys. That lead me a little closer to buying a CMOS astro camera for mostly DSO. Any suggestions ? ( budget limit £900) Prefrably OSC

[DaveS]

Under £900, OSC, ASI533.

[Gan]

Good suggestion. Thank you. I understand that its the latest reiteration of ASI 183 Pro ( colour) . Hope it is comparable to/ better than ASI 294 ( since they are similar in price). Unable to decide. Any help would be greatly aprpreciated.

[vlaiv]

Since you mentioned 8" SCT for imaging - I would say go for ASI294, but given that I've seen few people having some sort of issues with them (and I don't know what it is due to) - I can't give that recommendation.

Btw, 533 is not related to 183 in any way except maybe sensor size.

Both 183 and 533 are smaller sensors and are going to have rather tiny FOV on 2000mm of focal length. 294 is larger and FOV will be better.

183 has tiny pixels and 533 has larger but still rather small pixels (compared to 294 and DSLR sensors). 533 is square sensor, and initial impressions are that it is rather good.

Not sure what to recommend.

[Gan]

Let us suppose budget could be stretched a bit, which following parametrs would carry more weight? ( for DSO photography) : 1. Resolution ( arc sec/pixel) 2. FOV 3. resolution ( pixels) 4. Read noise 5. Pixel size 6. QE 7. Full well capacity 8. FPS

(The contenders then would perhaps include ASI 071, which is £500 more expensive)

[vlaiv]

3 hours ago, Gan said:

Let us suppose budget could be stretched a bit, which following parametrs would carry more weight? ( for DSO photography) : 1. Resolution ( arc sec/pixel) 2. FOV 3. resolution ( pixels) 4. Read noise 5. Pixel size 6. QE 7. Full well capacity 8. FPS

(The contenders then would perhaps include ASI 071, which is £500 more expensive)

That sort of question does not have a definite answer.

1, 2, 3 and 5 are connected and depend on several factors - namely what scope you want to use it on and what is your mount performance like

4. depends a bit on mount performance - how long can your subs be, but also on level of light pollution - is there a need for longer subs

6. Very important and does not depend on anything else - aim for the best value once you balance all other things.

7, 8. Totally irrelevant

[vlaiv]

7 hours ago, Gan said:

Let us suppose budget could be stretched a bit, which following parametrs would carry more weight? ( for DSO photography) : 1. Resolution ( arc sec/pixel) 2. FOV 3. resolution ( pixels) 4. Read noise 5. Pixel size 6. QE 7. Full well capacity 8. FPS

(The contenders then would perhaps include ASI 071, which is £500 more expensive)

I'm going to expand a bit on this from my previous answer which was rather short (as I was short on time, but I have a bit time now to be more elaborate).

First items 1, 2, 3 and 5. - here is what I recommend:

For scopes up to 4" in diameter, you should keep your sampling rate below 2"/px (below meaning higher number of arc seconds per pixel or lower sampling resolution). For scopes between 4" and 6" you can go with 2"-1.5"/px, for scopes 6"-8" you can go 1.5"-1"/px and for scopes 8" and above you can go with 1"/px.

That is based on size. You need to account for seeing conditions and also for your mount and guiding accuracy. In any sort of good seeing you will be limited to 1.5"/px and you need really good seeing to go below that (excellent seeing to go at 1"/px). You also need guide RMS to be at least half of target resolution if not less than that.

This all means that your realistic sampling rate is 1.2" - 1.5" per pixel for larger scope, 1.5"-2.0" for medium size aperture (about 6") and 2.0"/px and above for 4" and lower. Going for 1"/px is extremely difficult and you need large aperture scope, excellent mount (guide RMS of about 0.3") and excellent seeing conditions.

What does that mean for FOV and pixel size and resolution? Well FOV is determined by sensor size and focal length of scope. Larger sensor will be "faster" sensor - but only if you pair it with suitable telescope.

If you have your FOV set - for example you want to image galaxies and you have idea of your FOV to be approximately 40' x 30'. You can make that work with almost any sensor size, but you will need suitable focal length. You can do it with sensor that is ~15mm in diagonal, let's go with ASI183 and you will need something like 1000mm of focal length for that.

If you choose F/5 scope for this - you will work with 8" scope. Take ASI294 now and do the same FOV calculation. You will need something like 1500mm of focal length - and if you work with F/5 scope - that will be 12" scope.

So in first instance you have 8" of aperture gathering light for your FOV, while in second case you have 12" aperture gathering light for your FOV. It is clear that second case is faster - larger chip wins because you can record same FOV with larger scope (of the same type - but this costs more).

Now that you have your FOV set, and let's go with above case - 45' width (0.75 degree) at 1500mm - now you need to see what sort of sampling rate will you get. You have flexibility here, because you can bin.

First thing to realize is that you are sampling at twice the rate when using OSC camera (that also means that there is no benefit in using any other debayer technique than one that provides proper resolution - like super pixel mode).

Let's say that you want to sample at 1.2"/px, that means for OSC sensor that single pixel should be 0.6"/px. We have 45' to cover, and that is 45x60 = 2700 arc seconds.  We need a sensor that has 4500 pixels in horizontal direction (2700/0.6). ASI294 has 4144, so we are going to sample a bit "coarser" than 1.2"/px (which is still ok). We will actually sample at 1.27".

So you see, all above variables are connected, and you need to take into account:

- what sort of scope are you going to use camera with (maybe multiple scopes?)

- what targets are you going to go after with that camera (and particular scope)

- what sort of performance can you expect from your mount and skies

- take into account any binning and such

Then you can do comparison of cameras based on points 1,2,3 and 5 combined.

4. Read noise

This one is essential thing when doing planetary imaging, but for long exposure AP it is not crucial. It will only impact duration of your subs. Only difference between stacking multiple subs and taking one long exposure is in read noise (well, this is not quite true, there are other benefits to stacking like hot pixel removal, airplane/satellite trails removal, rejection of frames that have quality issues - like wind shake or passing cloud and such), but in mathematical sense - for ideal exposures, only difference is in read noise. If read noise were 0, then stack of many subs would be exactly equal to single exposure of same total length.

Since read noise is not 0, more shorter exposures will always be lower in SNR than fewer longer exposures for same total imaging time. There is but - how different will depend on how high read noise is - compared to other noise sources (and not it's absolute value).

If there is strong light pollution - it will contribute quite a bit of noise that is larger than read noise - you can go with short exposures.

Your camera is not cooled and there is thermal noise? - Again you can go short exposures and you won't be able to tell the difference.

You are in very dark skies or you do narrow band imaging - you need to go with longer exposures, because difference to shorter exposures will show.

Overall - for regular long exposure AP, you don't need to worry about read noise levels. If they are really low - like less than 2e, you can use 1-2 minute exposures in LP and 3-4 minute exposures in darker skies. If it is a bit higher like 3-4e - you simply go with longer exposures - like 3-4 minute in LP and 5-6 in dark skies (or even up to 10).

In the end you don't want to go overboard with exposure lengths because of things mentioned - algorithms work better when there is more data (more subs) and if you end up throwing away data - it's better to throw away one sub of 2 minutes than one of 20 minutes in length.

FPS - important only for planetary imaging

Full well - completely unimportant for imaging. You will likely saturate your sensor on bright stars regardless of how deep your full well is and there is really simple technique to deal with that - just take couple (and I mean really just a couple - like less than 10 is perfectly fine) - short exposures - 15s or so (even less if you have bright star that is saturating that) - and stack them separately and just use that stack to fill in original stack in saturated places.

For stars you really need only color (but we are talking about OSC cameras, so we are capturing color only) and for bright parts of target you will also need luminance (in case of LRGB and OSC).

QE - yes, very important. Really is - higher the better, no doubt about that.

 

[Gan]

Wow, thats was really educational. Really enjoyed reading it. Slowly getting there in understanding these concepts.

(I am sorry. I should have given few more facts. My area is about 4 on Bortle scale. Not sure how to quantify seeing. Scope 8 inch Edge HD. Mount is NEQ 6 Pro with  Belt modification.. Guiding with QHY 5 II L Mono and PHD. decent exposures up to 180 sec.  Currently using Canon 80 D UnMod. Main purpose is to do DSO photography)

My confusion was between ASI 294 /ASI 071/ ASI 533. Not sure which one to choose. with the help of  above information I will take a close look at what is the best combination of camera and my existing scope. Have no plans currently to spend on new scopes.

Once again many thanks Vlaiv. Much appreciated.

Gan

[vlaiv]

Usually I would recommend 071 then 294 and 533 last, but lately there are a few people having issues with these OSC cameras (071, 294). I'm not entirely sure why is that - whether it is their particular setup or something else. For this reason I'm a bit reluctant to give recommendation without any reservations. It could be that these people have issues with the way they are using their cameras (light leak, driver version, improper flats, ... there could be rather large number of reasons), and there is probably much more people out there using these cameras that are not complaining (and have them working fine).

In any case - EdgeHD has very long FL and I think that you will be best served by 071 model. With x0.7 reducer that will give you be sampling at ~1.4 (if you use super pixel mode) and will have decent FOV for DSOs

Here is comparison of FOVs on some targets:

[Gan]

11 hours ago, vlaiv said:

Usually I would recommend 071 then 294 and 533 last, but lately there are a few people having issues with these OSC cameras (071, 294). I'm not entirely sure why is that - whether it is their particular setup or something else. For this reason I'm a bit reluctant to give recommendation without any reservations. It could be that these people have issues with the way they are using their cameras (light leak, driver version, improper flats, ... there could be rather large number of reasons), and there is probably much more people out there using these cameras that are not complaining (and have them working fine).

In any case - EdgeHD has very long FL and I think that you will be best served by 071 model. With x0.7 reducer that will give you be sampling at ~1.4 (if you use super pixel mode) and will have decent FOV for DSOs

Here is comparison of FOVs on some targets:

Looks like you showed me clear direction. The pictures are very informative regarding the FOV differences. I will be on the job and will keep you posted. Many thanks for your excellent help

Gan

[Gan]

I shouldn't have procrastinated. After Xmas the price has gone up for these cameras. Now they are costing at least £150 dearer! Incidentally I read few reviews re. ASI 071: about the frosting/icing on cooling below certain temperature. I understand that the QHY conuterpart ( 168C) hasn't got the same issues. Am I right?

[Allinthehead]

On 06/01/2020 at 21:35, Gan said:

I shouldn't have procrastinated. After Xmas the price has gone up for these cameras. Now they are costing at least £150 dearer! Incidentally I read few reviews re. ASI 071: about the frosting/icing on cooling below certain temperature. I understand that the QHY conuterpart ( 168C) hasn't got the same issues. Am I right?

Generally speaking that's is true however a quick internet search shows some issues with frosting/dew on the Qhy version too. There're also plenty of folks using the 071 without issue. I have a 071 pro which is a great camera but if i push it beyond -10 it can frost. I use it at -5 because there's very little to gain going beyond that. Either version should serve you well. 

[wimvb]

23 hours ago, Allinthehead said:

I use it at -5 because there's very little to gain going beyond that. Either version should serve you well. 

The graph is misleading. It seems to me that dark current halves every 7 degrees or so. It may well keep doing so beyond - 5 C, but the graph won't show that properly. 

Your remark about frosting is very relevant, though.

[vlaiv]

1 minute ago, wimvb said:

The graph is misleading. It seems to me that dark current halves every 7 degrees or so. It may well keep doing so beyond - 5 C, but the graph won't show that properly. 

Your remark about frosting is very relevant, though.

Not sure what would be misleading in that?

Sensors have something called dark doubling temperature and indeed it ranges between 5C and 7C. Not sure why would there be issue with graph showing halving of dark current - it indeed approaches 0 asymptotically as one would expect for that case.

In any case - good temperature is one that provides negligible increase in total noise. That happens when dark current noise is less than read noise (about couple of times less, if it is five times less then you practically won't notice it at all).

Let's say that dark current is about 0.003e/s/px at -5C. In 300s exposure that is going to be 0.9e/px and associated noise will be ~0.95e.

Typical read noise for this camera is about 2.5e. These two combined give ~2.674e noise, or only about 7% increase in noise over read noise - not much really.