To CMOS or not???

[DaveS]

You can bin mono CMOS, it's just done in software so not quite the advantage of hardware binning CCD.

[vlaiv]

30 minutes ago, petevasey said:

Thanks to everyone who has got involved in this topic.  Lots of useful feedback.  But I think I've rather boxed myself into a corner here, simply because of the wide range of focal lengths I use, ranging from 420 to 2000 mm.  With  a pixel size of 3.76 um  (ASI533 and ASI2600), I will be severely oversampled on my RC10 - at 2000 mm FL it's 0.39 a-s/pixel.  The ASI294 with 4.63 um pixels is slightly better at 0.48 a-s/pixel, but still very oversampled given my average seeing of between 2 and 3 arc-secs.  The 5.4um on my QSI683 is still undersampled at 0.56 a-s/pixel, but of course it's a monochrome camera so I can bin it , then 1.12 a-s/pixel is ok, and no problem unbinned at my shorter focal lengths (420, 650 and 950 mm).

So unless I take a chance on the oversampling not causing problems, OSC CMOS is out of the question for my 'big gun", there just don't seem to be any CMOS chips with large enough pixels. 

Guess I'll be sticking with my QSI 683.

Cheers,

Peter

You can bin OSC CMOS sensors as well - just keep in mind that you increase read noise.

Say you bin ASI2600 x2 - it will no longer be 1.5e read noise camera - it will be 3.0e read noise camera, and that only impacts your exposure duration - as you want individual exposures to swamp read noise.

Oversampling will not cause any problems - except it is a waste of SNR. As long as you can bin your data in software afterwards and adapt it to actual seeing conditions (and pay attention to read noise thing) - you can recover lost SNR.

I happily use ASI1600 with 3.8µm pixel size on 1600mm of FL (RC8) - but I never tried to process image at native resolution - I always bin at least x2 if not more.

[vlaiv]

1 minute ago, DaveS said:

You can bin mono CMOS, it's just done in software so not quite the advantage of hardware binning CCD.

Actually main advantage is the same - increase in pixel size and hence SNR per sub. Difference is only in how the read noise is treated and read noise is dealt with by setting appropriate sub length.

[Magnum]

1 hour ago, petevasey said:

Thanks to everyone who has got involved in this topic.  Lots of useful feedback.  But I think I've rather boxed myself into a corner here, simply because of the wide range of focal lengths I use, ranging from 420 to 2000 mm.  With  a pixel size of 3.76 um  (ASI533 and ASI2600), I will be severely oversampled on my RC10 - at 2000 mm FL it's 0.39 a-s/pixel.  The ASI294 with 4.63 um pixels is slightly better at 0.48 a-s/pixel, but still very oversampled given my average seeing of between 2 and 3 arc-secs.  The 5.4um on my QSI683 is still undersampled at 0.56 a-s/pixel, but of course it's a monochrome camera so I can bin it , then 1.12 a-s/pixel is ok, and no problem unbinned at my shorter focal lengths (420, 650 and 950 mm).

So unless I take a chance on the oversampling not causing problems, OSC CMOS is out of the question for my 'big gun", there just don't seem to be any CMOS chips with large enough pixels. 

Guess I'll be sticking with my QSI 683.

Cheers,

Peter

Hi again Pete, interestingly although I said I wouldn't recommend the 294 due to not liking the starburst amp glow and the seemingly common issue with blotchy flats on the OSC version, I have  been looking back at a conversation I had with Wissam Ayoub about his image of m16 taken with his 294mm and his RC10 which he uses with a 0.67x reducer. now he has some amazingly detailed images with this combo https://www.astrobin.com/mwcohq/ if go to the full size version the detail on the eagle is one of the best ive ever seen. 

Not sure if you are aware but the 294 quoted pixel size of 4.6um is actually in 2x2 bin mode as default, but more recently they have implemented something called Unlocked bin x1 mode which unlocks the native resolution and pixel size of 2.3um. if you look through Wissams images some of the later ones are in the bin 1x1 using his explorer scientific ED127 refractor, and its quite impressive how he's getting similar detail using both scopes in 2x2 or 1x1.

I own a Meade 127 so im quite attracted by the 2.3um mode but not sure my uk seeing would allow for that much oversampling as he can get in UAE desert.

Wissam did send me some raw subs back in June which I seem to have lost now, but I was put off by the starburst amp glow as previously stated, but I can't deny he's getting some awesome images with that camera.

I believe he owns both the 294mm and the 294mc plus an older 1600mm.

Edited November 10, 2021 by Magnum

[petevasey]

2 hours ago, DaveS said:

You can bin mono CMOS, it's just done in software so not quite the advantage of hardware binning CCD.

I'm a bit fuzzy here!  Bear in mind I'm considering One Shot Colour cameras in this thread.  Binning CCD afaik adds the pixel values together in the early stages of download, so that with 2x2 binning the resultant super pixel has approximately (depending on the way the light falls on the group of pixels) 4 times the value of the individual pixels.  But it seems to be different for CMOS.  Or is it?  Is the binning still done at some point during download, albeit in software, so that download is faster and the screen image is smaller and brighter, or is it done much later in user controlled software?  And is the colour information scrambled as it is in binning of OSC CCDs?   If not, then it seems the only downside is the increased read noise, and with suitably long exposures not that much of a problem.

Cheers,

Peter.

[vlaiv]

3 minutes ago, petevasey said:

Or is it?  Is the binning still done at some point during download, albeit in software, so that download is faster and the screen image is smaller and brighter, or is it done much later in user controlled software?  And is the colour information scrambled as it is in binning of OSC CCDs?

You can bin in firmware - on camera, prior to download - that speeds up download (as if that is needed) and reduces file size (this can actually be beneficial), or you can bin in software after calibration, or even after stacking - while stack is still linear.

Binning can scramble color information - if not done properly. Software needs to be aware that it is binning raw color data and then it will bin it properly - but just simple "mono" binning will destroy the color information.

Yes, you are right - difference in hardware vs software binning (CCD vs CMOS) is just amount of read noise that you get. For CCD it remains the same and for CMOS you get increase by bin factor (2x read noise for bin x2, 3x read noise for bin x3, etc ...).

[petevasey]

So, the crunch question!  Is anyone out there binning an ASI OSC camera?  As I say I'm considering the 294 or even the 2600.  But would definitely want to bin the 2600, certainly 2x2, perhaps even 3x3.  Yes, if binning can be done correctly the colour data won't be lost.  But does the ASI camera driver look after that?  Or is it done by the capture program?  I use MaximDL 5, and wouldn't want to change, it does everything I want without complaining.  I should mention I'd be using the ASCOM driver in Maxim - from tests I know that works with the ASI533 in Windows XP.  I have also installed the ASI camera driver, although of course that isn't 'seen' directly by Maxim, but I presume that ASCOM interfaces with it.

Cheers,

Peter

[ONIKKINEN]

11 hours ago, petevasey said:

So, the crunch question!  Is anyone out there binning an ASI OSC camera?  As I say I'm considering the 294 or even the 2600.  But would definitely want to bin the 2600, certainly 2x2, perhaps even 3x3.  Yes, if binning can be done correctly the colour data won't be lost.  But does the ASI camera driver look after that?  Or is it done by the capture program?  I use MaximDL 5, and wouldn't want to change, it does everything I want without complaining.  I should mention I'd be using the ASCOM driver in Maxim - from tests I know that works with the ASI533 in Windows XP.  I have also installed the ASI camera driver, although of course that isn't 'seen' directly by Maxim, but I presume that ASCOM interfaces with it.

Cheers,

Peter

Not a ZWO product, but still an IMX 571 sensor camera (by rising cam, manufactured by touptek). I always bin 2x2 which is just a 50% resampling in post as far as i have understood and i perceive no lost colour data. Read noise gets "read" from each pixel regardless if your capture software captures native or bin x whatever resolution. There is no need to do binning during capture with CMOS other than reducing filesize, you can do this in post in whatever software you use to post process.

[petevasey]

8 hours ago, ONIKKINEN said:

Not a ZWO product, but still an IMX 571 sensor camera (by rising cam, manufactured by touptek). I always bin 2x2 which is just a 50% resampling in post as far as i have understood and i perceive no lost colour data. Read noise gets "read" from each pixel regardless if your capture software captures native or bin x whatever resolution. There is no need to do binning during capture with CMOS other than reducing filesize, you can do this in post in whatever software you use to post process.

Hmmm.  I don't think that's what I'm looking for when binning.  With my monochrome CCD camera the pixel values are added together during downloading, so with 2x2 binning the 4 pixel  combined 'superpixel' is approximately four times brighter, although of course limited to the maximum value of the 16 bit ADC .  And the downloaded file is four times smaller.  Helps when focusing and framing (indeed I generally do that binned 4x4), and reduces imaging time for very faint objects.   Also of course allows for matching sampling for a given optical configuration.  But if for the colour CMOS it's just a resampling process, such as can be done in Photoshop by reducing the image size, I see no advantage whatsoever.

Cheers,

Peter.

[vlaiv]

8 hours ago, ONIKKINEN said:

I always bin 2x2 which is just a 50% resampling in post as far as i have understood

That might not be true.

Binning is a form of resampling, but resampling is not binning. There are many different methods of resampling (integer resampling - which can be equal to binning, then bilinear, bicubic, spline, lanczos, to name a few).

Binning is exclusively done by adding adjacent pixel value and is done prior to post processing - while data is still linear and not stretched. Out of all mentioned methods to reduce sampling rate - only binning has certain properties. It does not introduce pixel to pixel correlation in final image and has known SNR improvement factor - which is equal to bin factor.

Other methods introduce correlation in pixels - which introduces blurring of the image that reduces detail and provides larger SNR improvement, or keeps sharpness but does not improve SNR as much as binning.

When binning stacked color image after debayering subs and stacking them - SNR improvement won't be the same as binning individual subs in adequate way for OSC subs, then debayering them and finally stacking them.

[gorann]

3 hours ago, vlaiv said:

That might not be true.

Binning is a form of resampling, but resampling is not binning. There are many different methods of resampling (integer resampling - which can be equal to binning, then bilinear, bicubic, spline, lanczos, to name a few).

Binning is exclusively done by adding adjacent pixel value and is done prior to post processing - while data is still linear and not stretched. Out of all mentioned methods to reduce sampling rate - only binning has certain properties. It does not introduce pixel to pixel correlation in final image and has known SNR improvement factor - which is equal to bin factor.

Other methods introduce correlation in pixels - which introduces blurring of the image that reduces detail and provides larger SNR improvement, or keeps sharpness but does not improve SNR as much as binning.

When binning stacked color image after debayering subs and stacking them - SNR improvement won't be the same as binning individual subs in adequate way for OSC subs, then debayering them and finally stacking them.

So Vlaiv, what method do you think is best for my ASI2600MC? So far I have only tried superpixel when debayering in PI, which seems to work well (but generally I do not bin). I seem to remeber that the ASICAP program that I use to run the camera also have a "binning" option but I never used that.

[vlaiv]

1 hour ago, gorann said:

So Vlaiv, what method do you think is best for my ASI2600MC? So far I have only tried superpixel when debayering in PI, which seems to work well (but generally I do not bin). I seem to remeber that the ASICAP program that I use to run the camera also have a "binning" option but I never used that.

That depends on what your target resolution is.

I personally view OSC sensors as having twice the lower sampling rate than their pixel size would suggest. I don't debayer in classical sense - I extract red, blue and two green subs out of single raw sub after calibration.

Then I stack each color as if data were coming from mono + filter device. Green will have twice as many subs - which is good as green carries the most luminance data and it will have best SNR.

Once you separate colors - then you can additionally bin your data if there is need - but this really happens only with long focal length scopes where you need to bin more - like x4 if we consider regular pixel size.

 

[ONIKKINEN]

5 hours ago, vlaiv said:

When binning stacked color image after debayering subs and stacking them - SNR improvement won't be the same as binning individual subs in adequate way for OSC subs, then debayering them and finally stacking them.

Interesting, yet another imaging misconseption of mine you have pointed out😄.

How big of a loss (or how much less the improvement) would resampling in post do? In SIRIL there are a few options like bilinear, bicubic, lanczos, pixel area relation etc. I would love to pretend i can see a difference between them but i dont.

If there is a big difference, would it be better then to capture binned by setting the driver to bin2? I can see straight away from HFR readings that the night is not good enough for native resolution and lose nothing of value by binning on camera.

[vlaiv]

4 minutes ago, ONIKKINEN said:

Interesting, yet another imaging misconseption of mine you have pointed out😄.

How big of a loss (or how much less the improvement) would resampling in post do? In SIRIL there are a few options like bilinear, bicubic, lanczos, pixel area relation etc. I would love to pretend i can see a difference between them but i dont.

If there is a big difference, would it be better then to capture binned by setting the driver to bin2? I can see straight away from HFR readings that the night is not good enough for native resolution and lose nothing of value by binning on camera.

See these threads:

I would personally avoid binning in drivers - as binning in software offers more flexibility.

If you lack simple way to bin - download ImageJ - you can bin image easily with that tool.

[ONIKKINEN]

8 hours ago, vlaiv said:

See these threads:

I would personally avoid binning in drivers - as binning in software offers more flexibility.

If you lack simple way to bin - download ImageJ - you can bin image easily with that tool.

Good reading here, definitely learned new valuable info but the sift bin mentioned in the first thread really doesn't work for me as there are already hundreds of subs, each of which are 300mb after calibration and conversion to 32-bit meaning that it takes forever to complete. Also i found that there is no way to know which sub is of which color after saving the new stack from the sift bin plugin? Also found that imageJ binning removes color data, the fits file becomes mono, but i found that ASTAP has a binning feature that retains color. Think ill use that on the linear stack from now on.

[vlaiv]

8 hours ago, ONIKKINEN said:

Also i found that there is no way to know which sub is of which color after saving the new stack from the sift bin plugin?

That depends on your bayer matrix order. It is most likely RGGB - but it is easy enough to test it - just shoot something red. Green channels will have equal values - red will be higher than blue.

[petevasey]

Hi, guys,

Thanks again for all the interest in this thread.  Well, despite my earlier comment that I would stick with my CCD camera I've bought an OSC!   A secondhand but nice condition Altair 294 C Pro TEC, ie. cooled.  Made some suitable adapters for my various bits of kit and after some initial hiccoughs it's working well.  Just need to do a bit of imaging to see how it compares with my CCD.   But just briefly to return to the subject of binning.  CCD of course is additive so a 2x2 binned pixel is considerably brighter (and of course smaller) than a 1x1 image.  I had a OSC SBIG camera many years ago.  Quite big pixels at 7.4um so matched quite well with my 2 m focal length telescope.  Binning wasn't really necessary, but anyway destroyed the colour information.  But on my 2m RC10 I generally bin my 5.4 um pixels QSI 683.  The pixels in the 294 are even smaller at  4.63 um so again possibly should be binned.

I've just done a test of different binning levels, 1 through 4.  All the images had the same brightness which confirms that the binning is average, not additive.  But good news that the colour information was not scrambled in any of them 😄   Incidentally dark subtraction does a great job on the 'starburst'.

I use Maxim DL for all my initial capture, calibration etc. with Altairs ASCOM driver.  Presumably the driver does software binning during download - there is very little difference in download time between 1 and 4 binning.  But of course later processing including debayering is much quicker in the smaller images.  So the crunch question.  For my longer focal lengths is it worth binning, better to match sampling?   Does it matter when the pixels are averaged rather than summed?  The 2x2 binned image is still a very acceptable size at 2064 x 1404.  And of course post processing will be quicker and the file sizes much smaller.

First light was on Tuesday 22nd Feb.  M42 was getting a bit low in the West and the seeing was very poor with quite strong winds so guiding was compromised!  294C on TS 65 mm quad (420 mm fl).  Just 9 x 5 minute subs, ASCOM gain 30 (approx 300 in some forms).  I tried 2 minutes for the trapezium, but still blown out and I had to stop - weather!!!

LATER

I've given this a bit more thought while doing an imaging run looking at both binned and unbinned subs on my 2m telescope.  And decided to stick with unbinned.  Why?  Well the main reason for taking luminance images with a monochrome camera is to get the best definition.  The colour isn't so important - when added to the Luminance image its definition is retained.  But with an OSC the resolution is compromised by the Bayer matrix, and although the finished item looks very nice, I suspect that there is ultimately a loss of resolution.  If I'm wrong, I'll stand corrected!

Cheers,

Peter

 

 

Edited February 27, 2022 by petevasey
Added text

[Adam J]

On 02/11/2021 at 21:55, DaveS said:

I just wish someone would make a BSI CMOS sensor equivalent to the 16200. APS "H" format and 6 micron pixels. Could easily have a FWC greater than 100,000 like the Sony IMX 410.

Just bin the ASI6200MM Pro 2x2 in software, the read noise is so low that its going to make very little difference over having a dedicated 6um pixel. 

Adam

[Adam J]

My view in this is that CCD is dead, with the release of the ASI6200mm pro, ASI2600mm pro and the upcoming ASI533mm pro there is a CMOS for every budget that will either perform as well as older CCDs or more likely outperform older CCDs are a lower cost to own. 

I just cant imagine buying a CCD in 2022. 

Adam 

[DaveS]

14 minutes ago, Adam J said:

Just bin the ASI6200MM Pro 2x2 in software, the read noise is so low that its going to make very little difference over having a dedicated 6um pixel. 

Adam

My original point was in regarding to a camera with the sensor size and format of the 16200. I much prefer the squarer shape of the 16200, while finding the 3:2 shape of the current crop of repurposed photographic sensors too long and thin. Unfortunately square sensors are limited to either the 533 which is too small for my purposes or the various 4040 based cameras which are crazy expensive, even worse than the 16803 which appears to be defunct.

[Magnum]

51 minutes ago, Adam J said:

My view in this is that CCD is dead, with the release of the ASI6200mm pro, ASI2600mm pro and the upcoming ASI533mm pro there is a CMOS for every budget that will either perform as well as older CCDs or more likely outperform older CCDs are a lower cost to own. 

I just cant imagine buying a CCD in 2022. 

Adam 

Even if cmos is as good or better ( thats debatable ), I think the statement CCD is dead to be ridiculous, our existing CCD cameras don't just stop working lol, and I would happily buy a another 2nd hand CCD, I already have 2 mono CCD's and 1 OSC CMOS and really don't see a big difference at this point they are both good and im happy using both.

"CCD is dead" really aggravates me and something I see quoted by the retailers that only sell cmos. I really can't see why any end user would want to be saying that when its not true at all.

Cheers

Lee

Edited March 14, 2022 by Magnum

[Adam J]

2 hours ago, Magnum said:

Even if cmos is as good or better ( thats debatable ), I think the statement CCD is dead to be ridiculous, our existing CCD cameras don't just stop working lol, and I would happily buy a another 2nd hand CCD, I already have 2 mono CCD's and 1 OSC CMOS and really don't see a big difference at this point they are both good and im happy using both.

"CCD is dead" really aggravates me and something I see quoted by the retailers that only sell cmos. I really can't see why any end user would want to be saying that when its not true at all.

Cheers

Lee

It's just an expression. Second hand is fine, if you have for sure it's still going to be capable of great images in the right hands as with anything. New is another thing though. 

[petevasey]

Postscript.  Things didn't go as well as I'd hoped.

I probably expected too much of the one shot colour system after many years of using monochrome with its finer detail advantage. And also probably expected too much of the vaunted sensitivity of the modern CMOS chips. I think part of the difficulty I was having was down to my somewhat out of date computers (still using Windows XP for capture - if it ain't broke, don't fix it!), although my W7 desktop flies along quite well.

My test images were a bit 'noisy' at times, but I suspect that if I had more powerful computers I would have gone for many more but shorter sub frames - that would seem the way many people use these cameras. Also to fully realise the potential with my existing optics would have meant considerable further expenditure, and almost certainly still not produce images as good as I obtain with my QSI CCD camera. So the decision was taken to sell the camera on, which was done with little delay and very little monetary loss. But worth a try and good practise making parts and modifying my Celestron off-axis guider which is now a little more useable!  If anyone's further interested, there is a page on my web site cataloguing my experience with the Altair camera.

Edited April 28, 2022 by petevasey

[Magnum]

55 minutes ago, petevasey said:

Postscript.  Things didn't go as well as I'd hoped.

I probably expected too much of the one shot colour system after many years of using monochrome with its finer detail advantage. And also probably expected too much of the vaunted sensitivity of the modern CMOS chips. I think part of the difficulty I was having was down to my somewhat out of date computers (still using Windows XP for capture - if it ain't broke, don't fix it!), although my W7 desktop flies along quite well.

My test images were a bit 'noisy' at times, but I suspect that if I had more powerful computers I would have gone for many more but shorter sub frames - that would seem the way many people use these cameras. Also to fully realise the potential with my existing optics would have meant considerable further expenditure, and almost certainly still not produce images as good as I obtain with my QSI CCD camera. So the decision was taken to sell the camera on, which was done with little delay and very little monetary loss. But worth a try and good practise making parts and modifying my Celestron off-axis guider which is now a little more useable!  If anyone's further interested, there is a page on my web site cataloguing my experience with the Altair camera.

Yes as with my experience I haven't found CMOS to be an improvement really, my ASI533mc is a fine camera and looking at the the data ive taken with it is about on par with my previous Atik 428osc, ive stuck with the 533 as its a decent size sensor and I like the 16 mm square format, I only sold the 428 as it was too small.

Comparing to my mono cams they are still better though as you say the worsening weather in the uk means I hardly ever manage to complete a colour image and usually just stick to Ha monochrome images, which is fine as Im a fan of monochrome astro images. ive recently had some mono ASI294 data to play with on m63 and its pretty much identical to my own data using the the Atik 383.  so for me cmos are now about even with CCD but I cant say ive noticed any improvement.

Ive gone down the hundreds of short subs route a couple of times with the 533 but came swiftly back to using long subs asI find they need far less stretching, and saves so much storage space and stacking time. Also I dont find that the cmos data smoothes up as quick as CCD data. the 383 can produce a very smooth noise free image in just 12 x 5 min exposures, while the 533 still needs noise reduction even with 100 x 5 min subs, though thats comparing OSC to mono again, but considering the Kodak 8300 chip in the 383 is supposedly one of the slowest sensors It stacks up remarkably fast.

People seem to get hung up on the lower cmos read noise , but if you live in an area with any light pollution at all, Ive found that read noise is not a factor at all, even the highest read noise ccd cameras are swamped by light pollution and shot noise, o to me I dont care about read noise at all, even in narrowband LP noise is greater than the read noise. 

im going to keep using both for now as I can produce nice images with either, but the OSC cmos takes longer.

Lee

 

[Whirlwind]

On 14/03/2022 at 16:10, Magnum said:

Even if cmos is as good or better ( thats debatable ), I think the statement CCD is dead to be ridiculous, our existing CCD cameras don't just stop working lol, and I would happily buy a another 2nd hand CCD, I already have 2 mono CCD's and 1 OSC CMOS and really don't see a big difference at this point they are both good and im happy using both.

"CCD is dead" really aggravates me and something I see quoted by the retailers that only sell cmos. I really can't see why any end user would want to be saying that when its not true at all.

Cheers

Lee

From the perspective new development for the amateur market the era of the CCD is generally over.  For science / specialist applications it is still going strong and there are pros/cons for both when used in this way.

The problem for amateur astronomy, is that CMOS is more adaptable - in effect there is no reason to develop CCDs from a business perspective.  Which is a shame as I imagine the new tech in CMOS if it could have been transferred to CCD architecture would have brought them up to similar standards in terms of read noise/sensitivity. However this is not going to be an option (unless you win the lottery). 

However CCDs will be around for many years to come, and will still provide a strong showing against CMOS cameras.