Choosing 1st OSC Cam...HELP!.

[paul mc c]

Hello all,i am considering buying a cooled osc camera,i want to upgrade from a modded dslr.

I will have to buy a used one and have allowed my self a budget of around £ 550,what would any one recommend at this budget,any help would be greatly appreciated.

Paul.

[JemC]

I'm in a similar situation and asked about changing from DSLR to OSC or Mono

I got good advice regarding this ( not what i wanted to hear, but good advice anyway), have a read through, basically you need more money 😀

Regards

[Stuart1971]

5 hours ago, paul mc c said:

Hello all,i am considering buying a cooled osc camera,i want to upgrade from a modded dslr.

I will have to buy a used one and have allowed my self a budget of around £ 550,what would any one recommend at this budget,any help would be greatly appreciated.

Paul.

Hi,

I would have a look for a good secondhand Starlight Xpress SXVR M25c, they do come up and are around that budget, I paid £850 for mine about four years ago and it’s been superb, it’s CCD as opposed to CMOS, has an APS-C size sensor, approx 26mm x 17mm and is super sensitive for a OSC, with its large pixels...darks are not really needed...Probably one of, if not the best OSC CCD made....

just my 2p worth, and others will inevitably point you towards CMOS...but that’s just not my bag at all... 👍😀

Edited December 17, 2020 by Stuart1971

[AKB]

Possible, perhaps, to find a second hand QHY8L around that price...

https://www.qhyccd.com/index.php?m=content&c=index&a=show&catid=59&id=12

 

[AstroRuz]

10 hours ago, paul mc c said:

Hello all,i am considering buying a cooled osc camera,i want to upgrade from a modded dslr.

I will have to buy a used one and have allowed my self a budget of around £ 550,what would any one recommend at this budget,any help would be greatly appreciated.

Paul.

A cooled OSC for £550 probably puts you in the range of second hand 183 colour sensors, or a second hand ZWO 533mc Pro with some good haggling I reckon. Unfortunately in these times with supply and demand the way they are, it might be a bit thin on the ground

[rnobleeddy]

I got an Atik 383L+ for a little more than that. I considered the SXVR M25c too, but the two I saw on sale were a little more again. 

[Stuart1971]

6 minutes ago, rnobleeddy said:

I got an Atik 383L+ for a little more than that. I considered the SXVR M25c too, but the two I saw on sale were a little more again. 

How is the Atik 383 colour as they get bad press...?? But the Mono is very good....

Edited December 18, 2020 by Stuart1971

[rnobleeddy]

28 minutes ago, Stuart1971 said:

How is the Atik 383 colour as they get bad press...?? But the Mono is very good....

I doubt it's the best camera in the world (in fact, one forum post described it as the worst!), but it has a larger sensor that most of the other camera's I've seen for sale in this price range, and there are plenty of images out using it that are better than I was managing with my DSLR. I never understood exactly how the OSC color version could be that much worse than the mono version given it presumably has a similar bayer matrix to all other OSC, and like most of these debates, it's very hard to find actual evidence based comparisons.

So far I haven't had much clear sky. It's noticeably (by eye) less noisy than the DSLR and the set point cooling means I can use darks. There's none of the fixed pattern noise I see on both my DSLRs. However, an hour of data on M33 did not produce better results than 3 hours with my DSLR (although I imagine the DSLR was in much better seeing and this weeks 383 data probably had fine cloud), so it's clearly not a step change in the way that I've heard people talk about their new CMOS cameras.

 

My plan was to use if for a few months. I'm slowly working myself up to mono but I'd also like to see how much time I have once real life returns to normal. The good thing about buying second hand is that you can sell the item back and so you get to try stuff out pretty cheaply. In my search I also noticed that people are often extremely optimistic about the price of second hand cameras. 70% of new might be a good guide price for most equipment, but lot's of people expect to recover 80% of the new price of a 10 year old camera. 

 

Edited December 18, 2020 by rnobleeddy

[rnobleeddy]

17 hours ago, rnobleeddy said:

I doubt it's the best camera in the world (in fact, one forum post described it as the worst!), but it has a larger sensor that most of the other camera's I've seen for sale in this price range, and there are plenty of images out using it that are better than I was managing with my DSLR. I never understood exactly how the OSC color version could be that much worse than the mono version given it presumably has a similar bayer matrix to all other OSC, and like most of these debates, it's very hard to find actual evidence based comparisons.

So far I haven't had much clear sky. It's noticeably (by eye) less noisy than the DSLR and the set point cooling means I can use darks. There's none of the fixed pattern noise I see on both my DSLRs. However, an hour of data on M33 did not produce better results than 3 hours with my DSLR (although I imagine the DSLR was in much better seeing and this weeks 383 data probably had fine cloud), so it's clearly not a step change in the way that I've heard people talk about their new CMOS cameras.

 

My plan was to use if for a few months. I'm slowly working myself up to mono but I'd also like to see how much time I have once real life returns to normal. The good thing about buying second hand is that you can sell the item back and so you get to try stuff out pretty cheaply. In my search I also noticed that people are often extremely optimistic about the price of second hand cameras. 70% of new might be a good guide price for most equipment, but lot's of people expect to recover 80% of the new price of a 10 year old camera. 

 

Last day at my old job today so found a little time for googling.

 

Turns out the CCD on the 383L colour is pretty awful, with peak QE of 40%. I don't know how the QE is calculated for color CCDs (e.g. before or after the Bayer matrix is added) but this is on par with my DSLR so I guess the only real advantage of the 383L over a DSLR is that it's cooled and so much less noisy. Most other cooled OSC cameras list QE of 60% or more, so my naive understanding is that exposure duration would need to be 50% higher to capture as much light? Probably explains the bad reviews!

That said, I much prefer to see how well things actually work rather than the numbers. If only it wasn't cloudy every night!

 

[vlaiv]

17 minutes ago, rnobleeddy said:

Last day at my old job today so found a little time for googling.

 

Turns out the CCD on the 383L colour is pretty awful, with peak QE of 40%. I don't know how the QE is calculated for color CCDs (e.g. before or after the Bayer matrix is added) but this is on par with my DSLR so I guess the only real advantage of the 383L over a DSLR is that it's cooled and so much less noisy. Most other cooled OSC cameras list QE of 60% or more, so my naive understanding is that exposure duration would need to be 50% higher to capture as much light? Probably explains the bad reviews!

That said, I much prefer to see how well things actually work rather than the numbers. If only it wasn't cloudy every night!

 

Not necessarily. I use ASI1600 and it has peak QE around 58% so not a lot higher than that, but it produces excellent images.

Compare that to 40% peak QE of KAF8300 and that is increase of 45%, so yes, peak QE comparison means - 6 hours with KAF8300 will be needed to match 4h with ASI1600, but that is not as drastic difference as one would expect.

Problem with sensor like KAF8300 is the way it is handled/used.

Todays trend is to go for short / medium exposures and lots of them. That is not the style of imaging with sensors like KAF8300. How long are your exposures? Anything less than 10 minutes and results will be poor. Ideally aim for 15-20 minute per exposure.

One of the things that distinguishes CMOS sensors to CCD sensors is level of read noise. My ASI1600 has 1.7e of read noise. KAF8300, depending on camera vendor will have about 6-8 times more than that.

Only difference between 60x1minute and 3x20minute of exposure is in read noise - higher the read noise of camera - for longer you have to expose to get the same results. I would easily expose for x10-20 longer with CCD like KAF8300 than with low read noise sensor like ASI1600. If I opt for 1-2 minute exposures with ASI1600 - go figure how long your exposures need to be with CCD like KAF8300 to overcome read noise.

Another issue with KAF8300 might be in processing. These cameras produce 16bit data, and as soon as you start calibrating that - you want to be in 32bit floating point mode, but not many people use 32bit workflow from the start.

[wimvb]

16 minutes ago, vlaiv said:

Another issue with KAF8300 might be in processing. These cameras produce 16bit data, and as soon as you start calibrating that - you want to be in 32bit floating point mode, but not many people use 32bit workflow from the start.

this is what Juan Conejero wrote in 2011 about PixInsight:

"Internally, most processes work with real or complex-valued pixels in either 32-bit or 64-bit floating point format, irrespective of the target image's data format. For 32-bit integer images, all internal calculations are always carried out in 64-bit floating point. The goal is to minimize roundoff and truncation errors, even at the cost of degrading performance. There are exceptions such as LUT-based transformations (histogram and curves for example) working on 8-bit and 16-bit integer images, and temporary working images used for structure or edge detection."

https://pixinsight.com/forum/index.php?threads/inernal-processing-bit-depth.2776/post-20274

[rnobleeddy]

17 minutes ago, vlaiv said:

Not necessarily. I use ASI1600 and it has peak QE around 58% so not a lot higher than that, but it produces excellent images.

Compare that to 40% peak QE of KAF8300 and that is increase of 45%, so yes, peak QE comparison means - 6 hours with KAF8300 will be needed to match 4h with ASI1600, but that is not as drastic difference as one would expect.

Problem with sensor like KAF8300 is the way it is handled/used.

Todays trend is to go for short / medium exposures and lots of them. That is not the style of imaging with sensors like KAF8300. How long are your exposures? Anything less than 10 minutes and results will be poor. Ideally aim for 15-20 minute per exposure.

One of the things that distinguishes CMOS sensors to CCD sensors is level of read noise. My ASI1600 has 1.7e of read noise. KAF8300, depending on camera vendor will have about 6-8 times more than that.

Only difference between 60x1minute and 3x20minute of exposure is in read noise - higher the read noise of camera - for longer you have to expose to get the same results. I would easily expose for x10-20 longer with CCD like KAF8300 than with low read noise sensor like ASI1600. If I opt for 1-2 minute exposures with ASI1600 - go figure how long your exposures need to be with CCD like KAF8300 to overcome read noise.

Another issue with KAF8300 might be in processing. These cameras produce 16bit data, and as soon as you start calibrating that - you want to be in 32bit floating point mode, but not many people use 32bit workflow from the start.

Interesting, thanks. 

I'd basically settled on 5 minute exposures but not for any apparent reason other than it seemed like a lot compared to what my old mount count manage! I have an EQ6 that tracks relatively well, so it sounds like I should have a go at much longer exposures. 

[vlaiv]

1 minute ago, wimvb said:

this is what Juan Conejero wrote in 2011 about PixInsight:

"Internally, most processes work with real or complex-valued pixels in either 32-bit or 64-bit floating point format, irrespective of the target image's data format. For 32-bit integer images, all internal calculations are always carried out in 64-bit floating point. The goal is to minimize roundoff and truncation errors, even at the cost of degrading performance. There are exceptions such as LUT-based transformations (histogram and curves for example) working on 8-bit and 16-bit integer images, and temporary working images used for structure or edge detection."

https://pixinsight.com/forum/index.php?threads/inernal-processing-bit-depth.2776/post-20274

Up until recently I was convinced that calibration in DSS was done in 16 bit mode. Now the software is open source and can be checked, but if you look at some tutorials - they often recommend using median for method for stacking darks and flats. Why? Because rationale is that median is the same as mean with gaussian type distribution - except you don't need to do any rounding.

[Stuart1971]

1 hour ago, rnobleeddy said:

Interesting, thanks. 

I'd basically settled on 5 minute exposures but not for any apparent reason other than it seemed like a lot compared to what my old mount count manage! I have an EQ6 that tracks relatively well, so it sounds like I should have a go at much longer exposures. 

Five mins is fine with a KAF8300 sensor, it will work well, it does not need to be 10-20 mins as stated earlier in this thread....

[vlaiv]

44 minutes ago, Stuart1971 said:

Five mins is fine with a KAF8300 sensor, it will work well, it does not need to be 10-20 mins as stated earlier in this thread....

Suppose we are imaging with very fast system 8" F/4 newtonian in rather poor light pollution (mag20 skies)

ATIK 383L will have something like 8e of read noise. In above conditions for luminance (and remember that these are bayer filters so each will receive less signal by 1/3) in 5 minute exposure, sky background signal will be about 571e and associated noise will be around 23.9e - only x3 more than read noise.

In order to get to x5 read noise you need to push sky background to around 1600 or x3 more - or around 15 minutes. This is in bortle 5 skies / mag20 with F/4 system and mono camera.

In darker skies and slower systems - LP will simply be less and read noise will dominate things more and longer exposures will be beneficial (one of the reasons people use longer exposures in narrowband as LP is dramatically cut down).

[rnobleeddy]

3 hours ago, vlaiv said:

Suppose we are imaging with very fast system 8" F/4 newtonian in rather poor light pollution (mag20 skies)

ATIK 383L will have something like 8e of read noise. In above conditions for luminance (and remember that these are bayer filters so each will receive less signal by 1/3) in 5 minute exposure, sky background signal will be about 571e and associated noise will be around 23.9e - only x3 more than read noise.

In order to get to x5 read noise you need to push sky background to around 1600 or x3 more - or around 15 minutes. This is in bortle 5 skies / mag20 with F/4 system and mono camera.

In darker skies and slower systems - LP will simply be less and read noise will dominate things more and longer exposures will be beneficial (one of the reasons people use longer exposures in narrowband as LP is dramatically cut down).

I'm realizing that I have a lot to learn - is there a good intermediate level guide that explains this? 

In particular, the relationship between the sky background level and read noise isn't something I'd considered. I assumed it was all about the signal from the target vs the background noise.

Although I've never really looked that closely, I had a look at an un-stretched calibrated image from M33 from the ATIK 383L compared to a 550D. Same exposure time but different nights so entirely unscientific, but the ATIK is roughly averaging a value of 800 (in FITS liberator) in the center of the galaxy and maybe 600 in the dark background of the image.  Readings from a similar image with a 550D are 2100 background and maybe 2500 in the center of the galaxy, which (albeit a meaningless comparison) would suggest a better SNR for the ATIK? However, I honestly never realized the image was so faint in both cases, so I'm starting to realize why this all matters so much!

[wimvb]

9 hours ago, rnobleeddy said:

I'm realizing that I have a lot to learn - is there a good intermediate level guide that explains this? 

I think that Stan Moore in Robert Gendler’s book ”Lessons from the Masters” comes closest.

There are also some interesting articles on Craig Stark’s website

http://www.stark-labs.com

[alacant]

On 17/12/2020 at 13:48, paul mc c said:

upgrade from a modded dslr

Hi

What are you looking for in the upgrade which your dslr doesn't provide?

I've often been urged to dump my dslr for something 'better', but I've not yet seen anything which comes close. I once tried an asi294 and apart from it's smaller sensor (and you really notice the change even if it's only to 3/4) it was hard to distinguish its frames from the dslr. Anything APS-c is gonna cost you €silly! 

Just my €0,02. Cheers

Edited December 19, 2020 by alacant

[vlaiv]

11 hours ago, rnobleeddy said:

In particular, the relationship between the sky background level and read noise isn't something I'd considered. I assumed it was all about the signal from the target vs the background noise.

There are different types of noise that add up to form "background" noise. There is shot noise - which is associated with target itself and depends on strength of signal (square root of it), then there is LP noise - which is just the same thing but related to sky glow that is signal that we don't want and remove in processing, then there is dark noise - which is the same thing associated with build up of dark current signal (that we remove in calibration) and in the end there is read noise.

Out of these four main types of noise, only read noise is "per exposure" - all others depend on time - more time, more signal gathers and higher each associated component of the noise (as all three are equal to square root of respective signal).

If read noise was 0 then there would absolutely be no difference between 1x600s and 600x1s (or any other combination of exposure lengths that add up to the same total time). However, since read noise is not zero, there is difference and difference depends on how small read noise is in comparison to any of other noise values.

Common rule of the thumb is to have read noise be smaller about 5 times than the highest of other components - which is usually LP noise since we shoot with cooled cameras and image very faint targets. This ensures that SNR difference between the stack and one single long exposure is few percent.

Noise adds like linearly independent vectors (square root of sum of squares) while signal adds like regular addition.

Imagine you have R being read noise and LP being light pollution noise and 5R = LP, or read noise is 5 times smaller than light pollution noise. Let's add them up.

We will have sqrt( (5R)^2 + R^2 ) = sqrt(25R^2 + R^2) = sqrt(26R^2) = R*sqrt(26) = ~5.099 * R

So resulting noise is 5.1R or 1.02LP about 2% higher than LP alone. Now let's try that same thing if R=LP, or when read noise is equal to light pollution noise.

sqrt(R^2+R^2) = sqrt(2R^2) = R*sqrt(2) = R*1.4142 ....

Now we get result as if LP noise is increased by 41% - much higher increase.

This shows that if you select such exposure that read noise is 5 times less than light pollution noise - it is like imaging single exposure in 2% more noisy light pollution (or light pollution being only 4% as strong). Imaging when you have the same level of read noise and LP noise is like imaging single exposure in twice as strong light pollution!

11 hours ago, rnobleeddy said:

Although I've never really looked that closely, I had a look at an un-stretched calibrated image from M33 from the ATIK 383L compared to a 550D. Same exposure time but different nights so entirely unscientific, but the ATIK is roughly averaging a value of 800 (in FITS liberator) in the center of the galaxy and maybe 600 in the dark background of the image.  Readings from a similar image with a 550D are 2100 background and maybe 2500 in the center of the galaxy, which (albeit a meaningless comparison) would suggest a better SNR for the ATIK? However, I honestly never realized the image was so faint in both cases, so I'm starting to realize why this all matters so much!

In order to compare the actual signal - you need to convert ADU to electrons.

CCDs usually have fixed e/ADU conversion factor and ATIK 383L has it at 0.41e/ADU

(screen shot from ATIK website)

This means that you need to multiply pixel values from ATIK image by 0.41e/ADU to get number of captured electrons.

With DSLR it is not so easy as they have changing gain - each ISO setting has different e/ADU value and you need to figure out e/ADU value for the gain setting you used in order to get actual electron count so you can compare the two.

Only then you'll be able to tell which one recorded stronger signal.

 

[Stuart1971]

1 minute ago, vlaiv said:

There are different types of noise that add up to form "background" noise. There is shot noise - which is associated with target itself and depends on strength of signal (square root of it), then there is LP noise - which is just the same thing but related to sky glow that is signal that we don't want and remove in processing, then there is dark noise - which is the same thing associated with build up of dark current signal (that we remove in calibration) and in the end there is read noise.

Out of these four main types of noise, only read noise is "per exposure" - all others depend on time - more time, more signal gathers and higher each associated component of the noise (as all three are equal to square root of respective signal).

If read noise was 0 then there would absolutely be no difference between 1x600s and 600x1s (or any other combination of exposure lengths that add up to the same total time). However, since read noise is not zero, there is difference and difference depends on how small read noise is in comparison to any of other noise values.

Common rule of the thumb is to have read noise be smaller about 5 times than the highest of other components - which is usually LP noise since we shoot with cooled cameras and image very faint targets. This ensures that SNR difference between the stack and one single long exposure is few percent.

Noise adds like linearly independent vectors (square root of sum of squares) while signal adds like regular addition.

Imagine you have R being read noise and LP being light pollution noise and 5R = LP, or read noise is 5 times smaller than light pollution noise. Let's add them up.

We will have sqrt( (5R)^2 + R^2 ) = sqrt(25R^2 + R^2) = sqrt(26R^2) = R*sqrt(26) = ~5.099 * R

So resulting noise is 5.1R or 1.02LP about 2% higher than LP alone. Now let's try that same thing if R=LP, or when read noise is equal to light pollution noise.

sqrt(R^2+R^2) = sqrt(2R^2) = R*sqrt(2) = R*1.4142 ....

Now we get result as if LP noise is increased by 41% - much higher increase.

This shows that if you select such exposure that read noise is 5 times less than light pollution noise - it is like imaging single exposure in 2% more noisy light pollution (or light pollution being only 4% as strong). Imaging when you have the same level of read noise and LP noise is like imaging single exposure in twice as strong light pollution!

In order to compare the actual signal - you need to convert ADU to electrons.

CCDs usually have fixed e/ADU conversion factor and ATIK 383L has it at 0.41e/ADU

(screen shot from ATIK website)

This means that you need to multiply pixel values from ATIK image by 0.41e/ADU to get number of captured electrons.

With DSLR it is not so easy as they have changing gain - each ISO setting has different e/ADU value and you need to figure out e/ADU value for the gain setting you used in order to get actual electron count so you can compare the two.

Only then you'll be able to tell which one recorded stronger signal.

 

🤯🤯🤯🤯

[wimvb]

36 minutes ago, Stuart1971 said:

🤯🤯🤯🤯

As I wrote before, check out Stan Moore's chapter in "Lessons from the Masters"

[rnobleeddy]

3 hours ago, alacant said:

Hi

What are you looking for in the upgrade which your dslr doesn't provide?

I've often been urged to dump my dslr for something 'better', but I've not yet seen anything which comes close. I once tried an asi294 and apart from it's smaller sensor (and you really notice the change even if it's only to 3/4) it was hard to distinguish its frames from the dslr. Anything APS-c is gonna cost you €silly! 

Just my €0,02. Cheers

Perhaps if you stick with OSC this is partially true. I've certainly got good images from my DSLR with sufficient integration time. On the flip side, at least the Canon 450D/550D I've used tend to have noise that makes it to the final image if you don't have sufficient integration time and don't perform large dithers every sub. I imagine newer sensors are better, but then they're more comparable in price. For example, the Canon EOS Ra costs £2500 so it firmly in the same range premium cooled astro cameras.

Aside from the discussion here about how best to use my specific OSC cooled CCD, it's clear that they do offer advantages - the noise in a single sub is visibly lower, and with set point cooling you can easily use darks, which isn't realistic with a DSLR.  Whether this translates to significantly better final images and whether that justifies the cost remains to be seen!

 

[rnobleeddy]

1 hour ago, vlaiv said:

There are different types of noise that add up to form "background" noise. There is shot noise - which is associated with target itself and depends on strength of signal (square root of it), then there is LP noise - which is just the same thing but related to sky glow that is signal that we don't want and remove in processing, then there is dark noise - which is the same thing associated with build up of dark current signal (that we remove in calibration) and in the end there is read noise.

Out of these four main types of noise, only read noise is "per exposure" - all others depend on time - more time, more signal gathers and higher each associated component of the noise (as all three are equal to square root of respective signal).

If read noise was 0 then there would absolutely be no difference between 1x600s and 600x1s (or any other combination of exposure lengths that add up to the same total time). However, since read noise is not zero, there is difference and difference depends on how small read noise is in comparison to any of other noise values.

Common rule of the thumb is to have read noise be smaller about 5 times than the highest of other components - which is usually LP noise since we shoot with cooled cameras and image very faint targets. This ensures that SNR difference between the stack and one single long exposure is few percent.

Noise adds like linearly independent vectors (square root of sum of squares) while signal adds like regular addition.

Imagine you have R being read noise and LP being light pollution noise and 5R = LP, or read noise is 5 times smaller than light pollution noise. Let's add them up.

We will have sqrt( (5R)^2 + R^2 ) = sqrt(25R^2 + R^2) = sqrt(26R^2) = R*sqrt(26) = ~5.099 * R

So resulting noise is 5.1R or 1.02LP about 2% higher than LP alone. Now let's try that same thing if R=LP, or when read noise is equal to light pollution noise.

sqrt(R^2+R^2) = sqrt(2R^2) = R*sqrt(2) = R*1.4142 ....

Now we get result as if LP noise is increased by 41% - much higher increase.

This shows that if you select such exposure that read noise is 5 times less than light pollution noise - it is like imaging single exposure in 2% more noisy light pollution (or light pollution being only 4% as strong). Imaging when you have the same level of read noise and LP noise is like imaging single exposure in twice as strong light pollution!

In order to compare the actual signal - you need to convert ADU to electrons.

CCDs usually have fixed e/ADU conversion factor and ATIK 383L has it at 0.41e/ADU

(screen shot from ATIK website)

This means that you need to multiply pixel values from ATIK image by 0.41e/ADU to get number of captured electrons.

With DSLR it is not so easy as they have changing gain - each ISO setting has different e/ADU value and you need to figure out e/ADU value for the gain setting you used in order to get actual electron count so you can compare the two.

Only then you'll be able to tell which one recorded stronger signal.

 

Thanks for the detailed reply! What's more, it makes sense. My mistake was assuming the thing I'm imaging had a signal far above the background sky - of course, that's because of what I see in the final stretched image, but I'd do better to assume it's the same level as the background sky for these purposes.

[rnobleeddy]

1 hour ago, wimvb said:

As I wrote before, check out Stan Moore's chapter in "Lessons from the Masters"

I've ordered a copy. Whilst it may not be everyone's idea of fun, I remember a lot of CCD vs film debates from my PhD group many years ago, so I look forward to learning!

[Stuart1971]

40 minutes ago, rnobleeddy said:

Perhaps if you stick with OSC this is partially true. I've certainly got good images from my DSLR with sufficient integration time. On the flip side, at least the Canon 450D/550D I've used tend to have noise that makes it to the final image if you don't have sufficient integration time and don't perform large dithers every sub. I imagine newer sensors are better, but then they're more comparable in price. For example, the Canon EOS Ra costs £2500 so it firmly in the same range premium cooled astro cameras.

Aside from the discussion here about how best to use my specific OSC cooled CCD, it's clear that they do offer advantages - the noise in a single sub is visibly lower, and with set point cooling you can easily use darks, which isn't realistic with a DSLR.  Whether this translates to significantly better final images and whether that justifies the cost remains to be seen!

 

Why can’t you use darks with a DSLR, I did all the time...the only issue is you need to take on the night of imaging after or before the lights....but it works extremely well....same sub length and will be same temp too, which is the important bit... 👍🏼😀

In fact with the canon DSLR I had it could be set to automatically take a dark of the same sub length after each light frame, this obviously doubled your imaging time if you did after every image, so I just tended to do at the end in one go...20 or so...

Whereas with a dedicated Astro cam you can build a library at any time with different exposure lengths and temps...