Short Exposure DSO Imaging

[nmoushon]

*DISCLAIMER*  I have spent 2 year with DSLR and 2 years with CCD. Though I am not well versed in the technical side on CMOS/CCD I do have experience with using them.

I was thinking about starting this in the camera section but didn't want this to start focusing on just the camera but also the theory, so to say, as a whole of short exposure imaging; specifically in regards to DSO imaging.  

I am well aware of the norm and the general consensus that longer exposures are better than short exposures. Though relatively short to many of you, I thought this for the entirety of  my imaging career. I have seen the progress I have gained from not only switching from DSLR to CCD but also from a max of 2min unguided subs to 15min guided subs. So I am not refuting any of this at all. But I have come across several very good images that, to me, look like 8-10 hrs of 20-30min subs but are in fact no more than usually a few seconds of exposure length(I will post links below to examples). After seeing those images I had to really stop and think if I was missing something. I took a look at all the equipment and noticed that all these imaging where taken with uncooled mono or color planetary CMOS cameras. Not the traditional cooled mono CCD camera typical of DSO imaging. 

Example 1 - short exposures

Example 2 - short exposures

Example 3 - short exposures

Example 4 - short exposures

Example 5 - short exposures - 30 sec

Example 6 - short exposures - 64 sec

Most of the examples are very short, except for the last two but even those are considered short when it comes to long exposure DSO imaging. In my opinion ALL of these are comparable to almost any picture I could find of the on Astrobin. Especially when you consider the exposure lengths. I even think these example are better than a lot of the images that were taken using 15min exposures. Now I'm not saying these out do any CCD image. There are some examples of far superior images with CCDs and longer exposures. Especially the Rosetta Nebulae example as that is a popular target. 

After finding those examples I started looking into the cameras more and to find out why planetary cams could match CCDs, especially the entry level CCD cams. I found that the major difference between the two cameras (besides one using a CMOS sensor and the other CCD) was their QE% and read noise. These CMOS planetary cams have a MUCH higher QE and a MUCH lower read noise than any CCD outside of research grade equipment. Now there are plenty of other differences but these where the major one that would contribute to the quality of the images in the example. 

Now the biggest problem I saw was that they are all uncooled. This as been mentioned in several of the posts as a problem and limited their exposure lengths as the internal temp was getting too high. Also many of them have a built-in exposure length limit since they were not designed for long exposures. But as I was researching all of this I came across that ZWO was making cooled versions of their ASI cameras that all these examples have been using. These actually were just released for sale in the last week or so. So there is not a lot of info on how they are working yet but there was one very good thread that I am following in regards to someone who has a new cooled ASI 175MM and is testing it out...well mostly waiting on weather but he has posted some good first examples. You can find the thread here: http://www.cloudynights.com/topic/519359-asi174mm-cool/

Now this is exact same camera as the examples above just it was redesigned to allow for active cooling...just like a tradition CCD. So same CMOS sensor, same QE, same red noise, etc. There is an example image on post #17 on the first page of the thread that is very impressive. Very comparable to several images posted with 15min subs. It will be interesting to see what the results are of very short exposures and stacking hundreds, if not thousands of them, like the examples above. He is planning on testing this later and will update when he does. 

By now some of you are bound to be thinking that I am advocated that everyone throw out their CCDs and switch over to a cooled CMOS but I'm not. Part of me is saying this is incredable and a great alternative for people that want to image DSOs but dont have the money for a good quality CCD or a mount with the ability to take 15-30min exposures. And the other part of me is saying there is something I am missing here and it can't be this easy. How can 4 sec subs compare with 15min subs?! It goes against everything that I've learned while in AP!  So this thread is to serve the dual purpose of sharING some incredible images that I found and to bring up a discussion on these new cameras and how they open up a "new" door for DSO imaging...especially for those that can't afford the proper mount to guide for 15-30mins. 
 

Edit: Sorry, one thing I should mention is that the one main thing I found that all the first set of examples had in common was that they all where taken with large aperture scopes that where either natively fast or reduced to have a fast f/ratio which was usually around F/5ish. So I'm pretty sure its safe to say that the large apertures have a lot to add to light gathering capabilities and how deep those images goes. So this brings up a couple questions. Do these camera only work well for DSO imaging with large aperture scopes at fast f/ratios? So are they not very good when paired with say a small ED80 or does it work just as well? Who knows....its will be interesting to see where this all goes and if this imaging style catches on or not. Especially at the price point of the cameras. 

[riklaunim]

Sony CCDs are still comparable with Sony CMOS sensors. Some CCDs still offer lower read noise and similar high QE. For DS imaging it will be some time before they will go extinct. The reason why those CMOS based cameras are used on short exposures (aside some of them lacking cooling) is the low read noise combined with planetary photographer skills - if the signal is above noise level then stacking a lot of frames will do the job. Plus some of those cameras do have noticeable amp glow so very long exposures aren't the best for them.

In general the new cooled ZWO (and QHY too) Sony CMOS based cameras could be a good 2 in 1 cameras.

[Martin Meredith]

This imaging style is also catching on in the near-real time observing community where having the image build up via live stacking is a real game-changer as it allows one to go very deep yet not hang around for long (15s or so) for the first image to appear. 

And it is not just ultra low read noise CMOS that can do the trick. So long as sky noise exceeds read noise, capturing X photons in N short subs is approx equivalent to capturing N*X photons in one long sub. So In poor-to-moderate skies (say SQM of 18-19) you can get away with quite a bit of read noise and still use short subs with no loss in quality. Nowadays I choose a sub length based on sky quality; sometimes as short as 5s, sometimes as long as 30s….

This thread (post 27) has some plots which show the SNR loss from short exposures relative to longer exposures as a function of sky conditions.

Martin 

[Thalestris24]

Hi

I had a go with the qhy5l-ii mono recently - here. I normally only use it has a guide cam but it's high QE lends itself to some experimentation, especially when normal imaging time is in short supply. It is a noisy camera though. Qhy have brought out a tec cooled one and a version witth built-in fw and mini filters. I'm watching zwo's development. There are some good examples of what can be done in the video astronomy section. I so wish I had darker skies since that would make a huge difference to all my imaging efforts!

Louise

[sharkmelley]

I'm restricted to 30sec exposures on my Sony A7S because of the bulb-mode "star-eater" issue.  These short exposures are fine on my scope because the skyfog noise is much greater than the read noise.  This is the crucial criterion for imaging efficiency, as Martin has already mentioned.

The only problem with short exposures is that I end up having to stack hundreds of them.

Mark

[ollypenrice]

I'm restricted to 30sec exposures on my Sony A7S because of the bulb-mode "star-eater" issue.  These short exposures are fine on my scope because the skyfog noise is much greater than the read noise.  This is the crucial criterion for imaging efficiency, as Martin has already mentioned.

The only problem with short exposures is that I end up having to stack hundreds of them.

Mark

I'd have expected a near-catastrophic fall-off in the usefulness of more than abut 40 subs based on the square root principle. Is this not the case? I've no idea because I've never tried it. Interested though. Big reductions in read noise could influence the way we image.

Olly

[Martin Meredith]

I'd have expected a near-catastrophic fall-off in the usefulness of more than abut 40 subs based on the square root principle. Is this not the case? I've no idea because I've never tried it. Interested though. Big reductions in read noise could influence the way we image.

Olly

This 30-40 subs idea has always struck me as unhelpful, and actually false at a number of levels. 

For a start, subs shouldn't enter into the equation. What matters are photons collected, and it would be very odd indeed if somehow the 30-40 subs rules applied in the same way to 1s subs as to 1200s subs. (Each one of your 1200s subs contains 1200 x 1s worth of photons, but you don't truncate after 30s).

Let's consider the SNR. We often see the change in SNR 1/sqrt(signal) plotted (the middle curve). Here, yes, there is a fairly rapid change in gradient at 30-40. But the SNR (lower plot) shows no such 'cliff', and surely it is the SNR that ultimately matters? One way to think about why there is no sharp change in SNR is because SNR is actually just the top curve multiplied by the second curve, and although the lower curve levels out, the signal keeps increasing.

I'd argue that deciding when to call it a day re collecting photons depends on the style of observing/imaging. For me, I might stop when the SNR is high enough that I can make a clear detection of a faint object, or when some faint part of the image sits above the noise. For an astrophotographer the criterion might be to achieve an SNR such that the noise is no longer perceptually-salient. Neither of these depends directly on the number of subs but on the number of photons collected (per pixel).

Well, that's my thinking on the matter; happy to be shown the error of my ways though!

Martin

[dph1nm]

My crystal ball tells me this (i.e. essentially read-noise free cameras) is the future for astro imaging (although I am sure the driver for all this will be terrestrial imaging).  With zero or very low read-noise it makes no sense to do long exposure imaging - all that messing around with autoguiding and trying to build high precision mechanical mounts ...

NigelM

[Thalestris24]

I think you'd have to take account of the response of the sensor i.e. nothing = nothing however many times you do it! So the outcome will vary from camera to camera and hence a short exposure gets more from a high QE sensor. Just my thinking!

Louise

[ollypenrice]

My crystal ball tells me this (i.e. essentially read-noise free cameras) is the future for astro imaging (although I am sure the driver for all this will be terrestrial imaging).  With zero or very low read-noise it makes no sense to do long exposure imaging - all that messing around with autoguiding and trying to build high precision mechanical mounts ...

NigelM

Well I guess we all hope you're right!!!

This three hundred panel mosaic was captured last night using...

 lly

[Owmuchonomy]

This is an interesting topic although I don't begin to understand the intricacies of imaging to be honest. What I will say is that my last 2 DSO images were produced from lots (100+) of shorter 90 sec subs than my usual technique of say 20 to 30 of several minutes duration. The outcome was that I could process the latest images to a much much better noise free result than the former. Don't ask me why, I just can. Feel free to tell me why.

[StuartJPP]

Easy enough to test with real data...take your 100+ sub collection and only process 30 or 40 of them and see what comes out the other side...

[nmoushon]

Martin - Thanks for your comments. I read your post on the other thread and skimmed through the rest of the thread(theres a lot there for one lunch time worth of reading) and you have done some very well done studies. I know you took that data mainly for live observing but I think it should still apply for AP as well...at least for the most part.

I was amazed at how much sky quality really effected the data collected. I knew sky quality was important but never thought the change would be that dramatic. From one end to the other that is. Though your data also suggests that there is relatively not as much change in the duration of the exposure per sky quality as there would be from changing sky quality. Basically the better your skies the longer your subs should be. 

What would be an interesting study is to compare the SNR of a 2 typical 8hr images but one made of short exposures and one of long exposures under the same conditions. You could extrapolate that into different data sets. One using same CCD camera (typical read noise and QE), one using same CMOS camera(low read noise and high QE) and then one comparing the two cameras. I wish I had the time and money to work that all out. I think it would be a fun and very interesting experiment to see what it actually shows. Especially compared to the data you showed.

Now what I would also like to see is if aperture play any part into the data you collected. How would it effect it? Would it be a linear movement in all direction and the SNR doesnt change just to total amount of photons is higher? I dont know. This whole thing has got me worked up into a frenzy and that is bad! Christmas is around the corner and its bad timing as all I want to do now is buy a new camera to test this all out. lol 

[nmoushon]

My crystal ball tells me this (i.e. essentially read-noise free cameras) is the future for astro imaging (although I am sure the driver for all this will be terrestrial imaging).  With zero or very low read-noise it makes no sense to do long exposure imaging - all that messing around with autoguiding and trying to build high precision mechanical mounts ...

NigelM

I will be interesting to see where the technology goes and what becomes the new norm. I also would like to see larger pixels on these style of CMOS cameras. Low read noise, high QE and 7-9um pixels would be perfect for long focal length imaging.

[Martin Meredith]

I did a quick simulation of an increase in aperture and the only effect I can see is that the curves shift to the right by a constant amount. The curves in the original thread were for my 0.2m Quattro. Doubling the diameter, so quadrupling the aperture, shifts the curves by about 2 magnitudes. In practical terms, if I'm looking for say a mag 15 target with a low read noise camera even in SQM 22 skies, there is predicted to be only a 10% difference between 1s and 30s subs (whereas with the 0.2m mirror there would be a 20% loss in SNR in this extreme condition). With a super-large 2m mirror there is predicted to be no difference in this condition. So in short, and from a quick analysis, it seems the advantage of short subs in the low read noise setting is enhanced by aperture. The same aperture benefit is true for the moderate read noise Lodestar that I actually use but from a lower target magnitude baseline. That seems to be saying that the benefits of aperture and low read noise do not interact (i.e. read noise doesn't seem to leverage aperture -- hmm, the grammar sounds wrong but you know what I mean). Of the two, the biggest win comes from low read noise. 

So the combination of a low read noise camera + large dob + equatorial platform is interesting. What's lacking for those of us who like near real-time is the software, and I also wonder how good current platforms are. With my interest in all things galactic I personally would love to make use of this long FL approach for seeing more detail emerge in interacting galaxies during an observing session.

Martin

[sharkmelley]

I'd have expected a near-catastrophic fall-off in the usefulness of more than abut 40 subs based on the square root principle. Is this not the case? I've no idea because I've never tried it. Interested though. Big reductions in read noise could influence the way we image.

Olly

No, not at all.  I'm regularly using 300-400 exposures in an image now.  As Martin says above, it's the total number of photons collected that is important.  There's no reason for short exposures to give a lower quality of final image than long exposures with the same total exposure time - as long as the short exposures are truly skyfog limited.

It's time consuming to stack all those exposures though.

Mark

[nmoushon]

No, not at all.  I'm regularly using 300-400 exposures in an image now.  As Martin says above, it's the total number of photons collected that is important.  There's no reason for short exposures to give a lower quality of final image than long exposures with the same total exposure time - as long as the short exposures are truly skyfog limited.

It's time consuming to stack all those exposures though.

Mark

That is interesting to hear.

Also if you think 400 subs is time consuming you should look at the example above. He used 20,000 subs! Yes, 20k subs lol. Thats just crazy. But the final images proves itself. 

[Thalestris24]

That is interesting to hear.

Also if you think 400 subs is time consuming you should look at the example above. He used 20,000 subs! Yes, 20k subs lol. Thats just crazy. But the final images proves itself. 

I guess it's essentially taking a video with a 1 frame/sec frame rate and using Registax to stack them. The imager also did '000s of longer subs. He made a nice image out of it all!

Louise

[nmoushon]

I did a quick simulation of an increase in aperture and the only effect I can see is that the curves shift to the right by a constant amount. The curves in the original thread were for my 0.2m Quattro. Doubling the diameter, so quadrupling the aperture, shifts the curves by about 2 magnitudes. In practical terms, if I'm looking for say a mag 15 target with a low read noise camera even in SQM 22 skies, there is predicted to be only a 10% difference between 1s and 30s subs (whereas with the 0.2m mirror there would be a 20% loss in SNR in this extreme condition). With a super-large 2m mirror there is predicted to be no difference in this condition. So in short, and from a quick analysis, it seems the advantage of short subs in the low read noise setting is enhanced by aperture. The same aperture benefit is true for the moderate read noise Lodestar that I actually use but from a lower target magnitude baseline. That seems to be saying that the benefits of aperture and low read noise do not interact (i.e. read noise doesn't seem to leverage aperture -- hmm, the grammar sounds wrong but you know what I mean). Of the two, the biggest win comes from low read noise. 

So the combination of a low read noise camera + large dob + equatorial platform is interesting. What's lacking for those of us who like near real-time is the software, and I also wonder how good current platforms are. With my interest in all things galactic I personally would love to make use of this long FL approach for seeing more detail emerge in interacting galaxies during an observing session.

Martin

Thanks for doing that quick study. I'm actually surprised that aperture didnt have near as big of an effect as I thought it would. I would have placed my bets that aperture would have allowed you to go deeper and collect more photons. 

One of the persons in the examples above is actually using a 20" Dob. If i remember right I think he was the one that was saying that he was limited in sub length bc of field rotation. I know for your use that wouldnt be much of a problem as i doubt you would be using exposure lengths that long for live observing. But for us interested in AP it would. Im with you and wondering how well the typical Dob EQ platform performs for AP purposes. What I do see the advantage of large aperture is keeping the f/ratio fast but keep the FL long. A F/5 20" Dob is about the same FL as my 9.25" SCT at F/10. And I'm sure that would play a big difference in at least for the live viewing, especially with faint objects. Would also help in reducing the amount of total subs needed wouldnt it? (For AP that is)

[nmoushon]

I guess it's essentially taking a video with a 1 frame/sec frame rate and using Registax to stack them. The imager also did '000s of longer subs. He made a nice image out of it all!

Louise

Thats what he did. It just totals up to hours of video instead the usual minutes for planetary AP.

[Martin Meredith]

Aperture is helping for sure. Don't forget also that my plots show SNR loss rather than SNR itself, so not all of the points plotted correspond to a useable SNR (e.g. looking for a mag 22 target in a mag 18 sky); my expectation would be that an increase in aperture will also show up as an increase in the SNR associated with any given point on the plot, but I haven't yet tested that.

When I get more time I'll try to devise a representation that shows both SNR and SNR loss in the same plot. This would then define a useful operating range for any given combinations of sky conditions/camera parameters/sub length. The code is quite simple so if you're into coding send me a PM and you can play around with it (currently Matlab but apart from the plots it doesn't do anything particularly Matlab-esque).

Martin 

[Physicist13]

I always reccomend reading Craig Starks notes/papers on SNR...very well explained.

The "CCD" eqn has it all in it really...worth noting, is that in the absence of read noise (or when read noise << other(random) noise sources) then only the total exp time matters for a given SNR (ie 60x1sec=1x60sec) . In reality read noise is finite and often non negligible so preference is for longer subs (limited by setup and non saturation). APerature increases resolution and not SNR to first order.

Patrick

[dph1nm]

APerature increases resolution and not SNR to first order.

S/N per object is very much aperture dependent - hence the ever increasing aperture sizes for professional telescopes.

NigelM

[petevasey]

Lots of mentions of stacking hundreds of images.  But what method of stacking is the most effective?  Average, Median, Sum?  Or perhaps a combination of these?

Cheers,

Peter.

[jACK101]

I am puzzled by this discussion. I am coming at this issue after a lifetime as an amateur photographer. In conventional photography, let's say landscape, the exposure would be a simple function of aperture, shutter speed and film speed, or nowadays sensor ISO setting.

 

I cannot think of any situation where many hundreds of grossly underexposed frames when added together could give an acceptable result because the information would not be there on the film or sensor.

 

I can understand the need to have different exposures for different brightness levels which when stacked give an acceptable result, but why does astrophotography seem to work to different rules?

 

Jack