81 replies to this topic

[Shawnxc]

Recently there are a lot of interests and discussions in short exposure “lucky/high resolution” deep sky imaging. Very good results had been obtained and people have different opinions, but there is a lack of subjective evidence why it works, what are the trade-offs and how to optimize this technique. Personally, I think the best way to compare or optimize this technique is to be based on total integration time and the final signal to noise ratio in the finished pictures. Fortunately, there were similar discussions 10 years ago base on CCD technology at the time, and a few wise men has already figured out the math. Steve Cannistra had a nice write-up on this topic, still pretty relevant today’s topic. He simplified the calculations to a very simple formula: tsub = F^2*R^2 / [Sky*(1-F^2)]. Where F is the ratio of picture quality (S/N) of the finished image to that of the ideal achievable for an object with a given total integration time and sky condition. The beauty of this method is that it is total integration time based subjective image quality comparison with varying sub-exposure time. Details please read his write-up at:  http://www.starrywonders.com/snr.html

 

Putting this formula in a spreadsheet, the attached chart shows how sub-exposure time affects final S/N ratio for today’s low read noise cameras. The sky flux used is 80e/min, estimation of my light polluted suburban night sky.

 

As you can see, for a camera with read noise of 1e, ~3 seconds of sub-exposure can get 90% of S/N ratio of that of traditional longer sub-exposures with same amount of total exposure time. 0.7s(!) sub-exposure will still be able to get 70% of S/N. Most of the cameras are not there yet, but a few of the CMOS cameras are pretty close.

 

However, short sub-exposure has trade-offs: all these cases will result in F<1, which means a reduced S/N ratio, that will need increased total integration time to compensate. F>=0.9 is maybe negligible; it is reduced picture quality nerveless. Is this a reasonable price to pay? Personally I think if shot exposure can increase spatial resolution, then it is more valuable than increased integration time. I can always add more subs another day, but cannot increase resolution after it is taken. Expensive mounts required for long exposures have a price tag also.

Limitations of these calculations: thermal noise not considered; assume dim objects, sky flux >>subject flux

 

Attached Thumbnails

• 

[gregj888]

Shawn,

 

Nice... and I agree when looking at traditional imaging.

 

What's missing and I have no way yet to quantify these, are:

  1) exposures that freeze the seeing ( < 100ms typ)

  2) grading and selecting images, yes, means longer base exposure but a boost in SNR and resolution

  3) processing algorithms- starting with align and stack moving through triple correlation to what ever.  There

       is additional data in the stack of frames we can use...

 

By concentrating the signal into far fewer pixels, SNR goes up in multiples-- 2" to 0.8" would be ~6x just moving from 1.5" to 1" is a bit over 2x and that should be easy to demonstrate.    Without some heavy processing, this concept will break down as the scope gets bigger... so a lot of if, ands and buts...  Think adaptive optics- tip and tilt are align and stack, defocus (and higher orders) we throw out with lucky.  We want the math to move us to the higher orders.

 

I am not saying we are there yet either, at least in a meaningful way.  Call this the promise of CMOS... will see if it materializes.

 

So, I agree with the above, as a straight SNR from stacking problem.  I don't when talking about the future of imaging.

[Raginar]

Magic.  Weird no one is doing it. Can't wait to see the magic from the new cameras!

[gregj888]

"People" are playing with it from the sounds of things.  New tools, takes a while... and we still need some of the software pieces.

[Jon Rista]

The ASI1600's shipped this week. Should start seeing a bunch of data from them in a week tops. I think some people have already received theirs (at least the color ones, not sure if anyone has received a mono one yet.) 

 

I am more interested in seeing some direct comparisons, particularly between the ASI1600 and KAF-8300 for direct SNR vs. SNR examples. Very curious how much the lower read noise of the ASI will matter in a real-world context. It may be that under average conditions, it won't matter as much as we think due to skyfog...

[Shawnxc]

Since both cameras has similar QE, the theoretical prediction based on these formulars is that they can achieve the same quality of images with a same amount of integration time. The differences are the minimum sub-exposure durations. KAF-8300 cameras will need 3.5min subs to get 90% SNR vs ASI1600 cameras can use much shorter subs maybe down to 10-15 sec to get the same quality. one thing I'm not sure yet is the effects of pixel size. These values maybe skewed some due to this. The main benifit of low read noise cameras seems to be having the flexibility of using short subs.

[Jon Rista]

Since both cameras has similar QE, the theoretical prediction based on these formulars is that they can achieve the same quality of images with a same amount of integration time. The differences are the minimum sub-exposure durations. KAF-8300 cameras will need 3.5min subs to get 90% SNR vs ASI1600 cameras can use much shorter subs maybe down to 10-15 sec to get the same quality. one thing I'm not sure yet is the effects of pixel size. These values maybe skewed some due to this. The main benifit of low read noise cameras seems to be having the flexibility of using short subs.

 

Wouldn't you be able to achieve the same quality of images with less integration time with the ASI? Both cameras would achieve 90% SNR in different exposure times, which implies that if you stack the same number of subs, you would get the same quality. However because the subs are shorter with the ASI, you would get that quality in less total time. 

 

Note that this assumes normalization of the final results. The ASI images would be higher resolution, however if you downsampled those images to the same dimensions as the KAF, quality would improve by a factor of ~1.4x (the SQRT(16.4/8.3)). So while the pixels of the ASI are smaller, the sensor dimensions are the same. Used on the same scope, the ASI should still be gathering the same light, but it's lower read noise should be giving it the advantage, right?

[Shawnxc]

It will still need enough photons to get to certrain SNR. with sufficient long integration time, read noise contribution is negligible in the asymptote SNR calculations. As a result, cameras with similar QE will end up with similar picture quality with the same and sufficient long integration time. At least that is my understanding, could be wrong. Section 8 of Steve's article talked about this. 

[Jon Rista]

It will still need enough photons to get to certrain SNR. with sufficient long integration time, read noise contribution is negligible in the asymptote SNR calculations. As a result, cameras with similar QE will end up with similar picture quality with the same and sufficient long integration time. At least that is my understanding, could be wrong. Section 8 of Steve's article talked about this.

I reread your original post as I missed the skyfog estimate. I guess if you have 80e/min skyfog flux, then that would be swamping read noise pretty thoroughly with L imaging. What about a context where read noise is more critical? A 21.5mag/sq" or darker dark site, or more critically, narrow band imaging? With narrow band imaging, skyfog contribution would be extremely low, usually low enough with a narrower band to make read noise and eventually dark current noise the primary noise terms until you reach extremely long exposure times (45, 60, 90 minutes.)

My primary use case with the ASI1600 will be narrow band imaging, hence my curiosity.

Edited by Jon Rista, 13 May 2016 - 11:41 PM.

[Shawnxc]

Wouldn't you be able to achieve the same quality of images with less integration time with the ASI? Both cameras would achieve 90% SNR in different exposure times, which implies that if you stack the same number of subs, you would get the same quality. However because the subs are shorter with the ASI, you would get that quality in less total time. 

 

Jon, I may have miss understood your question. You are correct, with limited integration time, when photon noise is not overwhelming read noise, read noise contribution matters. Read noise contribution is squared, which made it worse. Steve's ratio method assumed long total integration time when read noise is not a factor anymore. So with limited total exposure time and everything equal, say 30min total with 3min subs, camera with less read noise will have better picture quality. How big a difference is function of sky flux. Attached is a chart showing the difference in SNR with 1 hour total integration time.

 

I am working on a set of calculations and graphs for ASI1600 based different sky conditions, takes some time. Will post them when ready.

Attached Thumbnails

• 

Edited by Shawnxc, 15 May 2016 - 01:53 AM.

[Jon Rista]

Shawn, thanks for revisiting. That new chart is showing more what I would have expected given the lower read noise. Am looking forward to seeing what else you come up with at different skyfog levels. 

 

Based on that second chart, by the time you hit 120 second subs with the ASI, you have maximized your potential, and even at 60s your getting excellent S/N. With the KAF, you are still getting improved S/N results by 420 second subs. 

Edited by Jon Rista, 15 May 2016 - 08:43 AM.

[rkayakr]

Sawn

What read noise (gain) are you assuming in the graph?

[Shawnxc]

1.75e for ASI1600 and 8.5e for 8300

[rkayakr]

Thanks

Sorta related question. If your doing short subs is cooling useful?

example - unity gain, read noise 1.75 e; sensor at 25C, dark current .4 e/s/pix

3 s exposure -> dark current 1.2 e < read noise

[Shawnxc]

Dark current can be mostly calibrated out, so it wouldn't matter. Non-cooled version is harder to match dark frame temperature; vs cooled version, dark calibration will be more precise. 

[Jon Rista]

Thanks
Sorta related question. If your doing short subs is cooling useful?
example - unity gain, read noise 1.75 e; sensor at 25C, dark current .4 e/s/pix
3 s exposure -> dark current 1.2 e < read noise

If you were really doing short subs, I think you would be using high gain mode rather than unity gain. That would mean your read noise would be 1.2e- as well. I think it would be valuable to cool in that case, so that the dark current term is fully swamped by read noise. At -15C you would have around 0.008e-/s dark current, which would amount to a 0.024e- total dark current signal. Basically, meaningless.

[Shawnxc]

Thanks

Sorta related question. If your doing short subs is cooling useful?

example - unity gain, read noise 1.75 e; sensor at 25C, dark current .4 e/s/pix

3 s exposure -> dark current 1.2 e < read noise

Even after perfect dark substraction, dark current noise still remains, which is sqrt 1.2= 1.1e, missed this part; this will go into the noise, along with read noise^2=3.1e and shot noise. Shape of the curves will change some

[Jon Rista]

Shawn, keep in mind, rkayakr was talking about the camera operating uncooled. Uncooled, at 25C, you would have 0.4e-/s dark current. However, the ASI1600 can be cooled, and when cooled properly, dark current is around 0.008e-/s or less. 

 

I received my ASI1600MM-Cool yesterday. I did some testing last night. At high gain, setting of 300, read noise was 1.13e-. Dark current at -15C was so low it couldn't even be measured. At unity gain, read noise was about 1.56e-, and again dark current couldn't even be measured. At the setting I believe I'll be using for NB imaging, 60 gain, read noise was 2.27e-. 

Edited by Jon Rista, 18 May 2016 - 10:06 AM.

[Shawnxc]

Right, cooled version is prefered. Added dark noise will always degrade the pictures or need longer exposures to compensate. Jon, you did some very nice and informative analysis of the chip. I ordered one of this camera as well. The sub-exposure calculations for this camera are coming together. Will post them soon.

[Jon Rista]

Looking forward to the next chart. I have some more analysis coming. I am trying to find the balance point, where the dynamic range of the camera equals it's bit depth (which actually does not occur at unity, BTW...you get less DR than 4096 at unity because even though it's 1e-/ADU gain, you still have read noise.) I think it will be somewhere around a gain setting of 65, which might have read noise around 2e- and an FWC around 8000e-. Anyway, when I find that point, if I find it, I'll share the numbers. I am hoping that will give me a good balance of large FWC for long narrow band exposures, and yet still offer low read noise for good low-signal performance.

[Shawnxc]

We discussed this topic on a separate thread. I thought it might be better to start a new thread here so that it will be easier for current and future ASI1600 owners (myself included) to find these information as references. So here it is. These charts are organized by telescope focal ratio and sky brightness levels, and unity gain setting in the camera (read noise~1.7e). Sky brightness used are 4.0, 4.8 and 5.5 V-mag which is equivalent to 18, 19 and 20 mag/arcsec^2. This covers suburban to fairly dark sites. Focal ratio f=4, 6.3 and 10. 

 

To use these charts, you'll look up the chart that matches your scope focal ratio and follow the curve that is close to your sky brightness level. The vertical axis is the final picture SNR after 2hr total integration time. (approximate value, object is M51 in these charts, other objects will have different S/N but will have little impact on the shape of the curves). 

 

Clear skies,

Shawn

 

ps: found an error in the f4 chart. updated

also to clarify, these are just information not "recommendation" as the title suggests, the intention is to help users to make informed selection of sub-exposure time, knowing the impact on the final picture quality.

Attached Thumbnails

• 
• 
• 

Edited by Shawnxc, 19 May 2016 - 07:56 PM.

[Shawnxc]

Here is the chart for f4 scope at unity gain. As you can see, the lines are all pretty flat within 1min, especially in light polluted skys. Only at extreme dark sites (21mag/sqrsec?) one can benefit sub exposures longer than 1min. More limiting factors are the total file size and processing time. These low read noise cameras certainly will give people more flexibility in selecting sub durations and opens the opportunity for very shot sub astro-photography.  

 

More charts in a separate thread in the beginner forum.

Attached Thumbnails

• 

[syscore]

So, we are talking pretty short exposures? That is a lot of stacking.

[Jon Rista]

Out of curiosity, could you produce a chart for gain setting 60, which gives read noise of 2.27e-? With exposures that short, you are going to need to stack hundreds of them...if not thousands.

[A. Viegas]

are these at unity gain? Would there be any impact to shift the curves to the right at very low or no gain?