I see that Ray replied while I was writing this. I'll post it anyway in case it helps someone understand this apparently slippery subject...
Quote:
Originally Posted by Peter Ward
I don't agree...as the argument falls apart with very low flux emitters, where you can't ignore shot noise.
The argument doesn't ignore shot noise. Shot noise is a function of the number of photons detected in a total integration but is indifferent to the length of the individual subs.
Quote:
Originally Posted by Peter Ward
If you are only capturing say...one photon every five seconds...then a 1 second sub will likely fail to capture anything 4 or so times out of five.
So, if you take an "average" 10 second sub it will see 2 photons. If you take 10 x 1 second subs, on average two of them will see one photon and the rest will see none. You still get two photons in the same integration time.
Quote:
Originally Posted by Peter Ward
This is assuming you have zero read noise and perfect QE....which is simply not available at present....the reality is you won't capture anything but noise.
Read noise has relevance as already discussed. QE has the same effect on long or short subs.
Quote:
Originally Posted by Peter Ward
But if we go deeper...a five second exposure will capture that photon. A ten second exposure will capture two etc.
Doesn't matter because we're integrating the subs. It's the averaged total that's important.
Quote:
Originally Posted by Peter Ward
Yes, you need exposure time, but reduced to is absurd conclusion, a million millisecond exposures is unlikely to produce anything useful.
Not with current sensors, but without read noise a million one millisecond exposures will produce the same result as one 1,000 second exposure or 100 x 10 second exposures. BTW, we're not suggesting you can ignore read noise as should be clear from the discussion above...
The maths is very clear. The problems arise when we use intuition to reason about signal and noise. Like conditional probability it seems like our brains aren't wired to grok it.
I've been thinking about this recently... In the above example, the signal is present in just over 2% of the subs. You'd be at risk of these being rejected in the stacking process by the same rejection parameters that protect you from sat trails and cosmic rays, so unless you were to go with a straight average (i.e. no rejection), there might still be a benefit in longer subs for very weak signals.
That isn't how it works Peter - it is perfectly sensible to count subs that have no target photons. The signal is the total of the photoelectrons and the noise is the combination of shot noise and (read noise*sqrt(numbersubs)). For deep narrowband, the shot noise may be almost non-existent compared to the read noise, so the SNR = totalphotoelectrons/RN*SQRT(numbersubs). This equation applies to any sub length - there is no requirement for every sub to have signal in it.
To prove the point, have a look at M&Ts wonderful deep Helix. http://www.iceinspace.com.au/forum/s...d.php?t=127240
The outermost regions are identified to be equivalent to 9 photoelectrons/pixel - detected in 18x 1 hour subs ie, at least half of the subs had no target electrons at all. But there's the signal - really beautiful and very faint shock fronts.
The 18 subs with the 16803 would have produced about 40e rms noise total. A camera with 2e read noise would produce the same 40e total read noise using 400 subs rather than 18, so you could get the same result using 400x 2.7 minute subs - and you would still only have target photons in 9 of the subs. ie, all else being equal, identical results to M&Ts could have been produced by a hypothetical low noise camera using vastly more subs. Of course, with broadband the situation is dominated by sky noise, but but the two systems would be equivalent in this environment as well.
Sorry, still don't agree.
I have taken a similarly deep image to M&T's helix (attached) ..and my early attempts with 10 minute subs were no where near as successful as my later attempts at 30+ minutes.
Read noise wasn't the problem....it was a pure lack of photons...i.e. shot noise.
Hence I don't accept it is ".. perfectly sensible to count subs that have no target photons" if that is all you are counting.
The physical reality, as opposed the the mathematical nirvana, is you need to capture photons.
If you reduce subs to the point photons are not recorded more often than not, the noise from outside the camera will swamp the signal.
Last edited by Peter Ward; 06-10-2016 at 10:03 PM.
don't think it is a problem Lee. depend on the rejection algorithm, but 1 photoelectron is not going to be anywhere near the threshold needed to reject aircraft, cosmic rays etc. In any event, in a sub, a photoelectron will be hidden in the read noise and will only be visible at all after stacking - you probably could not reliably reject such low signals even if you deliberately went out of your way to do so.
I have a similarly deep image to M&T's helix..and my early attempts with 10 minute subs were no where near as successful as my later attempts at 30+ minutes.
Read noise wasn't the problem....it was a pure lack of photons...i.e. shot noise.
Hence I don't accept it is ".. perfectly sensible to count subs that have no target photons" if that is all you are counting.
The physical reality, as opposed the the mathematical nirvana, is you need to capture photons.
If you reduce subs to the point photons are not recorded more often than not, the noise from outside the camera will swamp the signal.
No need to apologise. I doubt that we are going to agree, even with M&Ts clear (non-Nirvana) results. Your better results at 30 minutes are easily explained - your camera has relatively high read noise. You would not have detected more photons in total using 30 minute subs cf 10 minute subs, but you would have generated much less total read noise when combining the longer subs.
I cant' think of any "noise from outside the camera" - once the signal is digitised, that's it. It could be sent on a mobile phone carrier from Pluto and still arrive in perfect condition.
No need to apologise. I doubt that we are going to agree, even with M&Ts clear results. Your better results at 30 minutes are easily explained - your camera has relatively high read noise. You would not have detected more photons in total using 30 minute subs cf 10 minute subs, but you would have generated much less total read noise when combining the longer subs.
I cant' think of any "noise from outside the camera" - once the signal is digitised, that's it.
Given the fact I routinely image through a glowing atmosphere, I can assure you the sky emits flux, and it is a noise source ...not sure why you are ignoring this.
surely sky noise is detected inside the camera Peter.
In any event, the physics is pretty much identical if you add in sky noise. I am sure that M&Ts marvellous image included a little sky noise (they must have imaged through the atmosphere), but they were still able to detect a clear signal where photoelectrons were present in only about half of the subs.
What advice would you offer about cameras given my gear, sky and goals? I see it as a stepping stone to learn up and get really serious about gear if that's my post retirement plan.
I've been thinking about this recently... In the above example, the signal is present in just over 2% of the subs. You'd be at risk of these being rejected in the stacking process by the same rejection parameters that protect you from sat trails and cosmic rays, so unless you were to go with a straight average (i.e. no rejection), there might still be a benefit in longer subs for very weak signals.
Lee,
The dimmest target I've managed is the jet, R4 in NGC 1097. I don't remember the actual numbers but it was at the level where you could count the photons on your fingers in a 15 minute sub and the number of e- measured was less than the read noise. Through the magic of stacking I got a faintly visible signal and pixel rejection wasn't a problem.
surely sky noise is detected inside the camera Peter.
In any event, the physics is pretty much identical if you add in sky noise. I am sure that M&Ts marvellous image included a little sky noise (they must have imaged through the atmosphere), but they were still able to detect a clear signal where photoelectrons were present in only about half of the subs.
OK just so we are clear.
Shot noise, is a noise source from outside the camera, that is due to the random arrival of photons to individual pixels.
If you sample at too short an interval, this noise will become dominant.
Due to its physical and statistical nature, Shot noise is *reduced* by longer individual exposure times, regardless of other parameters the camera might have.
That's not clear at all Peter. Shot noise increases with longer subs - it is the square root of the signal count and the longer the sub, the higher the count and the more shot noise you get. The read noise is fixed for a sub. With low signals from short subs, the read noise dominates. eg if you get 10 photoelectrons in a sub, the shot noise will be about 3e rms. That is much less than the read noise in an SBIG camera. You need about 100 photoelectrons to get up to the read noise level of such a camera.
When you stack, you get shot noise that is the square root of the total photoelectron signal - that is independent of the sub length and it doesn't matter at all which subs the photons appear in.
It's all useful learning curve for me Ray! There aren't too many guides to help you step beyond cooled, IT modified DSLRs - so the advise of this forum is priceless!
Not with current sensors, but without read noise a million one millisecond exposures will produce the same result as one 1,000 second exposure or 100 x 10 second exposures. ............
Yes, agreed, but CCD's in reality have all sorts of noise sources, and these simply can't be ignored. e.g.
Dark current
Dark Current non uniformity
Dark current spikes
Spurious Charge
Transfer noise
Residual surface image
Residual bulk image
Cosmic rays
Cosmetic defects
Blem Spill
ADC Quantizing noise
EM interference.
Read noise is but one of many sources that derail the "mathematical ideal"
way to capture an image being espoused here.....as to work... it requires a perfect camera that simply doesn't exist.
But if you expose longer with a less than perfect camera...gosh..you get more signal.
And if you stack hundreds of deep subs, the signal gets really impressive at the expense of noise.
Rest assured I get the math.
And when I can buy a camera with Perfect QE and zero noise, I'll seriously consider shorter subs.
That's not clear at all Peter. Shot noise increases with longer subs - it is the square root of the signal count and the longer the sub, the higher the count and the more shot noise you get.
I never meant to imply shot noise did not increase with flux. I did however say at low flux levels shot noise can dominate.
But by your own math, If the signal is 1, so is the shot noise. Making the exposure nine times longer, we get signal 9, shot noise 3. QED ?
Although I already have the 1600, I'm still very curious at the technical advantages of the sensor, learning ASAP how to take advantage of its strengths.
The cam is definitely a disruptor, price wise. Just under $2k from Bintel, compared to almost $7k for a QSI683WSG? I upgraded to a cooled Astro cam 1-2 yrs sooner than I expected due to the price difference. And I fully expect to upgrade again within 2 yrs assuming CMOS development goes as well as I think. Good times for the hobby.
The cam is definitely a disruptor, price wise. Just under $2k from Bintel, compared to almost $7k for a QSI683WSG? I upgraded to a cooled Astro cam 1-2 yrs sooner than I expected due to the price difference. And I fully expect to upgrade again within 2 yrs assuming CMOS development goes as well as I think. Good times for the hobby.
In terms of the cost, you need to add FW and OAG but it is still about half the cost of 683wsg; however, QSI cameras are extremely reliable and known to last for years. If one needs to "upgrade" / replace the camera every 2 years then it may not be so economical in the long term. Oh no! I just had a vision of landfills covered with astro cameras...
In terms of the cost, you need to add FW and OAG but it is still about half the cost of 683wsg; however, QSI cameras are extremely reliable and known to last for years. If one needs to "upgrade" / replace the camera every 2 years then it may not be so economical in the long term. Oh no! I just had a vision of landfills covered with astro cameras...
Except that you're likely to recover a portion of your costs in the upgrade as there will be a 2nd hand market. And that's also good news as the entry point for those buying 2nd hand will be a lot lower than it currently is now. It will make things more accessible all round.
Now we need a similar "revolution" in the tech/price of filters and filter wheels...
In terms of the cost, you need to add FW and OAG but it is still about half the cost of 683wsg; however, QSI cameras are extremely reliable and known to last for years. If one needs to "upgrade" / replace the camera every 2 years then it may not be so economical in the long term. Oh no! I just had a vision of landfills covered with astro cameras...
There is no evidence of any quality or reliability issues with the 1600 at this point (after nearly five months), and i am using mine a lot. The only upgrade i would considering changing over for is say the rumoured APS-C chip format that ZWO 'may be working on'., or commercialisation of sCMOS astro cameras. Rather than landfill i believe what is more likely is a big reduction in resale value of some cameras.
As to QSI, a brand is no guarantee of quality, i know of one member here on IIS, that bought a QSI for over $6k from Bintel, and was never able to get it to work. I watched him on dark site trips struggling to get it to produce an image. It didn't help that he was a beginner and probably got in over his head. He eventually got the s**ts with it and took a long caravan trip. By the time he tried again the warranty was over. He found out the camera was a dud from the factory i believe.
How many times do we hear of high end ccds with bad columns, or similiar pixel issues, right out of the box; or guys that have repeatedly returned these cameras to the manufacturer for fixes or debugging.
To date i have not heard, on CN or IIS, of one ASI1600 failure. Yes Beta testers uncovered some control concerns and as a result there have been a few driver updates but mostly these were required for inter-working with particular control software, like SGP. Not unusual for a new camera and Sam has quickly resolved any reported problem.
