EOS-1D X Mark II Claims of 15 Stops of DR [CR3]

[tron]

If they can generate the ISO1600 noise levels at ISO 3200 and/or 6400, that would be a good win. When I use the 1DX, I'm never below ISO 1600 and 75%+ as ISO 3200 - 6400.

I do not believe that it will have 15 stops of DR. I will be satisfied though if it will come very close to 14.

OK, that's High ISO performance but I am interested in it too. I use my 5D3 for landscape astrophotography and I have to use it up to ISO 10000. So, I would welcome a serious improvement in high ISO too....
Also, the fact that 1DxII will be 22Mp it will make it look like a super charged 5D3!

 

[Sporgon]

I can hear it already...Nikon fanboys saying "you have to spend $6,500 to do a 5-stop push"

Unless the 5D IV comes out with similar DR.

Or the EOS-M

Not as something that would directly compete with a Nikon D810 though. I'm not terribly impressed nor excited about Canon's mirrorless offerings...they would really have to pack it full of features including a high DR sensor to even twitch the needle as far as I am concerned. Canon is much farther behind on the mirrorless front than they are on the sensor front.

The 5D IV is overdue, and if there is any camera out there people are looking forward to for having a big jump in IQ, especially in comparison to Nikon cameras, it's the 5D IV.

Have you tried the M3 Jon ? The overall 'IQ' of the 24 MP digic 6 driven sensor is superb, so much so I'm sorely temped myself, though I don't like the mirrorless configuration with no built in viewfinder.

 

[JMZawodny]

If they can generate the ISO1600 noise levels at ISO 3200 and/or 6400, that would be a good win. When I use the 1DX, I'm never below ISO 1600 and 75%+ as ISO 3200 - 6400.

I do not believe that it will have 15 stops of DR. I will be satisfied though if it will come very close to 14.

OK, that's High ISO performance but I am interested in it too. I use my 5D3 for landscape astrophotography and I have to use it up to ISO 10000. So, I would welcome a serious improvement in high ISO too....
Also, the fact that 1DxII will be 22Mp it will make it look like a super charged 5D3!

You should always shoot Astro at the ISO setting that provides the gain to digitize 1 electron as one count. Typically that ISO is near 800. Anything less and you begin to lose information for the sake of increased dynamic range. More than that and you lose dynamic range. Shooting at 10000 simply reduces the dynamic range with absolutely nothing in return. Your ISO 10000 histograms very likely look like a comb.

 

[PhotographyFirst]

If they can generate the ISO1600 noise levels at ISO 3200 and/or 6400, that would be a good win. When I use the 1DX, I'm never below ISO 1600 and 75%+ as ISO 3200 - 6400.

I do not believe that it will have 15 stops of DR. I will be satisfied though if it will come very close to 14.

OK, that's High ISO performance but I am interested in it too. I use my 5D3 for landscape astrophotography and I have to use it up to ISO 10000. So, I would welcome a serious improvement in high ISO too....
Also, the fact that 1DxII will be 22Mp it will make it look like a super charged 5D3!

You should always shoot Astro at the ISO setting that provides the gain to digitize 1 electron as one count. Typically that ISO is near 800. Anything less and you begin to lose information for the sake of increased dynamic range. More than that and you lose dynamic range. Shooting at 10000 simply reduces the dynamic range with absolutely nothing in return. Your ISO 10000 histograms very likely look like a comb.

Strange. I don't know of anyone else who shoots astro landscapes at ISO 800 unless using a star tracker and blending sky and ground exposures.

Are you saying shoot at ISO 800, f2.8, and 20-30 seconds then push the exposure in post? That the exact same as shooting at a native ISO level of 6400 or greater?

 

[JMZawodny]

If they can generate the ISO1600 noise levels at ISO 3200 and/or 6400, that would be a good win. When I use the 1DX, I'm never below ISO 1600 and 75%+ as ISO 3200 - 6400.

I do not believe that it will have 15 stops of DR. I will be satisfied though if it will come very close to 14.

OK, that's High ISO performance but I am interested in it too. I use my 5D3 for landscape astrophotography and I have to use it up to ISO 10000. So, I would welcome a serious improvement in high ISO too....
Also, the fact that 1DxII will be 22Mp it will make it look like a super charged 5D3!

You should always shoot Astro at the ISO setting that provides the gain to digitize 1 electron as one count. Typically that ISO is near 800. Anything less and you begin to lose information for the sake of increased dynamic range. More than that and you lose dynamic range. Shooting at 10000 simply reduces the dynamic range with absolutely nothing in return. Your ISO 10000 histograms very likely look like a comb.

Strange. I don't know of anyone else who shoots astro landscapes at ISO 800 unless using a star tracker and blending sky and ground exposures.

Are you saying shoot at ISO 800, f2.8, and 20-30 seconds then push the exposure in post? That the exact same as shooting at a native ISO level of 6400 or greater?

Yup, my bad! I somehow missed the "landscape" part of that. My comment referred only to pure astrophotography where the image is used as a straight linearly scaled image without any enhancements (i.e., straight RAW from the sensor). Obviously, if landscape is included then other criteria come into consideration.

 

[rfdesigner]

You should always shoot Astro at the ISO setting that provides the gain to digitize 1 electron as one count. Typically that ISO is near 800. Anything less and you begin to lose information for the sake of increased dynamic range. More than that and you lose dynamic range. Shooting at 10000 simply reduces the dynamic range with absolutely nothing in return. Your ISO 10000 histograms very likely look like a comb.

If I may I'd like to disagree..

I've played astrophotography for a long while, built my own drive system, built a CCD camera, currently use a 383L+ and am in a long winded process of building an observatory.

You do not need to aim for 1ADU/e.. because the readout noise exceeds 1e. Most DSLRs readout noise is 2.5e or above... so the need to ensure you don't miss any steps is removed by the spreading effect of the noise.

You do need to aim for the lowest ISO where the readout noise is more or less unchanged from very high ISO.. this usually works out as 400~1600 on canon cameras, so in use terms we agree, but the reasoning is different. Nikons can go as low as ISO200 in this regard.. but Nikon has history of meddling with their RAW files, either actively hunting out hot pixels so deleating stars, or clipping blacks with the D800(e?)

For reference my dedicated astro CCD has a conversion factor of about 2.2ADU/e

I only bring this up to prevent a misunderstanding from propogating too far and wide.

 

[Lee Jay]

I thought the choice of ISO for astro subs was usually set by the longest tracking time your mount can sustain and/or the sky glow of your site.

http://www.samirkharusi.net/sub-exposures.html

"As long as we expose long enough for the entire skyfog mountain to be entirely detached from the camera's Read Noise we can be reasonably confident that stacking N exposures each T-minutes long will be closely equivalent to shooting one very long exposure that is NT-minutes long, yielding similar SNRs in the final processed images. "

http://www.cloudynights.com/topic/351257-stacking-efficiency-sub-length-skyfog-histogram/

 

[jrista]

You should always shoot Astro at the ISO setting that provides the gain to digitize 1 electron as one count. Typically that ISO is near 800. Anything less and you begin to lose information for the sake of increased dynamic range. More than that and you lose dynamic range. Shooting at 10000 simply reduces the dynamic range with absolutely nothing in return. Your ISO 10000 histograms very likely look like a comb.

If I may I'd like to disagree..

I've played astrophotography for a long while, built my own drive system, built a CCD camera, currently use a 383L+ and am in a long winded process of building an observatory.

You do not need to aim for 1ADU/e.. because the readout noise exceeds 1e. Most DSLRs readout noise is 2.5e or above... so the need to ensure you don't miss any steps is removed by the spreading effect of the noise.

You do need to aim for the lowest ISO where the readout noise is more or less unchanged from very high ISO.. this usually works out as 400~1600 on canon cameras, so in use terms we agree, but the reasoning is different. Nikons can go as low as ISO200 in this regard.. but Nikon has history of meddling with their RAW files, either actively hunting out hot pixels so deleating stars, or clipping blacks with the D800(e?)

For reference my dedicated astro CCD has a conversion factor of about 2.2ADU/e

I only bring this up to prevent a misunderstanding from propogating too far and wide.

I agree with this!

Unit gain is more a myth than anything. Most of the time we don't know enough about what the actual unity gain is for it to matter, and furthermore, we usually cannot actually use unity gain. Additionally, while sampling every electron as finely as you can tends to be better, it is still possible to image at lower ISO/gain settings where multiple electrons are required per ADU (i.e. ISO 400 or lower on my 5D III, which I have imaged at on many occasions). The read noise or quantization error are just too small to worry about in most cases.

The actual quantization noise that you get in any pixel is also actually quite small. It's around SQRT(RN^2 + (RN-SQRT(RN^2 + QN^2))^2). Quantization noise is +/-1 bit, pretty much regardless, so if you have 3e- RN (common for most DSLRs) you would have ~3.004e- noise with the quantization error (and we can only really measure it if we actually have proper low-level specs on the sensor, including input referred and output referred noise at the ADC). The quantization error is so small as to be meaningless in the grand scheme of things. Read noise itself is usually too small to matter in most cases...the exception being LRGB imaging at an EXCEPTIONALLY dark site, or narrow band imaging with a very narrow band pass (5nm or smaller). In both cases, getting significant background skyfog levels can be difficult, so you may not be able to sufficiently swamp read noise with background sky level.

FAR more important sources of noise in astrophotography is the noise from dark current, and the noise from light pollution. Both tend to DWARF any other sources of noise with a DSLR for sure, and often with a CCD (light pollution is usually dominant with CCD, as dark current is minimized by thermal regulation most of the time). My 5D III can have as much as 5e-/s/px dark current during the summer...which actually makes it the single most significant source of noise in the image, period.

 

[jrista]

I thought the choice of ISO for astro subs was usually set by the longest tracking time your mount can sustain and/or the sky glow of your site.

http://www.samirkharusi.net/sub-exposures.html

"As long as we expose long enough for the entire skyfog mountain to be entirely detached from the camera's Read Noise we can be reasonably confident that stacking N exposures each T-minutes long will be closely equivalent to shooting one very long exposure that is NT-minutes long, yielding similar SNRs in the final processed images. "

http://www.cloudynights.com/topic/351257-stacking-efficiency-sub-length-skyfog-histogram/

I have found it doesn't much matter in the end. I have imaged at ISO's from 400 through 3200 on my 5D III. While there is higher read noise at ISO 400, you are also able to gather twice the signal strength as at a higher ISO. Signal grows faster than noise, so 9.8e- RN @ ISO 400 with appropriately long subs (say 15-20 minutes) vs. 5.6e- RN @ ISO 800 with about half the sub length (7-10 minutes) vs. 3.6e- RN @ ISO 1600 with again half the sub length (3-5 minutes) usually means you have a higher SNR with the longer subs at lower ISO:

200/SQRT(200 + 9.8^2) = 11.62:1
100/SQRT(100 + 5.6^2) = 8.75:1
50/SQRT(50 + 3.6^2) = 6.3:1

If you throw in dark current, that normalizes things a bit, because read noise becomes a secondary noise factor. Throw in light pollution, and things tend to flatten out even more, as read noise becomes a distant tertiary noise factor. However I have never actually encountered a situation where imaging with longer subs at a lower ISO gave me worse results than imaging with shorter subs at a higher ISO. Half a dozen of one, six of the other. It tends to break even in the end. On a PER-SUB basis, anyway.

The real caveat with using shorter subs is you need to stack more of them. Then read noise compounding becomes a problem...you get a constant amount of read noise in every frame, and since you need to stack MANY more ISO 1600 subs than ISO 400 subs, you end up with more total read noise in the end. However, it still usually doesn't matter if you are dark current and/or light pollution limited, as the noise from both of those is still significantly higher.

 

[JMZawodny]

I thought the choice of ISO for astro subs was usually set by the longest tracking time your mount can sustain and/or the sky glow of your site.

http://www.samirkharusi.net/sub-exposures.html

"As long as we expose long enough for the entire skyfog mountain to be entirely detached from the camera's Read Noise we can be reasonably confident that stacking N exposures each T-minutes long will be closely equivalent to shooting one very long exposure that is NT-minutes long, yielding similar SNRs in the final processed images. "

http://www.cloudynights.com/topic/351257-stacking-efficiency-sub-length-skyfog-histogram/

I have found it doesn't much matter in the end. I have imaged at ISO's from 400 through 3200 on my 5D III. While there is higher read noise at ISO 400, you are also able to gather twice the signal strength as at a higher ISO. Signal grows faster than noise, so 9.8e- RN @ ISO 400 with appropriately long subs (say 15-20 minutes) vs. 5.6e- RN @ ISO 800 with about half the sub length (7-10 minutes) vs. 3.6e- RN @ ISO 1600 with again half the sub length (3-5 minutes) usually means you have a higher SNR with the longer subs at lower ISO:

200/SQRT(200 + 9.8^2) = 11.62:1
100/SQRT(100 + 5.6^2) = 8.75:1
50/SQRT(50 + 3.6^2) = 6.3:1

If you throw in dark current, that normalizes things a bit, because read noise becomes a secondary noise factor. Throw in light pollution, and things tend to flatten out even more, as read noise becomes a distant tertiary noise factor. However I have never actually encountered a situation where imaging with longer subs at a lower ISO gave me worse results than imaging with shorter subs at a higher ISO. Half a dozen of one, six of the other. It tends to break even in the end. On a PER-SUB basis, anyway.

The real caveat with using shorter subs is you need to stack more of them. Then read noise compounding becomes a problem...you get a constant amount of read noise in every frame, and since you need to stack MANY more ISO 1600 subs than ISO 400 subs, you end up with more total read noise in the end. However, it still usually doesn't matter if you are dark current and/or light pollution limited, as the noise from both of those is still significantly higher.

I guess we'll agree to disagree on some points. What ever works for you.

I'm a bit surprised by your variation in read noise vs ISO. Neither my 5D or 5D2 behave like that. Makes me wonder whether Canon did something different with the 5D3.

I gave up trying to use a DSLR for astro as the dark current and spectral filtering were severe issues. Given my site, I mostly use a cooled camera (KAF-8300 chip) and narrow band filters now.

 

[Don Haines]

I thought the choice of ISO for astro subs was usually set by the longest tracking time your mount can sustain and/or the sky glow of your site.

http://www.samirkharusi.net/sub-exposures.html

"As long as we expose long enough for the entire skyfog mountain to be entirely detached from the camera's Read Noise we can be reasonably confident that stacking N exposures each T-minutes long will be closely equivalent to shooting one very long exposure that is NT-minutes long, yielding similar SNRs in the final processed images. "

http://www.cloudynights.com/topic/351257-stacking-efficiency-sub-length-skyfog-histogram/

I have found it doesn't much matter in the end. I have imaged at ISO's from 400 through 3200 on my 5D III. While there is higher read noise at ISO 400, you are also able to gather twice the signal strength as at a higher ISO. Signal grows faster than noise, so 9.8e- RN @ ISO 400 with appropriately long subs (say 15-20 minutes) vs. 5.6e- RN @ ISO 800 with about half the sub length (7-10 minutes) vs. 3.6e- RN @ ISO 1600 with again half the sub length (3-5 minutes) usually means you have a higher SNR with the longer subs at lower ISO:

200/SQRT(200 + 9.8^2) = 11.62:1
100/SQRT(100 + 5.6^2) = 8.75:1
50/SQRT(50 + 3.6^2) = 6.3:1

If you throw in dark current, that normalizes things a bit, because read noise becomes a secondary noise factor. Throw in light pollution, and things tend to flatten out even more, as read noise becomes a distant tertiary noise factor. However I have never actually encountered a situation where imaging with longer subs at a lower ISO gave me worse results than imaging with shorter subs at a higher ISO. Half a dozen of one, six of the other. It tends to break even in the end. On a PER-SUB basis, anyway.

The real caveat with using shorter subs is you need to stack more of them. Then read noise compounding becomes a problem...you get a constant amount of read noise in every frame, and since you need to stack MANY more ISO 1600 subs than ISO 400 subs, you end up with more total read noise in the end. However, it still usually doesn't matter if you are dark current and/or light pollution limited, as the noise from both of those is still significantly higher.

I guess we'll agree to disagree on some points. What ever works for you.

I'm a bit surprised by your variation in read noise vs ISO. Neither my 5D or 5D2 behave like that. Makes me wonder whether Canon did something different with the 5D3.

I gave up trying to use a DSLR for astro as the dark current and spectral filtering were severe issues. Given my site, I mostly use a cooled camera (KAF-8300 chip) and narrow band filters now.

I am nowhere near jrista, but when I moved from my 60D to a 7D2, I found a sighificant jump in the quality of my long night exposures.... The 60D and the 5D2 are similar age of technologies, and the 7D2 and 5D3 are of similar age, so a jump in quality of Astro imaging from the 5D2 to the 5D3 would seem likely to me.

 

[jrista]

I thought the choice of ISO for astro subs was usually set by the longest tracking time your mount can sustain and/or the sky glow of your site.

http://www.samirkharusi.net/sub-exposures.html

"As long as we expose long enough for the entire skyfog mountain to be entirely detached from the camera's Read Noise we can be reasonably confident that stacking N exposures each T-minutes long will be closely equivalent to shooting one very long exposure that is NT-minutes long, yielding similar SNRs in the final processed images. "

http://www.cloudynights.com/topic/351257-stacking-efficiency-sub-length-skyfog-histogram/

I have found it doesn't much matter in the end. I have imaged at ISO's from 400 through 3200 on my 5D III. While there is higher read noise at ISO 400, you are also able to gather twice the signal strength as at a higher ISO. Signal grows faster than noise, so 9.8e- RN @ ISO 400 with appropriately long subs (say 15-20 minutes) vs. 5.6e- RN @ ISO 800 with about half the sub length (7-10 minutes) vs. 3.6e- RN @ ISO 1600 with again half the sub length (3-5 minutes) usually means you have a higher SNR with the longer subs at lower ISO:

200/SQRT(200 + 9.8^2) = 11.62:1
100/SQRT(100 + 5.6^2) = 8.75:1
50/SQRT(50 + 3.6^2) = 6.3:1

If you throw in dark current, that normalizes things a bit, because read noise becomes a secondary noise factor. Throw in light pollution, and things tend to flatten out even more, as read noise becomes a distant tertiary noise factor. However I have never actually encountered a situation where imaging with longer subs at a lower ISO gave me worse results than imaging with shorter subs at a higher ISO. Half a dozen of one, six of the other. It tends to break even in the end. On a PER-SUB basis, anyway.

The real caveat with using shorter subs is you need to stack more of them. Then read noise compounding becomes a problem...you get a constant amount of read noise in every frame, and since you need to stack MANY more ISO 1600 subs than ISO 400 subs, you end up with more total read noise in the end. However, it still usually doesn't matter if you are dark current and/or light pollution limited, as the noise from both of those is still significantly higher.

I guess we'll agree to disagree on some points. What ever works for you.

I'm a bit surprised by your variation in read noise vs ISO. Neither my 5D or 5D2 behave like that. Makes me wonder whether Canon did something different with the 5D3.

I gave up trying to use a DSLR for astro as the dark current and spectral filtering were severe issues. Given my site, I mostly use a cooled camera (KAF-8300 chip) and narrow band filters now.

The 5D and 5D II have insane levels of dark current. I complain about my 5D III's ~5e-/s/px dark current @ 25C, because it can be a real problem (a 300 second sub @ 5e-/s DC has 1500e- dark current (which is suppressed itself by CDS), which leaves behind almost 39e- dark current noise!) A 5D II has several times as much dark current. The majority of that horrid red color noise in a 5D II at higher ISO is primarily from dark current (it's high enough to add several electrons worth of noise in and of itself (even at sub-second exposure times), which at high ISO could be higher than read noise).

So I would venture to guess that your primary problem with the 5D and 5D II was not necessarily read noise (IIRC, the 5D II actually had LESS read noise, not more). The problem was the ridiculous levels of dark current. It's still fairly ridiculous with my 5D III. It got a bit better with the 6D, which dropped to around 2.5-3e-/s @ 25C, and scaled better as you went cooler thanks to a lower doubling temp. The 7D II is the first camera Canon produced that actually had truly competitive dark current levels relative to other brand sensors. It's around 0.2-0.3e-/s @ 25C, and also has a smaller doubling temp.

In my experiences over the last year, particularly summer and fall this year, I have learned that dark current is often the single most significant noise term when doing astrophotography with a DSLR. During winter, depending on where you are, that can change. My recent images have had sensor temps around -2C to -5C, so dark current doesn't matter (it's less than read noise with my average sub lengths.) But even during winter, it's rarer that I get to sub-zero (C) temps at the sensor, and usually it's around 8-10C or higher, and most of the year it's 20C or higher.

 

[JMZawodny]

I thought the choice of ISO for astro subs was usually set by the longest tracking time your mount can sustain and/or the sky glow of your site.

http://www.samirkharusi.net/sub-exposures.html

"As long as we expose long enough for the entire skyfog mountain to be entirely detached from the camera's Read Noise we can be reasonably confident that stacking N exposures each T-minutes long will be closely equivalent to shooting one very long exposure that is NT-minutes long, yielding similar SNRs in the final processed images. "

http://www.cloudynights.com/topic/351257-stacking-efficiency-sub-length-skyfog-histogram/

I have found it doesn't much matter in the end. I have imaged at ISO's from 400 through 3200 on my 5D III. While there is higher read noise at ISO 400, you are also able to gather twice the signal strength as at a higher ISO. Signal grows faster than noise, so 9.8e- RN @ ISO 400 with appropriately long subs (say 15-20 minutes) vs. 5.6e- RN @ ISO 800 with about half the sub length (7-10 minutes) vs. 3.6e- RN @ ISO 1600 with again half the sub length (3-5 minutes) usually means you have a higher SNR with the longer subs at lower ISO:

200/SQRT(200 + 9.8^2) = 11.62:1
100/SQRT(100 + 5.6^2) = 8.75:1
50/SQRT(50 + 3.6^2) = 6.3:1

If you throw in dark current, that normalizes things a bit, because read noise becomes a secondary noise factor. Throw in light pollution, and things tend to flatten out even more, as read noise becomes a distant tertiary noise factor. However I have never actually encountered a situation where imaging with longer subs at a lower ISO gave me worse results than imaging with shorter subs at a higher ISO. Half a dozen of one, six of the other. It tends to break even in the end. On a PER-SUB basis, anyway.

The real caveat with using shorter subs is you need to stack more of them. Then read noise compounding becomes a problem...you get a constant amount of read noise in every frame, and since you need to stack MANY more ISO 1600 subs than ISO 400 subs, you end up with more total read noise in the end. However, it still usually doesn't matter if you are dark current and/or light pollution limited, as the noise from both of those is still significantly higher.

I guess we'll agree to disagree on some points. What ever works for you.

I'm a bit surprised by your variation in read noise vs ISO. Neither my 5D or 5D2 behave like that. Makes me wonder whether Canon did something different with the 5D3.

I gave up trying to use a DSLR for astro as the dark current and spectral filtering were severe issues. Given my site, I mostly use a cooled camera (KAF-8300 chip) and narrow band filters now.

The 5D and 5D II have insane levels of dark current. I complain about my 5D III's ~5e-/s/px dark current @ 25C, because it can be a real problem (a 300 second sub @ 5e-/s DC has 1500e- dark current (which is suppressed itself by CDS), which leaves behind almost 39e- dark current noise!) A 5D II has several times as much dark current. The majority of that horrid red color noise in a 5D II at higher ISO is primarily from dark current (it's high enough to add several electrons worth of noise in and of itself (even at sub-second exposure times), which at high ISO could be higher than read noise).

So I would venture to guess that your primary problem with the 5D and 5D II was not necessarily read noise (IIRC, the 5D II actually had LESS read noise, not more). The problem was the ridiculous levels of dark current. It's still fairly ridiculous with my 5D III. It got a bit better with the 6D, which dropped to around 2.5-3e-/s @ 25C, and scaled better as you went cooler thanks to a lower doubling temp. The 7D II is the first camera Canon produced that actually had truly competitive dark current levels relative to other brand sensors. It's around 0.2-0.3e-/s @ 25C, and also has a smaller doubling temp.

In my experiences over the last year, particularly summer and fall this year, I have learned that dark current is often the single most significant noise term when doing astrophotography with a DSLR. During winter, depending on where you are, that can change. My recent images have had sensor temps around -2C to -5C, so dark current doesn't matter (it's less than read noise with my average sub lengths.) But even during winter, it's rarer that I get to sub-zero (C) temps at the sensor, and usually it's around 8-10C or higher, and most of the year it's 20C or higher.

Yes, I was never worried about read noise with the 5D or 5D2, but it was very constant vs ISO except at the high and low extremes of the ISO range. With the KAF-8300 cooled 30C below ambient, I'm no longer worried by dark current either. Now I wish that I had a darker site.

 

[jrista]

If they can generate the ISO1600 noise levels at ISO 3200 and/or 6400, that would be a good win. When I use the 1DX, I'm never below ISO 1600 and 75%+ as ISO 3200 - 6400.

I do not believe that it will have 15 stops of DR. I will be satisfied though if it will come very close to 14.

OK, that's High ISO performance but I am interested in it too. I use my 5D3 for landscape astrophotography and I have to use it up to ISO 10000. So, I would welcome a serious improvement in high ISO too....
Also, the fact that 1DxII will be 22Mp it will make it look like a super charged 5D3!

You should always shoot Astro at the ISO setting that provides the gain to digitize 1 electron as one count. Typically that ISO is near 800. Anything less and you begin to lose information for the sake of increased dynamic range. More than that and you lose dynamic range. Shooting at 10000 simply reduces the dynamic range with absolutely nothing in return. Your ISO 10000 histograms very likely look like a comb.

To address this more directly. There is always noise in the conversion. It actually is not possible to always convert 1 electron to 1 ADU because noise will usually mean you get 0 ADU, or >1 ADU.

Technically speaking, it is better to sample every electron as finely as possible. That means you want a gain (in terms of e-/ADU) of LESS than 1. Nyquist would dictate a gain of 0.5e-/ADU, however that only really works for something like an audio signal, however because of the various noise terms that play a role with digital imaging (including read noise and PRNU) and because we usually stack, a gain of 0.3e-/ADU is better, and higher is certainly not bad, especially with very faint signals. There is certainly a balancing act here...too high of an ISO and you will throw away too much dynamic range...and in the case of a Canon DSLR, too low of an ISO and read noise will increase to ridiculous levels (anything over 10e- and you have to start wondering why your bothering...the 5D III has ~35e- RN @ ISO 100...square that, you add 1225e- noise in your noise term when calculating SNR!!!)

There is a caveat here. Outside of the >10e- read noise scenario...read noise usually barely matters! This is because for most imagers, there are other sources of noise that are more significant. Dark current is one, which can add as much as 1500e- signal or around there, which is 30e- noise. Another significant source of noise is light pollution, which can add anywhere from 200 to 2000e- additional signal outside of a nice dark site (yellow through white zones on the bortle scale). With minimal dark current and no light pollution, an SNR calculation might look like this (assuming 50e- signal from space in say a 120 second exposure ):

50e-/SQRT(50e- + 10e-^2) = 4.1:1

At high ISO, where your better sampling each and every electron, your read noise (in relative terms) is smaller:

50e-/SQRT(50e- + 3e-^2) = 6.5:1

An improvement of almost 60%. Quite significant. However it's never that simple. Add in dark current at say 10e-/s/px @ 20C (i.e. 5D II):

50e-/SQRT(50e- + (10*120) + 3e-^2) = 50e-/SQRT(50e- + 1200e- + 9e-) = 1.41:1

That dark current totally decimated our SNR. Increasing read noise to 10e- really doesn't matter much at that point:

50e-/SQRT(50e- + (10*120) + 10e-^2) = 50e-/SQRT(50e- + 1200e- + 100e-) = 1.36:1

Less than a 4% difference here. Read noise matters even less when we factor in light pollution (say deep in a red zone, suburbia central):

50e-/SQRT(50e- + 1500e- + (10e-/s*120s) + 10e-^2) = 50e-/SQRT(50e- + 1500e- + 1200e- + 100e-) = 0.94:1

Even if you had a mere 1e- read noise, it wouldn't matter:

50e-/SQRT(50e- + 1500e- + (10e-/s*120s) + 1e-^2) = 50e-/SQRT(50e- + 1500e- + 1200e- + 1e-) = 0.95:1

It really doesn't matter if you image at ISO 400, 800, or 1600...in the end, the amount of read noise barely affects the results. Dark current and light pollution dominate by such a significant margin. That changes if you can regulate your sensor temp, and find dark skies, though:

50e-/SQRT(50e- + 30e- + (0.02e-/s*120s) + 3e-^2) = 5.23:1

Bump read noise up to 10e-:

50e-/SQRT(50e- + 30e- + (0.02e-/s*120s) + 10e-^2) = 3.7:1

Read noise becomes a more significant factor when your imaging under pristine dark skies (i.e. 21.5mag/sq" or better, deep blue zone, gray zone, black zone) with very low dark current. Dark current itself could easily be the single most devastating noise term you may have to deal with (i.e. dark sky...in the summer):

50e-/SQRT(50e- + 30e- + (5e-/s*120s) + 3e-^2) = 1.9:1
50e-/SQRT(50e- + 30e- + (10e-/s*120s) + 3e-^2) = 1.4:1

 

[JMZawodny]

To address this more directly. There is always noise in the conversion. It actually is not possible to always convert 1 electron to 1 ADU because noise will usually mean you get 0 ADU, or >1 ADU.

That would depend upon the what the gain of the system is as well as the nature of the read noise. I must note that a proper signal chain must have a noise level of at least 0.5 ADU (assuming you plan to stack). You say as much when you cited Nyquist. You must agree that digitization is the final step in the process. These factors are also why placing the ADC chain on chip improves performance so much. You can control the nature of the noise.

Technically speaking, it is better to sample every electron as finely as possible.

I'm sorry, but in my quantum world measurements of electrons always yield an integer result. Sampling them more finely only serves to reduce the DR of the system.

Read noise becomes a more significant factor when your imaging under pristine dark skies (i.e. 21.5mag/sq" or better, deep blue zone, gray zone, black zone) with very low dark current. Dark current itself could easily be the single most devastating noise term you may have to deal with (i.e. dark sky...in the summer):

Naturally.

 

[jrista]

To address this more directly. There is always noise in the conversion. It actually is not possible to always convert 1 electron to 1 ADU because noise will usually mean you get 0 ADU, or >1 ADU.

That would depend upon the what the gain of the system is as well as the nature of the read noise. I must note that a proper signal chain must have a noise level of at least 0.5 ADU (assuming you plan to stack). You say as much when you cited Nyquist. You must agree that digitization is the final step in the process. These factors are also why placing the ADC chain on chip improves performance so much. You can control the nature of the noise.

I was speaking explicitly for unity gain, which is by definition 1e- to 1 ADU. I was trying to point out that you aren't going to always get 1ADU because of noise, including PRNU.

Technically speaking, it is better to sample every electron as finely as possible.

I'm sorry, but in my quantum world measurements of electrons always yield an integer result. Sampling them more finely only serves to reduce the DR of the system.

Your misunderstanding. At a gain of 0.3e-/ADU, one single electron would result in 3 to 4 ADU. That would be a more finely sampled electron VS unity gain. At a gain of 0.15e-/ADU, one single electron would result in 6-7 ADU. That would be even more finely sampled. The quantization error becomes smaller with the finer sampling as well, as +/-1 ADU the discrepancy is only 0.15e- rather than 0.3e-.

Whether you use all the DR depends on what your imaging, and how much clipped stars matter to the end result. I tend to always clip my brighter stars (and I usually image around ISO 800-1600, which is sampling each electron well, but not necessarily as ideally as possible), and my results are usually excellent, and very deep:

I could expose less, and avoid clipping any stars at all...but then I'm getting SIGNIFICANTLY less faint detail in each sub. Stars saturate at a much higher rate than nebula and dust, so when you reduce exposure enough to prevent star clipping, you've usually stuffed all the faint details down well into the noise floor. That can greatly increase the number of subs required to fully reveal all those faint details.

If push comes to shove, you can always do an HDR blend to bring out immense dynamic range. This Orion Sword image has around 18 stops, and was the result if a very meticulous linear fitting and HDR blending process in PixInsight:

This was sampled at 0.36e-/ADU, which was probably closer to ideal, resulting in great outer dust SNR, but requiring several sets of shorter subs to add in the ultra bright details around the Trap.

 

[JMZawodny]

To address this more directly. There is always noise in the conversion. It actually is not possible to always convert 1 electron to 1 ADU because noise will usually mean you get 0 ADU, or >1 ADU.

That would depend upon the what the gain of the system is as well as the nature of the read noise. I must note that a proper signal chain must have a noise level of at least 0.5 ADU (assuming you plan to stack). You say as much when you cited Nyquist. You must agree that digitization is the final step in the process. These factors are also why placing the ADC chain on chip improves performance so much. You can control the nature of the noise.

I was speaking explicitly for unity gain, which is by definition 1e- to 1 ADU. I was trying to point out that you aren't going to always get 1ADU because of noise, including PRNU.

Technically speaking, it is better to sample every electron as finely as possible.

I'm sorry, but in my quantum world measurements of electrons always yield an integer result. Sampling them more finely only serves to reduce the DR of the system.

Your misunderstanding. At a gain of 0.3e-/ADU, one single electron would result in 3 to 4 ADU. That would be a more finely sampled electron VS unity gain. At a gain of 0.15e-/ADU, one single electron would result in 6-7 ADU. That would be even more finely sampled. The quantization error becomes smaller with the finer sampling as well, as +/-1 ADU the discrepancy is only 0.15e- rather than 0.3e-.

Whether you use all the DR depends on what your imaging, and how much clipped stars matter to the end result. I tend to always clip my brighter stars (and I usually image around ISO 800-1600, which is sampling each electron well, but not necessarily as ideally as possible), and my results are usually excellent, and very deep:

Yes, most impressive results to be sure. Better than what I was able to do during my Pre-CCD days at the university in Boulder back in the 80's. Yet what I am saying is that you are being most conservative in your approach to manipulating noise.

 

[jrista]

To address this more directly. There is always noise in the conversion. It actually is not possible to always convert 1 electron to 1 ADU because noise will usually mean you get 0 ADU, or >1 ADU.

That would depend upon the what the gain of the system is as well as the nature of the read noise. I must note that a proper signal chain must have a noise level of at least 0.5 ADU (assuming you plan to stack). You say as much when you cited Nyquist. You must agree that digitization is the final step in the process. These factors are also why placing the ADC chain on chip improves performance so much. You can control the nature of the noise.

I was speaking explicitly for unity gain, which is by definition 1e- to 1 ADU. I was trying to point out that you aren't going to always get 1ADU because of noise, including PRNU.

Technically speaking, it is better to sample every electron as finely as possible.

I'm sorry, but in my quantum world measurements of electrons always yield an integer result. Sampling them more finely only serves to reduce the DR of the system.

Your misunderstanding. At a gain of 0.3e-/ADU, one single electron would result in 3 to 4 ADU. That would be a more finely sampled electron VS unity gain. At a gain of 0.15e-/ADU, one single electron would result in 6-7 ADU. That would be even more finely sampled. The quantization error becomes smaller with the finer sampling as well, as +/-1 ADU the discrepancy is only 0.15e- rather than 0.3e-.

Whether you use all the DR depends on what your imaging, and how much clipped stars matter to the end result. I tend to always clip my brighter stars (and I usually image around ISO 800-1600, which is sampling each electron well, but not necessarily as ideally as possible), and my results are usually excellent, and very deep:

Yes, most impressive results to be sure. Better than what I was able to do during my Pre-CCD days at the university in Boulder back in the 80's. Yet what I am saying is that you are being most conservative in your approach to manipulating noise.

Manipulating noise at the level you are talking about only matters if you are truly limited by the noise of the camera, though. I am FAR from being that limited, even though I use a dark site. If I was still imaging under 21.6mag/sq" skies, then it MIGHT be a different story...maybe...if airglow was at a minimum and I had no inversion layer, and I was getting 15 minute color subs. At 21mag/sq" skies, I'm working under skies almost 1.8 stops brighter, and am totally skyfog limited (I have more signal from LP than from object, so it doesn't take all that much to make SQRT(SkyFog) my most dominant source of noise.

If I was doing narrow band imaging with 3nm filters on ultra faint objects, say OU4 in OIII, then I would be more concerned about read noise, FPN, PRNU, quantization error, etc. I still wouldn't be all that concerned about any of them except maybe read noise if I was doing 3-5nm Ha imaging, as again it's not terribly difficult to get extremely high contrast subs with Ha.

Very few astrophotographers need to be that concerned about noise at this level, because it is such a minuscule amount of noise in the grand scheme of things. Light pollution at the very least will be the most significant source of noise for any DSLR or mirrorless astrophotographer, followed by dark current noise. The two together are at least a factor of ten more significant than read noise, quantization error, etc.

If you do a lot of narrow band imaging with that Atik CCD camera, then sure, you should pay attention to the read noise. Either that, or just get some seriously long sub exposures, 60-90 minutes or so, and actually get your subs skyfog limited.

 

[JMZawodny]

Manipulating noise at the level you are talking about only matters if you are truly limited by the noise of the camera, though. I am FAR from being that limited, even though I use a dark site. If I was still imaging under 21.6mag/sq" skies, then it MIGHT be a different story...maybe...if airglow was at a minimum and I had no inversion layer, and I was getting 15 minute color subs. At 21mag/sq" skies, I'm working under skies almost 1.8 stops brighter, and am totally skyfog limited (I have more signal from LP than from object, so it doesn't take all that much to make SQRT(SkyFog) my most dominant source of noise.

If I was doing narrow band imaging with 3nm filters on ultra faint objects, say OU4 in OIII, then I would be more concerned about read noise, FPN, PRNU, quantization error, etc. I still wouldn't be all that concerned about any of them except maybe read noise if I was doing 3-5nm Ha imaging, as again it's not terribly difficult to get extremely high contrast subs with Ha.

Very few astrophotographers need to be that concerned about noise at this level, because it is such a minuscule amount of noise in the grand scheme of things. Light pollution at the very least will be the most significant source of noise for any DSLR or mirrorless astrophotographer, followed by dark current noise. The two together are at least a factor of ten more significant than read noise, quantization error, etc.

If you do a lot of narrow band imaging with that Atik CCD camera, then sure, you should pay attention to the read noise. Either that, or just get some seriously long sub exposures, 60-90 minutes or so, and actually get your subs skyfog limited.

In my day job, I guess I have to quantify the detailed nature of noise more than most do (google "zawodny sage" if you have not already). As a result, I have a refined expectation for the performance of measurement HW and the justified expectations for a proposed approach to making a measurement. All I can say is that you are being rather conservative in your approach to making the measurement. You are certainly most practical in your methodology, but you may not be pushing the state of the art.

Let me be quite clear here. Jon, I find your technique both in data capture and post processing to be first rate. Your imagery is masterful and very pleasing to the eye. I must ask, however, what will it take to be the master 5 years from now? Do not be afraid of digitization/quantization (sp?) noise.

 

[jrista]

Manipulating noise at the level you are talking about only matters if you are truly limited by the noise of the camera, though. I am FAR from being that limited, even though I use a dark site. If I was still imaging under 21.6mag/sq" skies, then it MIGHT be a different story...maybe...if airglow was at a minimum and I had no inversion layer, and I was getting 15 minute color subs. At 21mag/sq" skies, I'm working under skies almost 1.8 stops brighter, and am totally skyfog limited (I have more signal from LP than from object, so it doesn't take all that much to make SQRT(SkyFog) my most dominant source of noise.

If I was doing narrow band imaging with 3nm filters on ultra faint objects, say OU4 in OIII, then I would be more concerned about read noise, FPN, PRNU, quantization error, etc. I still wouldn't be all that concerned about any of them except maybe read noise if I was doing 3-5nm Ha imaging, as again it's not terribly difficult to get extremely high contrast subs with Ha.

Very few astrophotographers need to be that concerned about noise at this level, because it is such a minuscule amount of noise in the grand scheme of things. Light pollution at the very least will be the most significant source of noise for any DSLR or mirrorless astrophotographer, followed by dark current noise. The two together are at least a factor of ten more significant than read noise, quantization error, etc.

If you do a lot of narrow band imaging with that Atik CCD camera, then sure, you should pay attention to the read noise. Either that, or just get some seriously long sub exposures, 60-90 minutes or so, and actually get your subs skyfog limited.

In my day job, I guess I have to quantify the detailed nature of noise more than most do (google "zawodny sage" if you have not already). As a result, I have a refined expectation for the performance of measurement HW and the justified expectations for a proposed approach to making a measurement. All I can say is that you are being rather conservative in your approach to making the measurement. You are certainly most practical in your methodology, but you may not be pushing the state of the art.

Oh, I'm NOT pushing the state of the art. I have no illusions there. But I don't really need to, either...not with the LP and dark current I have these days. I also have no good specs, as with a CCD, for my DSLRs...I don't know what the actual input referred read noise is, what the actual PRNU the pixels is, etc. so I couldn't really be all that accurate about perfectly optimizing my exposure times even if I wanted to. Things change enough throughout a night at my dark site that it would be difficult to get a truly optimal exposure time and stick with it as well...post-midnight, my dark site can improve by 0.2-0.4mag/sq" as cities go to sleep and LP fades, and the shift in the signal peak can easily render too many faint details into the noise floor. So indeed...I choose a more conservative approach. But it works for what I'm currently doing. I've been doing astrophotography for just shy of two years now (started Feb. 12, 2014), and I have results that rival well known DSLR imagers who have been doing it for a decade or more.

(Although I'm never satisfied...heh, I'm always looking to do more, get more, etc. Last year my average dark site integration time with the DSLR was 4-5 hours, which was decent. This year I am aiming to make my minimum integration time with the DSLR 8 hours, preferably over 10 hours, and go after even more faint stuff. I'm never satisfied! )

Out of curiosity, what do you do for a living?

Let me be quite clear here. Jon, I find your technique both in data capture and post processing to be first rate. Your imagery is masterful and very pleasing to the eye. I must ask, however, what will it take to be the master 5 years from now? Do not be afraid of digitization/quantization (sp?) noise.

Thank you.

I plan to get into narrow band imaging this year (if the funds pan out...and if the camera pans out, looking at the new APS-H sized KAF-16200 cameras from FLI, QHY, Moravian), and I'll be furnishing myself with a full set of Astrodon 3nm Narrow Band filters. Once that rig is going, I'll be far more concerned about minute noise factors than I am now. I've been processing narrow band images for most of this year, using data from various friends, so I am quite well versed in the processing techniques. I am also aware of how clean CCD data can be...even if it has higher read noise. I am particularly partial to QSI and FLI CCD cameras...incredibly clean noise, pure gaussian, small standard deviation...wonderful. Despite 5-7e- read noise, which is higher than the 2-3e- common with DSLRs at high ISO. The 5D III is just NOT a clean camera, even after removal of the bias signal, after cosmetic correction, after everything I can do to clean it up, it's just not all that clean. I'll be quite happy to move to CCD, narrow band, and to deal with the worries over quantization noise.