heptagon said:
pedro said:
heptagon said:
1.5 stops better high ISO? In RAW? No way!
In cooked JPEG - quite possible with additional processing power in easy scenes.
I'm always in for surprises. But if I read the sound comments of our experts here in the forum, 1/3 to 1/2 a stop would be a huge step forward already, given the current tech. Sorry for comparing apples to oranges now: in comparison to an a7s we talk about 24 MP vs 12 MP, and a different sensor tech as well. Anyway, if a 24 MP 6DII would deliever an 1.5 stops improvement in RAW, this would be fantastic. 3200ish ISO 8000, 6400ish ISO 18000 and 12800ish ISO 31000. Bring it on, put the same sensor into the 5DIV and I am a happy camper...;-)
Well, current sensors have about 50% quantum efficiency and 3 electrons of read noise.
Imagine you could double the quantum efficiency to 100% (which is the limit), you would gain 1 stop. (1 stop = doubling of light collected).
Then at these very low numbers we're at poisson noise and I don't have the numbers at hand but 3 electrons is not much. You need 100 electrons to get a signal-to-noise ratio of 10 with a perfect sensor. With 3 electrons of read noise that would be a signal-to-noise ratio of 100/13 = 7.7. Not much less. Not much to gain here either. And that's the limit for a single sensor camera. A physical limit which no amount of research or marketing can break.
But image processing is an area where significant gains should be obtainable. Intelligent algorithms which guess what was photographed and "paint" it for you without noise. What you see after noise reduction is not the original image, but an image painted by an algorithm showing what that algorithm "thinks" is in the picture. Improve that algorithm a lot and you can fish out more details and improve the subjective image quality by some stops.
You could attack the problem from three fronts:
1) Reduce minimum Read Noise (What's the minimum possible? Maybe 0.4?)
2) Increase Full Well Capacity (What is the maximum for 6 micron pixels? Maybe 79,000?)
3) Increase Quantum Efficiency (What is the maximum? Maybe 87%)
Please feel free to challenge my assumptions above.
Current 6D:
RN = 1.6
FWC = 74,256
QE = 47%
(1.6/0.4) x (79,000/74,256) x (87/47) = 4 x 1.06 x 1.85
= 7.84
Converting to stops: log(7.84)/log(2) = 2.97 stops
FWC depends on pixel size and limits of materials. I think the biggest potential improvements would be from reducing read noise not increasing QE.