the em12 has better DR as the a93 has incamera noise reduction added to the raw files.
March 23, 2024
40
-
-
Read Noise.
-
thanks for that, is that because of the more complex electronics used in a stacked sensor produces more read noise ?
-
One thing is that these figures don't mean very much. You are looking at read noise in DN (digital number) which is not a defined unit, it's simply a count on the ADC. The size of a DN is different from camera to camera. To compare you need to look at 'input-referred read noise', which translates the read noises into electron charge equivalents. We can make a rough estimate of how the two compare. The A9 II's pixel has 3.33 times the area, so will collect approximately 3.33 times the charge for the same exposure. On the other hand, its DN are 1/4 the significance of the E-M1 II's. So, if all else was equal (including input-referred read noise) we would expect the A9 III to give about 1.2 x the read noise in DN.
It's obviously reading about twice that. This is likely in part down to the global shutter rather than the stacked sensor. A global shutter effectively has two pixels in each pixel. One is not light sensitive, and serves to hold the charge from the light sensitive one until it can be read out. Both pixels produce a share of read noise - which is one of the problems with global shutter. But check the 'input-referred read noise' for a better comparison.
I don't know how Bill diagnoses NR. I'd be surprised if Sony was really using raw NR across the range. -
Thanks for the explaination Bob, it all makes a sense now.
-
-
Other than the dual conversion gain cusp, the Sony A9-III looks like a very "ISO-invariant" camera.
Note that the definition of "input-referred" that Bill uses does not mean exposure input; it is the electron input into the photosite charge wells themselves, not the number of photons striking the sensor surface. QE is still a factor beyond what the graph shows.
-
and how does he measure that ? when all he does is collect images from people sending him them. how can you measure the different stages not just the output.
-
AFAIK, he does the necessary math to determine average numbers of electron charges for a given raw level; IOW, he derives electron counts from photon noise. He separates the photon noise from other noises by analyzing the total transfer curve.
-
FWIW here's my measurement of DR (from ISO 100% to SNR=2) compared with A7 II
-
WOW! ~17 stops DR from - what, a 14 bit linear(?) RAW file!??
Guess it's "engineering DR", much noise about... 😁 -
the base iso on the a93 is 250 , how can you plot 100 🤨
-
Especially as my monitor in my dim room is set to 85 cd/m2 white and about 5 cd/m2 black = a DR of about 4 EV - for which my 9 EV Sigma camera is quite adequate.😎
-
That looks like 125, actually.
I'd have to look at the manual, but maybe 125 exists as an "extended low ISO", which does not necessarily mean that it has less headroom than slide film, something that people worked with for many decades and didn't even know was "extended low ASA".
Many current digital cameras, especially ones with dual conversion gain, can be operated at base ISO like slide film, way below an ISO exposure index of 100, with evenly-lit scenes with matte subjects.
-
Apples, oranges, and pomegranates. DR metrics are meaningless without context and definitions.
-
Yes, and there is an ISO Standard for DR, namely ISO 15739.
But Doug Kerr doesn't regard it very highly:
-
Goes down to Lo -1.0, 125 ISO on A9 III, 50 on A7 II
-
Yes, DR is a statistical measure - on a suitably large sample you can get more than the bit depth. Sample for this is about 250k pixels.
As I said it's 100% ISO - which is about 2/3 stop short of saturation here, down to SNR = 2, normalised to A3 at 360 ppi, which as it happens is about 24MP. SNR = 2 is a sensible limit for deep shadows. On the zone system it's zone 1 'near black, with slight tonality but no texture'. SNR=2 is fine for that, so long as there isn't a load of pattern noise in there.
