This is part of a series of posts about the Nikon D5. The series starts here.
I took all the data gathered under to protocol of the preceding post and told my model-making program to only look at the ISOs at the base of each of the D5 “sawteeth” (isoList = [100 200 400 640 800 1250 2500 5000 10000]) and to try to model the camera using a standard model.
Here’s what it spit out:
I should explain my version of the standard model: full well capacity is obvious to those skilled in the art, but the two read noise numbers may not be. The model says that, sorta like Gaul, all read noise is divided into two parts*, the read noise prior to an amplifier whose gain changes with the ISO knob, and read noise after that amplifier, including ADC noise. The first read noise is measured in electrons, and the second in least-significant bits (LSBs), counts, or, if you really must, DN.
Interestingly, when the modeler is forced to work over many ISOs, we don’t see such a big disparity in FWCs. I’ll have to look at that, but my policy is in general to get this kind of stuff out there and clean up the mess, if any, later.
The orange balls are the per ISO modeled results, and the blue ones the results of the attempt to model the whole camera in one swell foop.
The fit isn’t great, but it’s close enough to make me think we’re looking at a camera with an impressively low preAmp read noise, and a so-so post amp read noise.
*When I’m modeling an Apinta DR-Pix camera, then the noise Gaul is indeed divided into the requisite three parts.
Maybe they did use less-than-16 bit ADC for speed concerns. is it a possibility?
To my knowledge, Nikon has never used an ADC with more than 14 bit precision. Actually, with the high noise level, 12 bits would have probably been enough.
Michael Demeyer says
Takes me back to High School Latin class. And that was a long time ago… 😉