Nikon D5 — what’s going on with the sensor?

This is part of a series of posts about the Nikon D5. The series starts here:

Nikon D5 — read noise vs ISO setting

As I mentioned in a previous post, I’ve been unable to model the behavior of the D5 sensor as the ISO control is changed using the standard model. That’s not the first time that’s happened with a camera I’ve been testing. All of the Sony alpha cameras that employ the Aptina-developed DR-Pix conversion gain changing technology defy modeling with the standard model as well. However, since — at least for now — the conversion gain change takes place at only one point in the ISO adjustment range, it’s been pretty easy to find where it happens as model it.

Lacking similar clues about what’s going on in the D5 sensor, I’m confused — confused at a fairly high level, but still confused. The D5’s departures from sensor orthodoxy are not confined to one ISO.

Here’s are the parameters that I got after I modified my model to take into account what appears to be Nikon’s use of digital gain in the D5, with analog gain only at ISOs 100, 200, 400, 640, 1259, 2500, 5000, and 10000,

And here’s a graph of the modeled total read noise in the raw file versus ISO of the overall modeled camera (the blue dots) and the model results at each individual ISO (the orange dots):

You can see the two sets of points match pretty well at low ISOs, but the per-ISO numbers start looking better than the overall ones at the middle ISOs, only to have that situation reverse itself at the high ISO settings.

If I come at it another way, and use the per-ISO results to set the FWC and read noise for the ISO 100 point and the ISO 10000 point, we get this for the modeling parameters:

And this for the comparison:

Now the low and high ISOs are well modeled, but the overall modeled camera does too well at the middle ISO.

I am open to — more than that, I’m eager for — ideas about what Nikon might be doing with this sensor.