the Canon 1D Mark II, over 10 stops of information!". This is consistent with Ray's observations.
Bill, don't you think considering a 10 f-stop DR in a 12-bit RAW based camera is a bit too optimistic? if we assume linear behaviour of the sensor into this usable DR, the entire 10th lowest f-stop would be codified in just 4 different tone non-interpolated levels. I think that's too poor even if noise is not present in such deep shadows, isn't it?
0EV: 2048 levels, 2048..4095
-1EV: 1024 levels, 1024..2047
-2EV: 512 levels, 512..1023
-3EV: 256 levels, 256..511
-4EV: 128 levels, 128..255
-5EV: 64 levels, 64..127
-6EV: 32 levels, 32..63
-7EV: 16 levels, 16..31
-8EV: 8 levels, 8..15
-9EV: 4 levels, 4..7 -10EV: 2 levels, 2..3
-11EV: 1 level, 1
The only thing I can think for reaching such a figure (10 f-stops DR) is to consider a non-linear low end of the response curve of the sensor as a usable range that could include more than 4 effective levels to codify degrees of lightness (of course being these levels "stolen" from upper f-stops, -8EV and up). However the dots in the log-log graphic don't seem to show any non-linearity in the input-output response curve:
And in any case he is using a 16-bit reference, with which I don't agree as those new levels appearing in the 16-bit range and filling it, are purely interpolated. Real captured levels are in a 12-bit range, and only 4 different of them represent the whole 10th f-stop.