I made a few quick calculations from the data in the DPREVIEW
version of the data table.
First, we see the RMS readout noise of 21 electrons rms at maximum
readout speed. We also see that the pixel saturates at 40000. Thus,
for a maximally exposed pixel, the expected dynamic range based on
read-out noise alone is:
20*log(40000/21) = 66 dB, consistent with the entry in the table.
This also equates to 11 bits worth of range. You can get this by
dividing the dynamic range in dB by 6 by taking the base-2 log of
the dynamic range on a linear scale:
base-2 log of (40000/21) = 10.89
On the dark current front, we see
pixel fill-factor listed, but if we assume that the active pixel is
a full 6.8um x 6.8um, this equates to 2.312E-18 A for a pixel. Amps
are Coulombs/sec, with 1.6E-19 Coulombs/electron. Thus, we have a
dark current of 14.5 electrons/sec at 25C. It would seem that it
would take about a 1.5 sec. exposure for the dark current to
introduce pixel noise of the same order as the read-out noise. It
would take about an 11 sec exposure for the noise to limit the
dynamic range to 8 bits and thus for any noise to be visible at all
when viewed in 256-step grayscale - assuming no further noise
reduction algorithm is applied. All of this is at the native ISO
rating of the sensor and ignores any noise introduced later in the
signal stream.
I'm not sure if these calculations are useful. I did them quickly,
so they may have errors. They are certainly without context, since
I don't have another table to compare against!
