Ilya Zakharevich wrote in
:

[A complimentary Cc of this posting was sent to
David J. Littleboy
], who wrote in article
:
Of course, my reading of figure 5 on this page indicates that there's
no point to bothering with a 14-bit A/D converter on either this or
the 40D.

My hunch is that N-bit converter would deliver at most N-1 or N-1.5
usable bits. If so, then using 14-bit converter makes sense even if
only 12bit is needed FOR STORAGE.

That's my thinking as well; if there is some benefit in the 14-bit ADC,
you might not need 14 bits in the output.

Comparing the deep shadows of a 14-bit ISO 100 RAW from the 1Dmk3, and a
12-bit version of it (the 2 LSBs zeroed), with exactly the same RGB
interpolation and WB, there is no significant difference, even pushed to
ISO 2800. So, even *if* the 14-bit data had less analog noise underlying
it, having the 2 LSBs in the RAW is of questionable value, in terms of
what you get vs what it costs you. Personally, I think that only the
cleanest of DSLRs warrant even 12 bits at ISO 100 (Pentax K10D, Fuji S5,
Canon 1D(s)mk*, etc), and that 10 or even less bits are needed at ISO
1600. The D2X could probably get away with 8 or even 7 bits at ISO 1600
- read noise is about 60 ADU at 12 bits, and that would still be almost 4
ADU at 8 bits.

I just tried it - zooming in to 640 RAW levels from the shadows of the
blue channel of a D2X ISO 1600 file (IOW, pushed to about ISO 5100),
mapped to 0 to 255 on the screen, I can see almost no difference
whatsoever in the quality of the original 12 bits, and after integer-
dividing and -multiplying by 32, for an effective 7 bits of RAW data.

At 6 effective bits (read noise is now about 1 ADU), I start to see a
slight unevenness in tonalities. At 5 bits, it is obvious, and at 4
bits, the greyscale image becomes severely posterized.

--


John P Sheehy