A couple of recent reviews (here
and particularly here
) have shown dynamic ranges from a camera's RAW files which exceed the 12 bit resolution of the camera's analog to digital converter. How might that work?
Well, both my assertion that this is what is happening and the explanation are a guess but for a grey test subject, such as one would use for a DR measurement, the red, green and blue photosites will have different quantum efficiencies so it's entirely likely that when, say, the green sites have reach full well capacity there's still room for some more electrons in the blue and/or red sites. A clever RAW converter might use the extra headroom from those adjacent differently coloured sites to reconstruct a luminance signal in excess of the theoretical 12 EV available from a single pixel.
Is this cheating? In a way I suppose one should say "Yes" but in the real world swapping a little spatial resolution to recover detail from otherwise blown highlights seems like a very good trade off provided the RAW converter reverts to a more expected behaviour when pixels haven't reached full well capacity. I would also argue that it doesn't invalidate comparative tests between cameras provided the test conditions are the same and the RAW converter used is the same
and one can assume that said RAW converter uses the same algorithm to extrapolate dynamic range.
There is a potential "gotcha" in that when a camera is tested it can be so new that the only RAW converter that can be used is the proprietary one that ships with the camera. In such cases the red flag would be "converted to TIFF" before the software used to measure dynamic range is deployed as that implies that the test is more likely to be of the supplied software rather than a best case result of what can be achieved from the hardware once more capable third party RAW converters are available.
The first test I linked to above did use the same software so comparison between the two cameras is fair. I strongly suspect that the second test I linked to used different RAW converters which raises a huge question mark in my mind about its validity.
Just a thought. I might be wrong.