One thing to keep in mind in this discussion is that many photographers still think that DR is measured by the amount of blown highlights they manage to recover with their favorite raw converter...
Bernard
Bernard,
Quite true, but a camera that has a lot of room for highlight recovery is not desirable. Highlight recovery is possible because the red and blue channels of a typical sensor are less sensitive than the green channel. This results in a white balance multiplier for the red and blue channels. For the Nikon D3 with Solux illumination, the red and blue multipliers are 1.55 and 1.61 respectively, as shown in the Rawnalize screen capture shown below. The green channel here is just short of clipping.
[attachment=19853:16_Rawnalize.png]
If we render this image in ACR at defaults, the image appears overexposed.
[attachment=19854:16_ACR_Default.png]
Using negative exposure, we appear to be recovering 1.05 stops of overexposure. However, there was actually no overexposure but merely a hot tone curve in ACR (which has a baseline exposure of +0.5 EV and a default brightness of +50).
[attachment=19855:16_ACR_MinusExp.png]
With more actual exposure of this image, the green channel would clip first but the blue and red channels would be intact and allow highlight recovery. However, there would likely be color shifts. The blue multiplier of 1.6 corresponds to approximately 0.69 stops, and this would be the maximal highlight recovery for this camera. It would be desirable if the blue and red channels had the same sensitivity as the green channel. Some photographers place a magenta filter over the lens to hold back some of the green light and balance the channels, allowing increased dynamic range.
Since the response of a CMOS or CCD sensor is linear, it makes no sense to talk about highlight and shadow dynamic range on either side of mid gray. If you expose for mid gray (18% saturation), the highlights will be 2.5 stops to the right. However, for digital exposure it does not make sense to meter from the midtones, but rather one should place highlights that must not be clipped just short of clipping. Shadow noise and effective DR will then be determined largely by the read noise of the sensor for a given full well. I'm sure you know all this, but others may be interested.
Finally a few words on DR. Dynamic range is defined as the full well capacity/read noise (both expressed in electrons). This is the engineering definition, and photographers may set the noise floor higher. For a recent MFDB sensor (the Kodak KAF 50100), the full well is 40,300 e- and the read noise is 12.5 e-, giving a DR of 11.7 stops, about the same as the Canon 1D MII (
Roger Clark). The current champion of DR for dSLRs is the Nikon D3x due to its low read noise and decent full well capacity. I do not have the actual figures for that camera, but DXO reports the DR at 12.84 stops (screen), and they report 11.07 stops DR for the 1DMII (similar to Clark's data). This is per pixel DR, reported as "screen" on DXO. The screen DR for the Phase 1 p65+ is 11.51 stops. If one normalizes for a given print size, the DRs for the D3x, P65+ and 1D MII are 13.65, 12.97 and 11.11 stops respectively. I prefer these scientific data to subjective impressions DR derived by looking at pictures of clouds.