This is obviously somewhat "shocking". I tried to explain it in my #84; now I try a different way.
Let's take the Canon 40D as an example, based on DPReview's measurement (the 1DsMkII has not been reviewed yet). I don't agree with their result in general, they are JPEG based, which does not have much to do with the camera's capablity, but their result is suitable for the explanation.
1. The 40D loses exactly one stop from the highlights by one stop ISO increase from ISO 200 to ISO 1600. This does not appear that way in DPReview, because the middle gray moves as well. However, the fact remains, that ISO 400 cuts off the very top stop captured by the sensors.
2. DPR measured a loss of *total* dynamic range as 0.3EV from 200 to 400, 0.1EV from 400 to 800 and 0.1EV GAIN (this is ridiculous) from 800 to 1600. Together, the loss from 200 to 1600 is only 0.3EV. I am "generous", I assume 1 full stop loss.
Point 1 is correct. To be more precise on point 2, engineering DR (raw saturation/read noise) on the 40D is 11.3 stops at ISO 100, 11.3 at ISO 200, 10.9 at ISO 400, 10.2 at ISO 800 and 9.3 at ISO 1600. Are the DPR numbers you are quoting based on measurements of raw or jpeg? They do both; the jpeg numbers are rather devoid of content. The raw numbers are the result of a somewhat flawed procedure. Some people prefer a notion of "useful" DR which is somewhat more subjective, but strongly correlated to, engineering DR.
3. The loss in the highlights zwischen 200 and 1600 is *three* stops, but the total loss is only *one* stop. This means, that the sensor can deliver all together *two* stops higher DR than it is doing at ISO 200.
Loss of engineering DR is two stops, compared to exposure change of three stops between ISO 200 and 1600.
4. Why does the 40D NOT deliver this higher DR? Because it delivers always a *crop* of the total dynamic range. With ISO200 it cuts off two stops from the shadows, with ISO1600 it cuts off three stops of the highlights.
This is a misunderstanding of the capture process. Canon is not "throwing away" DR by not recording it (subject to a caveat). ISO 200 doesn't have more DR in shadows because the sensor readout is limited by the noise of the variable gain amplifier that implements the ISO setting. At high ISO this is a rather minor component of read noise, but is the dominant effect at low ISO.
The caveat is that, because sensor readout is a non-destructive procedure in CMOS technology, one could preserve the state of the sensor while doing a read of the capture data at low ISO amplification, then do another read of the same data using high ISO amplification, and finally combine the two reads in an in-camera version of HDR blending to get at the full DR delivered by the sensels. This would combine the total DR in the way you are imagining, but it would cost a factor of two in buffer capacity and be a big drag on processing speed. Not out of the question for the near future though.
This is, what most MFDBs are doing differently: they deliver everything in one - but that requires the 16bits.
No. They simply record more bits, which does not in and of itself confer more DR. They do not garner more DR due to that; even the caveat above is not available, since a CCD sensor cannot be read non-destructively.
To the extent that the sensels are gathering more photons, there may be an improvement in quality over a larger portion of the DR for MFDB's over the 40D, since the S/N ratio climbs out of being dominated by read noise faster for larger photosites (this is at the pixel level; smaller pixels need not be under such a disadvantage when compared on a fixed spatial scale instead of the pixel scale).
If the 40D converted the ISO1600 data in 13 bits and added three bits to keep the highlights, its dynamic range would be two stops larger and I would not have to throw away some of my shots I just made in the Rocky Mountains, because the 40D could not capture the scenery even with the best possible exposure.
[a href=\"index.php?act=findpost&pid=201903\"][{POST_SNAPBACK}][/a]
Again, this is a misunderstanding of how the circuitry that processes the sensor data works. It's not simply a question of adding more bits to record overamplified highlights, camera company engineers are not so naive. The ISO gain amplifier and the ADC are matched to provide as much DR as they are capable of.
Emil Martinec
