Yeah...there appears to be a lot of confusion here...and the geek in me feels the need to try and clear some of it up. And manufacturers do a lot to obscure facts, so that doesn't help us, either.
Here goes:
1) There are 2 bit-depths at play here, the bit depth of the output file, typically 8-bit or 16-bit *per channel (RGB + optional Infra-Red), and the bit depth of the A/D converter in the scanner...anywhere from 8-16 bits (or 24, with the HP, apparently). We're interested in the bit depth of the A/D converter, a 16bpc output file can accommodate any A/D converter up to 16 bits. i.e. it's easy to put 3/4 a gallon of milk into a gallon container.
2) 8 bits = 256 distinct levels (2^8)
10 bits = 1024 distinct levels (2^1024)
...
16 bits = 65536 distinct levels (2^16)
3) What really matters is dynamic range (Dmax-Dmin). But for practical purposes, lets assume the Dmin of slide film=0, so that Drange = Dmax, since on many scanners, the scanning time or exposure can be increased so that Dmin of the film yields a full-scale reading from the A/D, effectively rendering Drange=Dmax. It seems when most manufacturers talk about Dmax, the either mean Drange, or they assume Dmin=0, which is effectively the same thing.
4) The A/D converter operates linearly. All gamma encoding, ICC profiles, etc. are all applied after the scan. So for an 8-bit scanner, the very brightest thing it can sense will yield an output value of 255, and the darkest, 0. This is how you get 256 steps for an 8-bit scanner. Therefore, the darkest thing it can scan is 256 times darker than the brightest thing it can scan. A brightness range of 256 = 8 stops = 2.4 logD. As you can see from (1) every additional bit of bit depth = a doubling of the number of levels, which results in a halving of the minimum brightness level, so it's easy to equate 1 bit = 0.3 logD of brightness range, and that's theoretically correct. Now, this theoretical 8-bit scanner can scan through something that's denser than 2.4 logD, say 3.2. But in order to do so, everything lighter than 0.8logD must all register as full-scale (completely white), because the dynamic range (Drange) can *never* exceed 2.4 for a theoretical 8-bit scanner, 3.0 for a theoretical 10-bit scanner, etc.
4A) Now, if you're scanning, say B&W neg film, which may only have a density of 2.0, any decent 10-bit scanner and up should be able to handle it without much problem. In this case, the scanner is capable of a greater brightness range than the subject presents. So film base will be white (or nearly so), but the maximum density on your film will render as some dark gray. No problem, simply set the black point (a la PS "Levels" tool) and you're good to go. The scanner has basically "exposed-to-the-right" (for all you digi-cam users out there), using the least-noisy bits of the A/D, and setting the black point then (mathematically) "stretches" the data to cover the full 16-bit range.
5) So theoretically, a 16-bit A/D in a scanner has a Dmax (Drange) of 4.8 logD. But from a practical standpoint, that number is lower, due to any number of factors, which together we refer to as "noise". An excellent sampling system may only have 1 bit of equivalent noise, meaning that the average noise level in a 16-bit scanner is 65536 times smaller than the full-scale signal. An ordinary system may have 2-3 bits of noise, and a poorly-designed system may have as many as 6 bits of noise (meaning that the average noise level is now only 1024 times less than the full-scale signal, effectively creating a 10-bit scanner). Now, some noise is random, so if we acquire each sample multiple times (or do multiple scan passes in perfect registration), we can reduce the random noise in a system. But there is also non-random noise in the system. Hi-ISO fans of DSLR's may know this as "banding" or "fixed-pattern" noise. Nothing short of re-designing the scanning system will eliminate this noise.
So a disreputable scanner manufacturer might say 16-bit output file = 4.8 Dmax.
A lazy manufacturer (or a good one driven by it's marketing department) may say 16-bit A/D = 4.8Dmax (theoretically correct, but less, in practice, due to noise).
A good manufacturer not driven by it's marketing department will actually test it's design and report a number that comes out of that test...but I don't know if there are any standards that govern that, kinda like "Watts" for a speaker or amp.
Here are some examples, all with 16-bit A/D's, Dmax quoted by the manufacturer:
Nikon LS-5000 : "Density Range: 4.8D"
Epson V750 : "Optical Density: 4 Dmax "
PrimeFilm 7250Pro3 : "3.6 Dynamic Range"
Plustek 7600i : "Dynamic Range: 3.5"
Nikon appears to be the marketing scofflaw, here. Though, in practice, it probably does out-perform the other 3 in this regard (Dmax/Drange).
Anyway...that's a pretty dense post...I hope somebody benefits from it...
--Greg
Pardon my hyperbole.
