side by side comparison... large print digital back VS 4x5 color film

[Tim Povlick]

It sounds like there are many parallels to digital audio at work here, but with some optical twists. In essence we are talking about the fundamentals of digital sampling. In the digital audio world an anti aliasing-filter cuts out the garbage that is out of the range of the sampling frequency. This reduces noise that would be introduced by audio signals above the sampling frequency. It seems the aperture of a scanner serves the same purpose, at least it appears so from the scans Lenny posted a link to.

I think this is a good analogy, only small distinction is an anti-alias filter prevents higher frequency signals from folding into the sampled data and masquerading as an in band signal. If we sample at 20K samples / sec we must cut off all frequencies above 10KHz to prevent them from aliasing down into the in-band signals. That is, a 12KHz tone would end up being sampled as a 2KHz tone if not for the anti-alias filter. I am sure this is what you meant by the above, I simply am amplifying to it.

Are drum scanners oversampling up to the mechanical limits of the machine (38,000 samples for every 1/16,000th of an inch)? Where the aperture is determining the frequency of oversampling by controlling the physical size of the sample area?

Doesn't the scanner have to be scanning every grain more than once or it would not be able to decipher individual grains. At a minimum, you must be taking samples at twice the frequency/size of the smallest grain you would like to render accurately, correct?

To continue the anti-alias filter concept, I suspect the camera lenses we are using act as such a filter. The really high frequency components that would cause alias problems are stopped by our camera lenses. At the scanner one can push the aperture down and down into the single micron range but the film (at least commercial film and not 'spy' film) has nothing more to give up other than the grain or grain clumps (or whatever the term). This is my theory on the subject feel free to take it with a grain of silver.

Wouldn't a higher bit depth substantial increase the accuracy of scanning? 8-bits is only 256 possibilities. That is not very many in the digital sampling world. 16-bit is 65,536 possibilities. Shouldn't scanning at a higher bit depth lower the noise within the sample, as well?

It should, that is the signal to noise ratio (SNR), should improve as bit depth increases otherwise the added bits are totally meaningless. PMT's do have some noise associated with their operation. The interested student is encouraged to check the Hamamatsu site for some interesting reading on PMTs. Including photon counting with PMTs.

Best,

Tim