> To represent a 120 lp/mm pattern reliably in digital, the zeroth level
> requirement is that your sampling be at least at the Nyquist frequency.
> Which requires two pixels per line pair. But in practice, you can't even
> represent a 120 lp/mm pattern with 240 pixels/mm, since that only works at
> the exact phase. It turns out that even 3 pixels/mm is pretty
> poor, but that
> 6 pixels/mm is adequate.
>
> See:
Hi David,
Notice he says "100 lines/inch", not line pair, and the sampling rate is 200
DPI for Nyquist...hum.
It's actually required to sample at LEAST at slightly more than 2x (not
simply 2x) what you are trying to sample to reliably DETECT the signal,
since, as you said above, you can not guarantee phase coherency. If you
sample at exactly 2x, you can end up not detecting the signal at all (as
shown by the transition point in the 1x Nyquist sample), since you can not
guarantee at what point along the signal you are sampling, and technically
could sample at the zero crossing for each of the two samples! This is why
CDs sample at slightly more than 2x 20kHz, or 44.1kHz. Just because you
sample at Nyquist, does not mean you don't get distortion...in fact lots of
distortion...and the higher the sampling rate, the lower the distortion, all
shown by the above reference. You do not get accurate information
reconstruction at (slightly more than) Nyquist, simply reliable detection.
Some people argue, at least in the audio world, that 20x is actually what is
needed for "accurate" reproduction...but that issue was somewhat mitigated
in the audio world with output interpolation filters (sometimes misnamed
oversampling filters), now commonly 8x...or giving 16x waveform
reconstruction.
Austin
----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or
body
