What you are saying is at least 99% wrong and I'd bet on it :-)
The CCD presents a sequence of voltage samples to the A/D, one sample per
pixel, and the A/D makes one sample per pixel. This is time domain sampling
per Shannon and not phase-sensitive frequency domain stuff where Nyquist
criteria usually prevail. Your comments about sampling frequencies are
correspondingly incorrect, since this is a discrete event sampling function
in the time domain. No pixel is sampled repeatedly to reconstruct a
frame-to-frame waveform, so the Nyquist issues that might apply to a
per-pixel waveform recovery don't apply: this is just one sample per pixel
stuff. My comments about sampling are only intended to help make this
distinction and were not intended for frame-by-frame integration concerns
(image enhancement) which I think you weren't speaking of either. There may
be some difference in our topical focus that would explain why we mostly
agree but are differing on the techie minutia that few scanner owners would
ever care to know about.
Note that 2x is not a bit "greater than" 2x, by the way, it's equal. If the
typical noise in a film's "signal" is the approximate 1 part in 512 that HP
lit states, then the 2x criterion would be met by a 10 bit A/D and anything
else (eg, 12-16 bit A/Ds) would be mismatched to the film's signal if I
understand your language and intent. None of that makes sense.
I'm well-stocked on references and experience here. I had checked before
citing Shannon since I often confuse the two myself.
cf Engineer's Handbook, which I cited in my earlier post.
Must be differences between that language now spoken in the UK and good
English in the USA :-)
Oh.. maybe it's the list serve's fault: I don't know that you are in the UK,
do I? :-)
Regards,
Chuck
> -----Original Message-----
> From: filmscanners_owner@halftone.co.uk
> [mailto:filmscanners_owner@halftone.co.uk]On Behalf Of Austin Franklin
> Sent: Thursday, July 17, 2003 4:12 PM
> To: cncole@earthlink.net
> Subject: [filmscanners] RE: scanner dmax discussion
>
>
> Chuck,
>
> > > Actually, it's Nyquist, and it's 2 to 1.
>
> > Check your Engineer's Handbook: it's Shannon for sampling and
> Nyquist for
> > stability. The concepts overlap at times, but not this time.
>
> I don't have to check, I know that you have to only sample at 2x
> the highest
> frequency to reliably detect that frequency, and you simply to NOT need 3x
> as you say. It is good to sample at slightly more than 2x. This
> is XY, and
> to sample in a diagonal (worst case), the number is different, of course,
> but it's simple geometry, and 2x applies, just the dimension used
> is 1.414 x
> the X/Y dimension.
>
> Perhaps you ought to look at some sampling theory primers:
>
>
>
> The scanner IS sampling a frequency, and Nyquist perfectly applies. BTW,
> you CAN detect at f, but not reliably. What a sampling system
> gives you is
> in fact a range, 1/2f to f.
>
>
> > > That is simply part of basis
> > > sampling theory, and is inherent in any discussion of
> sampling, scanner
> > > design included. Your comment said that the "steps" (of the A/D) had
> > > nothing to do with the intrinsic properties of the CCD and that is
> simply
> > > NOT true, which was my point, and the A/D IS in fact sized to
> the noise,
> > > and that sizing is 2:1 typically.
>
> > I think you are wrong on this technical point.
>
> What I said above is %100 true, and I'd bet on it. I'm not sure why you
> don't believe it, but clearly one of us is missing something.
>
> > The film "noise"...
>
> Film noise has nothing to do with the discussion. The scanner
> doesn't care
> a wit about what is on the film. I think this is the reason we're not
> connecting.
>
> > is
> > adequately and wholly captured by HP's 12 bit A/D, according to
> their tech
> > notes.
>
> Yeah, but that has nothing to do with sampling theory. The
> scanner doesn't
> care if it's noise on the film (what EXACTLY do you mean by "film noise"
> anyway"?) or actual image data.
>
> > I believe that the CCD noise in scanners is substantially smaller
> > than film variabilities (ie, "noise"), though it is significant.
>
> That depends on the film.
>
> > CCD
> > offsets (eg, fixed cell-to-cell zero light values) are likely to be much
> > larger than what is called "noise" like Johnson noise on CCD
> spec sheets.
>
> Larger or not, and it's deterministic and easy to correct to the noise
> resolution of the CCD, analog front end, and A/D.
>
> > That would mean that the models with 14 and 16 bit A/Ds are
> using a finer
> > step that is not directly sized to film noise or intrinsic CCD noise.
>
> Film noise has nothing to do with this, and I believe that's the reason
> you're not understanding. As I said, the A/D IS in fact matched to the
> noise level of the CCD, period, which is what the question was.
> It has not
> a thing to do with film noise.
>
> > > I'm not sure what you mean by "The A/D is scaled..." because
> it is not.
> > > The analog front end simply "matches" the output voltage range of the
> CCD
> > > with the input voltage range of the A/D, which is not scaling
> the A/D at
> > > all. It IS "scaling" the voltage TO the A/D.
>
> > Difference of viewpoint only I think:
>
> My statement isn't a viewpoint, it's a statement of how the
> analog front end
> is designed in a typical film scanner. There is no other way/reason to
> design it otherwise in this application.
>
> > I'm used to doing basic design where the A/D
> > subsystem characteristics can be scaled but the physics of
> light detection
> > or film variabilities cannot.
>
> Who said anything about physics of light detection or film variabilities?
> They, as I've said a number of times already, have nothing to do with the
> design of the CCD/AFE/AD.
>
> > > The size of the lsb is just not relevant nor is it significant (as
> > > for frame-to-frame waveform analysis/synthesis) in these
> > scanner systems.
> >
> > The size of the LSB IS in fact relevant, and IS a property of the noise,
>
> > Maybe we differ in terminology reference
> > alone since we're not debating existence of the phenomena.
>
> Having designed them that way, and having reviewed many other
> designs, I've
> got a pretty good idea that what I said is terminologically and
> technically
> accurate.
>
> > > and again, the choice of number of bits used in the A/D matched to the
> > > noise (by a factor of two, but it's still matched to the
> noise). If you
> > > have more bits, fine, but you can't have less, or you lose good data.
>
> > Specific matching by a factor of two would be bad design IMHO, and "not
> the
> > best" according to handbook criteria for quality signal
> recovery with low
> > distortion.
>
> Specifically WHAT handbook? I know adding more bits gives you nothing.
> I've done thousands of hours of testing and analysing different
> designs, and
> my statements are conclusive with my personal experience.
>
> > According to HP's and other tech lit, that matching for film noise would
> > occur at about 9 bits of A/D. We're not connecting on this
> point somehow.
>
> You're right. For some reason, you think film noise has something to do
> with this, and it doesn't (at least in what I was discussing).
> If you could
> explain why you think that, perhaps we can find out where the
> misunderstaning is...
>
> Regards,
>
> Austin
>
> ------------------------------------------------------------------
> ----------------------
> Unsubscribe by mail to listserver@halftone.co.uk, with
> 'unsubscribe filmscanners'
> or 'unsubscribe filmscanners_digest' (as appropriate) in the
> message title or body
>
----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or
body
