"Dr John Stockton" wrote in message
...
JRS: In article , dated
Tue, 14 Feb 2006 15:52:28 remote, seen in news:uk.sci.astronomy, Roger
Hamlett posted :

Difference is between quantisation noise, and quantisation _error_.
Quantisation 'noise', would be a random process, and as such ammenable
to
multiple images finding the real data underneath. Unfortunately,
quantisation error, is a non random process (it is _influenced_, by the
randomness of the data underneath). You get ADC's, that over particular
ranges, will tend to 'stick' on a particular value. If you generate
psuedo
random noise, on a small scale signal, and feed it through most ADC's,
and
then try to regenerate the signal by stacking, you find this fixed
pattern
being seen, rather than the small scale data. I was involved some time
ago, with a number of tests, trying to perform stacking like this on
small
scale audio signals, and in practice, these effects destroyed the
ability
to reconstitute the data, beyond perhaps about 5 bits extra, (using in
excess of 1024 samples).

How about adding a larger-scale analogue signal, such as a slow sine
wave or sawtooth, in such a manner that it averages out in the final
digital result? That way each portion of the result would result from
measurements using many portions of the core converter.

The analogue might be added electrically, or maybe as illumination on
the sensor.
That ought to work. Given that CCD cameras have a bias voltage applied, it
suggests an interesting experiment.
There are a number of seperate 'terms' involved, and some have quite a lot
of stuff about it on the web. Differential non linearity (covers the
'jump' behaviour at some points). However one I have never seen much
about, is the tendency of CCD's in particular, to not completely 'empty'
their wells after a flush. On some cameras, this can result in a ghost
reading for muliple frames.
In audio, there is the term 'SINAD', which is used for how well the ADC
really performs in reproducing frequencies, versus it's theoretical
performance. I have never seen anything similar, for CCD maging though.

You've read all you can about how various types of ADCs and DVMs work, I
trust? Schemes intended for getting slower conversions at sub-ppm
resolution may be adaptable for different speed/accuracy regimes.

Best Wishes