In message <35549D02.5D8E@xxxxxxxxxxxxxxxxxxxx>, Richard Schtzl
<Richard.Schaetzl@xxxxxxxxxxxxxxxxxxxx> writes
>Kennedy wrote:
>
>> I think you are using a different definition of dynamic range from me.
>> Normally it is defined in dB, as 20log(saturation limit/noise level)
>> with both levels expressed in volts, or 10log(.....) if levels are in
>> power or intensity. That definition certainly gives far higher figures
>> than the 1.8 - 3.8 you are quoting, which sounds more like the gamma
>> response than anything else I can think of. What are you defining as
>> dynamic range?
>
>I´m relating to the log (dec) relative optical density or brightness,
>used in print (photographic and paperbased), based on the brightest
>(paper, lightsource) and darkest spot (dye) of an picture.
>
So you are quoting the dynamic range of the film or print in Bels,
rather than decibels (dB), so for comparative purposes your figures
should be multiplied by 10, or mine divided by 10. When that is done
then the figures correspond quite closely - 18dB corresponds to an
intensity ratio of 63:1 - or roughly 6 bits, an extra 20dB gives another
factor of 100, or a ratio of 6300, just over 12bits. These are the
minimum resolution A to D converters required to digitise, without loss
of information, the information recorded on the film or print of those
dynamic ranges.
>> Whilst the number of bits used in the ADC is unrelated to the dynamic
>> range of a CCD, using less bits than the CCD is capable of simply limits
>> the dynamic range further, whilst increasing the bits simply quantises
>> noise - increasing cost without any increase in performance.
>
>Logic, a 1bit device would produce either black or white with nothing in
>between, but a high bit device (f.ex. 12bit) would not be of much use if
>the combination of CCD, AD converter and optics did not allow to produce
>more than 1024 shades (8bit).
>
Whilst your concept is correct, the figures are a bit suspect - 1024
individual shades requires a 10 bit convertor (2^10 = 1024, 2^8=256).
The point is that the CCD AND the ADC must have sufficient resolution.
In the CCD, this is determined by the dynamic range, or peak signal to
noise ratio. In the ADC it is determined primarily, but not only, by
the bit resolution. Whichever is smaller inevitably limits the true
performance of the sensor. The ADC limits the performance if there are
insufficient bits - as in the 1-bit example you quoted above (bitstream
and sigma-delta technology notwithstanding). But there is no point in
fitting a high resolution ADC to a CCD which has sufficient dynamic
range (other than market hype), since all of the lower bits are simply
noise. This is the point that Peter made earlier when he pointed out
that in commercial units "the lower n bits are always garbage". What I
pointed out is that even in high performance CCD's the dynamic range is
equivalent to only a 10bit ADC - so in a so called 12-bit unit at least
the lower 2 bits are simply noise - of no consequence unless you do off-
chip digital integration, when it is important that the noise itself is
sufficiently quantised.
>> Also recall that the figures I gave were for two-dimensional CCD's,
>> where the storage capacity is limited by the space available under each
>> pixel. In the examples that you have quoted, a linear CCD is used where
>> the storage capacity can be much larger since it is ultimately limited
>> by the mean free path in the substrate if sufficiently large gates are
>> fabricated. Linear CCD's with storage capacities of 50-100 million
>> carriers are not unknown, giving a dynamic range up to 10,000 and
>> requiring at least 14 bits to fully quantise into the noise floor.
>>
>> Of course, the time required to scan a frame with such devices makes
>> them totally impractical for use in a digital camera back - unless you
>> only want to make pictures with exposure times of the order of several
>> tens of seconds.
>
>You wrote earlier:
>
>> Astronomical units - as Lee will doubless confirm - achieve higher
>> dynamica ranges by reading the CCD out numerous times, converting the
>> signal to digital form and accumulating the result. This increases the
>> effective photon noise limit by a factor of the square root of the
>> number of integrations, but pretty soon this becomes dominated by other
>> effects - particularly dark current noise etc. Whilst such multiple
>> integrations are fine in the relatively stable astronomic field, they
>> would not be suitable for normal photography since the subject movement
>> would corrupt the image.
>
>Didn´t Nikon use(d) this technique with there Coolscans? I´d remember to
>have read, that the scaning time is relative to the slides density.
>
AFAIAA, its not something they offer in their 35mm Coolscan products,
the LS-20 and LS-1000.
I just did a quick check with my Nikon - a clear full 35mm frame (no
film in the caddy) scans in 63seconds whilst a full black frame scans in
130seconds. Other models might be different of course.
They adjust the integration time (ie. the time the linear CCD used in
the scanner views each point) relative to the image density. This
ensures that the CCD storage capacity and the ADC range are almost
optimally utilised.
This is important in CCD's since as well as the photon noise limit there
are additional sources of noise and, of course, the dreaded dark
current. If the slide to be scanned is dark the storage is only
partially filled, and the photon noise becomes more significant. Whilst
a well designed, low noise, million carrier CCD might come close to
10bit performance - as shown earlier - if each pixel is only 10%
illuminated then the signal to noise is reduced by at least the square
root of 10 - probably more due to the dominance of other noise sources,
such as gate transfer noise and noise in the output amplifier. So the
CCD may only provide a dynamic range equivalent to less than 8 bits
under these circumstances. Integrating for longer will permit the CCD
pixels to get closer to saturation where the other noise sources are
less significant and closer to the full performance can be obtained.
Even with longer integration times, however, the performance will not be
as high with a dark slide as it is with a bright one since the CCD will
also suffer from dark current which not only adds noise to the signal
but introduces a fixed pattern superimposed on the image. The only way
to reduce that effect is to keep the CCD cold - and I am sure that any
astronomers are more than familiar with peltier technology required to
do this. Anyway, its this dark current is why the Nikon limits the
maximum integration time even for completely black frames.
Its also worth noting that the Nikon provides substantially better scans
when it is cold - hence the integrated unit that fits into your PC has a
lower practical performance than the stand-alone unit - the heat from
the Pentium processor just heats the CCD up, increasing the dark
current. I decided to fit a small peltier device on a heatsink and fan
to keep my Nikon Coolscan cool inside the PC case since I upgraded to a
Pentium II!
Whilst multiple integrations increases the dynamic range of the system
it will not 'bring out the blacks' from saturation level, which the
longer integration time will also do. So although they adjust the scan
time to match the slide density, they're not doing multiple integration.
If you wanted to get into writing your own software to control the Nikon
(not something I would even consider!) you might be able to get it to
adjust the integration time relative to the slide density AND do
multiple integrations to increase the dynamic range. That way you could
possibly get an LS-1000 to give more than the 10bit per colour output
that it is currently limited to, but it would make scanning slides even
longer. A 36-bit colour scan would require at least 4 integrations per
pixel, meaning the entire scan would take over 5 minutes per frame.
Not exactly the sort of technology you would want to see on the back of
a potential OM-5, I suspect. :-)
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's pissed.
Python Philosophers
< This message was delivered via the Olympus Mailing List >
< For questions, mailto:owner-olympus@xxxxxxxxxxxxxxx >
< Web Page: http://Zuiko.sls.bc.ca/swright/olympuslist.html >
|