Re: Megapixels and noise

"Roger N. Clark (change username to rnclark)" <username@xxxxxxxxx> wrote:
Floyd L. Davidson wrote:
"Roger N. Clark (change username to rnclark)" <username@xxxxxxxxx> wrote:
Remember, though, that most of that noise in the shadows
is photon noise, and we need higher QE and deeper full wells
in order to improve that. Or simply bigger pixels. At low
ISOs on current cameras, 16-bit A/Ds would help too for the
very lowest signals.
A 14-bit A/D has a dynamic range of 84 dB. Until you
have
a sensor that has more dynamic range than that, using a
16-bit A/D simply is not going to provide any improvement.

14-bits gives an 84dB range, but unfortunately, A/D converters
above 12 bits that have to work as fast as those in digital
cameras (several tens of megahertz) are not that accurate.
For example, see:
http://www.analog.com/IST/SelectionTable/?selection_table_id=124
for an extensive list of current A/D products (Analog Devices
is a major manufacturer). Look at lower power devices that
run in the ~30 to 100 megahertz range.
12 bit A/Ds: ~68 - 71 dB SNR (12-bit range = 72.2 dB)
14 bit A/Ds: ~70 - 75 dB SNR (14-bit range = 84.3 dB)
16 bit A/Ds: ~70 - 80 dB SNR (16-bit range = 96.3 dB)

I hadn't looked up the numbers, and that is indeed worse
than I had assumed. I don't know if these are or are
not representative of what is currently being used in
digital cameras, but if those are representative, and if
your figures for the sensor (given below) are correct,
then yes it appears to be right on the edge of potential
benefit.

But again, the move to 16-bit ADC's is just like the
move to 14-bit ADC's, and will wait until the cost in
speed and memory/storage is less of a burden.

With any luck, that means the same thing that everyone
rattling about why Canon and Nikon didn't go to 14-bit
ADC's with the previous models... which is to say that
they didn't, but in the *next* models that is exactly
what we will indeed see. And along with it, the much
more significant improvements in other areas that
enabled the use of higher bit ADC's is what we'll really
care about too.

Now consider the large pixel DSLRs, like the 1D Mark II.
With a full well ~80,000 electrons and read noise ~3.9
electrons, the sensor could have a dynamic range
of ~20,500 or 86 dB. Looking at the 16-bit Analog Devices
A/Ds, none will currently match SNR potential of the
1D Mark II sensor.

And I'm sure the sensor manufacturers are trying to
reduce read noise further. Imagine 2 electron
read noise with 80,000 electron full well!

We live in exciting times. The arguments about whether
film or digital is better are going to vaporize (into
religious movements).

Do the A/Ds in digital cameras do better?
See Figure 4 at:
http://www.clarkvision.com/imagedetail/evaluation-1d2
The model shows the 12-bit A/D in the range of 70.5 dB,
right in line with the top end of the Analog Devcies
A/Ds.
I don't know what the current crop of sensors looks
like
for dynamic range, but I doubt that it approaches 84 dB
without cooling.

They do. See above.

Looks good to me!

But is there any advantage to a 16-bit A/D at all, to
compensate for the slower speed and higher memory
requirements?

Hopefully, with the above data, you will see that there
is a need.

Oh, there is the need. But obviously we aren't there
yet, and manufacturers are *not* using 16-bit ADC's.

Scientific CCDs use 16-bit A/Ds running much slower to get
the SNR to match the sensor. They often take a minute
to read out the chip!

Is it the ADC or the sensor that is slower? Regardless,
speed is the reason 16-bit ADC's are not in DSLR's yet.

--
Floyd L. Davidson <http://www.apaflo.com/floyd_davidson>
Ukpeagvik (Barrow, Alaska) floyd@xxxxxxxxxx
.

Relevant Pages

  • Re: Megapixels and noise
    ... 16-bit A/D simply is not going to provide any improvement. ... for dynamic range, but I doubt that it approaches 84 dB ... the limitation is the sensor not the ... should get the same shadow detail at ISO 100 as I would ...
    (rec.photo.digital.slr-systems)
  • Re: Questions about optimizing Sensor outputs to dsp/microcontroller A/D inputs
    ... > sensor with a high voltage output and a A/D with a high voltage input ... > so that the LSB is well above the noise floor. ...
    (comp.arch.embedded)
  • Re: Questions about optimizing Sensor outputs to dsp/microcontroller A/D inputs
    ... > sensor with a high voltage output and a A/D with a high voltage input ... > so that the LSB is well above the noise floor. ...
    (sci.electronics.design)
  • Re: ISO 200000 ?
    ... Doesn't suggest that you can achieve 3e read noise in a sensor fitted to a dSLR. ... That way you could select the ISO after the event, and improve the effects that you are showing in your shadow assessment page. ... Compare that to your device saturation of 20 photons, which can have a maximum dynamic range of 20. ...
    (rec.photo.digital)
  • Re: 25 MP sensor of Sony
    ... Having more data in the highlights in the correctly exposed 14 bit raw file in the first place is an advantage over that method as you have more room to move at both ends (presuming that the sensor is capable of recording that depth of data). ... then 12 bits can cover the sensor's dynamic range. ... Roger M. Clark has argued that the dominant noise is "shot ... so does the limiting signal-to-noise ratio. ...
    (rec.photo.digital)