Sensor Dynamic Range, where to find specifications Options

[marcmccalmont]

With all the discussions of sensors dynamic range I spent some time looking for the sensors specifications (signal to noise ratio) It was straight forward to find the information for both the Kodak and Dalsa sensors but I could not find any specifications for either Canon or Nikon (Sony) sensors. Has any one found these specifications? It is nice to see F stops and A/D converter bit depth quoted but the limiting factor would be the sensors signal to noise ratio.

Thanks
Marc

What got me thinking of this was with the Mamiya zd back (73db S/N, 12 stops, 14 bit A/D) would the new Canon 1Ds III compete in dynamic range and thus be a better investment?

[bjanes]

With all the discussions of sensors dynamic range I spent some time looking for the sensors specifications (signal to noise ratio) It was straight forward to find the information for both the Kodak and Dalsa sensors but I could not find any specifications for either Canon or Nikon (Sony) sensors. Has any one found these specifications? It is nice to see F stops and A/D converter bit depth quoted but the limiting factor would be the sensors signal to noise ratio.

Thanks
Marc

What got me thinking of this was with the Mamiya zd back (73db S/N, 12 stops, 14 bit A/D) would the new Canon 1Ds III compete in dynamic range and thus be a better investment?

Roger Clark has published tests showing the dynamic range and noise characteristics of a number of Nikon and Canon cameras.

Electronics engineers define dynamic range as the sensor's full well capacity divided by the read noise (both expressed in electrons). In practical photography, one is interested in the amount of shadow noise that can be tolerated and posterization from an insufficient bit depth. At higher ISO, DR is usually limited by noise rather than posterization.

Norman Koren discusses bit depth and dynamic range on his web site. If you require 8 levels in the darkest f/stop to avoid posterization (banding), then a 14 bit ADC would give 11 f/stops of dynamic range and a 12 bit ADC two stops less.

Norman's excellent Imatest program can determine dynamic range for a given level of quality (limited by shadow noise). Here are the results for a Nikon D200 at ISO 100. For the highest quality, one gets only 6.5 stops of DR.



Bill

This post has been edited by bjanes: Aug 25 2007, 07:53 AM

[marcmccalmont]

Thank you for such a good answer!
Marc

[John Sheehy]

Electronics engineers define dynamic range as the sensor's full well capacity divided by the read noise (both expressed in electrons). In practical photography, one is interested in the amount of shadow noise that can be tolerated and posterization from an insufficient bit depth. At higher ISO, DR is usually limited by noise rather than posterization.

That's true at all ISOs with current cameras. Only the Pentax K10D comes close to having quantized RAW capture, and then only in the very deepest shadows at ISO 100.

Norman Koren[/url] discusses bit depth and dynamic range on his web site. If you require 8 levels in the darkest f/stop to avoid posterization (banding), then a 14 bit ADC would give 11 f/stops of dynamic range and a 12 bit ADC two stops less.

That way of looking at things doesn't apply when the noise is much higher than the ADU steps. It's all dithered noise. There is no clear posterization in the RAW data in any DSLR you can buy. You see it in *CONVERSIONS* because the conversions posterize the output. That's why you see colored blotches in blue sky; in an attempt to smooth away pixel-level noise, the converters smooth things out so that the red values posterize in multi-pixel blotches.

Norman's excellent Imatest program can determine dynamic range for a given level of quality (limited by shadow noise). Here are the results for a Nikon D200 at ISO 100. For the highest quality, one gets only 6.5 stops of DR.

That is measuring the TIFF, however. It is not a spec of the sensor, or even a measurement of RAW data.

The sensor itself, at a very minimum, would be capable of at least 4 stops more DR than what is available at ISO 1600, for a camera that starts at ISO 100.

[marcmccalmont]

From what I read then is it correct to say that the limiting factor (Dynamic Range)on a DSLR is the on chip A/D converters and for a MF digital back it is the sensor?
Marc

[John Sheehy]

From what I read then is it correct to say that the limiting factor (Dynamic Range)on a DSLR is the on chip A/D converters and for a MF digital back it is the sensor?
Marc

That would depend on the specific camera.

If a camera could truly count electrons, DR of the pixel would be determined soley by the number of maximum photons countable. Dark current noise, sensor read noise, analog ADC noise, or ADC quantization noise limit it further with real cameras. The bit depth is not likely to be a factor per se, except as it may force a converter to work in a higher precision.

[bjanes]

That is measuring the TIFF, however.  It is not a spec of the sensor, or even a measurement of RAW data.

The sensor itself, at a very minimum, would be capable of at least 4 stops more DR than what is available at ISO 1600, for a camera that starts at ISO 100.

That's quite a sweeping statement! Do you have any data to back it up?

Bill

[John Sheehy]

That's quite a sweeping statement! Do you have any data to back it up?

How about the fact that ISO 1600 is not recording 4 stops of the sensor on a camera that starts at ISO 100?

[BJL]

The sensor itself, at a very minimum, would be capable of at least 4 stops more DR than what is available at ISO 1600, for a camera that starts at ISO 100.

That is the rule of thumb I use to dig through misleading DR tests based on 8-bit per channel conversions (JPEG or TIFF): I look at the results for Exposure Index three or four stops above minimum (800 or 1600) and add those three or four stops to the shadow range. Looking at minimum EI JPEG or TIFF output probably misses a lot of shadow handling ability. DPReview DR tests often give near identical DR at EI 100, 200, 400 and even 800: clearly the lower EI settings are losing shadow details measured by the sensors in the 8-bit conversions (The loss is apparently not in A/D conversion, since the DR at even lowest ISO is still below what the A/D convertors can handle.)

[Graeme Nattress]

You've really got to get the raw data out and measure the values, and plot them to see at what point you can no longer see what is going on, and that does mean getting beyond 8bit data, which, even when gamma encoded, is not really good enough.

[BJL]

You've really got to get the raw data out...

In principal I agree that RAW is needed for definitive DR measurements. But at high exposure index like 1600, the signals from dimly lit pixels are amplified enough that they fit comfortably into the range of JPEG data, and I doubt that there is anything lost at the shadow end of the range in JPEG conversion.