Substantially correct. I have a few tweaks below...
TriX wrote:
.... The digitizer has a theoretical dynamic range which is usually specified in bits.
Analog to Digital Converter (ADC) bit depth is simply bit depth (eg. 14-bits). Regarding dynamic range, see below.
TriX wrote:
...typically an amplifier between the sensor and the digitizer, and it contributes its own noise to the chain. When you change the ISO setting of your camera, you are changing one of two variables - one being the gain of the amplifier. As you raise the gain and amplify the signal, you also amplify the noise, decreasing the dynamic range (the dynamic range of the digitizer is limited on the low end by noise from both the sensor and the amplifier).
When you raise the ISO setting (one stop for example) you lose an equal amount of highlight dynamic range because the ADC gets clipped.
Surprisingly, you actually get a small increase in shadow dynamic range (noise goes up a little less than signal so Signal to Noise Ratio (SNR) actually improves in the shadows).
The net effect is that you loose less than one stop of dynamic range per stop of increased ISO setting.
WWhen the shadow improvement reaches diminishing returns you achieve "ISO Invariance".
TriX wrote:
...How ISO is set depends on the camera manufacturer, but in Canons (which I know best) it controls both the gain of the amplifier (as stated above) and digitally manipulates the digitized data from the A/D. Up to about ISO 800-1,000, it’s accomplished by changes in amplifier gain, but after that, changing the ISO setting multiplies the digital value of the A/D mathematically. I believe that most cameras rely only on manipulating the digital value beyond ~ ISO 1000, but I’m glad to be corrected if I’m mistaken. If you look at the DR vs ISO curve of a typical Canon, you’ll see the steps as the amplifier gain changes up to about ISO 1000, after which it smooths out because the changes are accomplished digitally.
...How ISO is set depends on the camera manufactur... (
show quote)
Canon cameras often use either digital scaling or a second analog gain to accomplish intermediate ISO settings which can give dynamic range curves a "jagged" look.
At PhotonsToPhotos I detect digital scaling and other manipulations which are indicated by various symbols on the charts.
For example, an open triangle (pointing up) indicates digital scaling whereas an open triangle pointing down indicates noise reduction.
TriX wrote:
...There are other contributors of noise in the acquisition/digitizing chain including quantization error (how accurately the A/D quantifies/digitizes the incoming analog signal) ... the net-net is that the effective dynamic range of an A/D is always less than the base DR in bits...
Quantization error is not added to the noise. As you indicated it is a limiting factor on the smallest measurement that can be made.
Theoretically quantization error is limited to 1/sqrt(12) = 0.29 bits but in practice things fall apart below about 0.6 to 0.7 bits.
So, strange as it seems, you can get about 1/2 stop more dynamic range out of an ADC than it's bit depth. In other words 14.5 stops for a 14-bit ADC.
Regards,