Does the "system gain" (i.e. electrons/ADU) change when one changes the ISO of a digital camera? To put this another way is there an analog amplifier between the sensor and the ADC whose gain is programmed to change when the ISO setting is changed?


is there an analog amplifier between the sensor and the ADC whose gain is programmed to change when the ISO setting is changed

Quite often there is. It can be a multiplier after the ADC, too. Can be that gain does not change (Sigma/Foveon, some digital backs, some other cameras). Can be that it is not for all ISO settings, like "extended low ISO" on Nikons and Canons. Can be a combination of gain and multiplier (intermediate ISO settings on many Canons, Hi ISO on some cameras).

| improve this answer | |

The exact recipe will be a closely guarded trade secret for manufacturers but certain things hold true. Chipworks might give you some background that may help for specific cameras (or it might not).

The actual light captured and charge generated will always be the same at the sensor. There are multiple points where the results can be adjusted.

CMOS sensors have an on-chip gain control / amplifier which makes adjustments during the readout (CCD's don't).

There can a separate amplification stage en-route to the ADC. Necessary for CCD but some CMOS systems have been seen to do this too.

Finally a third layer of digital adjustment may be applied.

| improve this answer | |
  • 1
    Noise and signal analysis allows to finŠ² out what is happening. With digital multiplication, the histogram contains gaps. With analogue amplification stage raw data is the same for different ISO settings (given the ISO setting is used to compensate for less light), but Std.Dev. is different. – Iliah Borg Jan 13 '15 at 22:40
  • Not really, since in the real world what happens is going to be a blend of all the techniques there rather than any specific one. Also what is used is likely to be specific to each sensor (if not each body). So far as I'm aware none of the main applications include 14-bit histograms (even if you had a 16k pixel wide display to show it on) you probably would not see a gap and if there was then it would be easily buried by the s/n ratio anyway. – James Snell Jan 14 '15 at 9:37
  • The blend it is, but the signatures of each method are clear and thus ingredients can be established. There is no need to display the full histogram - a segment that fits display is enough, or one can dump it to something like CSV. Gaps are clear, non-masked by noise, as digital manipulation is the last operation, obviously. I happen to be one of the authors of a program, one of the uses is precisely that - statistical and visual analysis of raw data. – Iliah Borg Jan 14 '15 at 13:24
  • Comments are not a place for discussion. If you have evidence of what happens then you have an answer you can edit to include it, the fact that you didn't suggests that you are likely to be talking rubbish. – James Snell Jan 14 '15 at 15:38
  • Sorry, but rubbish is your "you probably would not see a gap". – Iliah Borg Jan 14 '15 at 17:01

Not the answer you're looking for? Browse other questions tagged or ask your own question.