1
\$\begingroup\$

My question are based on two things.

  • SNR can be modeled by the poisson distribution and with increasing illuminance, noise increases (shot noise) but so does the SNR.

  • Cinema camera sensors have a set sensitivity (native iso) and changing the iso (or perhaps more correctly formulated exposure index) merely represents a shift in latitude, as different EI represent different log curves.

Question: This brings me to my question. When you look at double native ISO the latitude distribution is different between the two, so I’m guessing it is another set of curves. But what I’m wondering is in regards to the amount of noise between EI 1250 at base ISO 800 and EI 640 at base iso 3200.

Assumption: If I had to surmise I would think that they have the exact same amount of noise, but that the setting at ISO 3200 requires less illuminance to strike the sensor, but perhaps due to the Quantum efficiency of the sensor more photons get converted into photoelectrons.

The only issue I have with this conjecture is that it would imply that it makes more sense to just simply use the second ISO as it is more sensitive, so obviously there must be something wrong in the assumption.

I would appreciate if someone with more knowledge on this topic could give an in-depth explanation and if my assumption is correct. This is based on digital cinema cameras but I can imagine that digital stills/photography cameras similarly function.

Here is the photo I’m referencing from the Sony Venice 2 dual native ISO, showing the latitude distribution for each EI at the two settings.

enter image description here enter image description here

Improve this question
\$\endgroup\$

1 Answer 1

Reset to default
3
\$\begingroup\$

First, photosite exposure is simply a ratiometric value of the voltage the photodiode collects (acting as a capacitor in electro conductive mode) versus the maximum voltage it can collect (the full well capacity). So if a photodiode/pixel is measured (read out) at .5mv when it could hold 1mv it is recorded as 50% exposed.

Dual ISO/dual gain places a second capacitor in parallel with the pixel well... in the case of the Z8/Z9 it approximately doubles the capacity. But that also means it has to receive 2x as much light, and there are two capacitors to switch (read noise). Requiring 2x as much light means it has a lower conversion gain... it is less sensitive to light. It is also noisier in dark regions for the same reason, and it may have greater read noise due to the additional readout/switching. The tradeoff is that it has a much greater capacity to record brighter scenes or scenes of a greater dynamic range... which is the reason you would (normally) use a lower ISO.

Note that there are other variants called dual ISO... but they are not "native" sensor sensitivities. And video ISO's tend to be higher than native sensor ISO due to the gamma curve applied to the output.

Aptina white paper on dual gain/dual ISO.

Improve this answer
\$\endgroup\$
11
  • \$\begingroup\$ Is it correct that this dual conversion gain feature does not mean that the readout gain is fixed? \$\endgroup\$
    – Euri Pinhollow
    Commented Mar 27 at 9:44
  • \$\begingroup\$ And video ISO's tend to be higher than native sensor ISO due to the gamma curve applied to the output. - I do not understand. ISO is definitely a sensor characteristic (even if dynamic). What does gamma curve has anything to do with ISO? Gamma curve does not even change saturation point. \$\endgroup\$
    – Euri Pinhollow
    Commented Mar 27 at 10:22
  • \$\begingroup\$ @EuriPinhollow, yes, this conversion gain change does not change the readout gain (analog amplification). And native ISO is certainly a sensor characteristic, however the reported/stated numerical ISO isn't quite that. The CIPA standard for the recommended exposure index only requires that the stated ISO result in an exposure comparable to what a handheld light meter would report; which is also not standard... But it is the resulting apparent exposure that is the result of the ISO number rating; not the cause of it. I.e. analog amplification (ISO) is also not sensor sensitivity. \$\endgroup\$
    – Steven Kersting
    Commented Mar 27 at 13:00
  • 1
    \$\begingroup\$ @vannira, The smaller FWC of the HCG state means less DR capability; but that's not an issue because the scene is (presumably) darker. The increase noted for the Z9 is due to the N-Log curve, not the sensor. And sensor/camera noise (read/etc) is only one part of image noise. In the LCG state it is probable that any increased sensor/read noise will be overwhelmed by the exposure (light/signal) at low ISOs. Most cameras with this type of dual gain automatically switch at a relatively low ISO (200-400) when there is no longer benefit to LCG. \$\endgroup\$
    – Steven Kersting
    Commented Mar 28 at 14:33
  • 1
    \$\begingroup\$ There are also cameras (typically cinema) that have two settings for the minimum amplification of the pixel well readout, which they call "dual ISO." But this has nothing to do with sensor performance and it is all about manipulating the signal after. And when you are manipulating the signal (noise reduction/curves/etc) you can do most anything you want... IDK about the Venice2 in particular. premiumbeat.com/blog/dual-native-iso-explained \$\endgroup\$
    – Steven Kersting
    Commented Mar 28 at 14:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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