1
\$\begingroup\$

As I understand it, there are two potential limiting factors on full well capacity.

a) adc bit depth and gain,

b) pixel size and ability to absorb photons

Whichever is the smaller value (in electrons), is limiting full well capacity.

Now I test a few different cameras. Data is found at www.photonstophotos.net under the headings, (i) "DxO Mark Gain Analysis" and (ii) "DxO Mark Derived Sensor Characteristics Chart and Table" - the former, providing gain values at each iso, the latter providing a full well capacity value. Data for adc bit depth is found at DxO website in camera specifications "bits per pixel".

In a few examples we get a clear match between the two.

Eg1. Nikon D3300, 12 bit adc, 8.41 gain (lowest ISO). So, (2^12)*8.41 = 34,439. This is very similar to the value given at (ii) = 34,446

Eg2. Olympus E-M10 Mark II, 12 bit adc, 6.403 gain (lowest ISO). So, (2^12)*6.403 = 26,220. This is very similar to the value given at (ii) = 26,228

But there are examples where things vary wildly.

Eg3. Canon EOS 2000D, 14 bit adc, 5.915 gain (lowest ISO). So, (2^14)*5.915 = 96,905. This is very different to the value given at (ii) = 24,226. BUT if we change adc to 12 bit = 24,222 value is now very similar. Why would the full well of this camera be listed according to 12 bit adc when it has a 14 bit adc?

Eg4. Sony A6000, 12 bit adc, 2.019 gain (lowest ISO), So, (2^12)*2.019 = 8,268. This is very different to the value given at (ii) = 33,083. BUT if we change adc to 14 bit = 33,077 value is now very similar. Why would the full well of this camera be listed according to 14 bit adc when it has a 12 bit adc?

Meanwhile, none of the above yet takes into account b) pixel size and ability to absorb photons. My understanding is that full well capacity from this perspective = pixel pitch (microns)^2 * x, where x is a value that describes the ability of the silicon (pixel) to absorb photons, and is typically between 1000-2000. My question here, is there anyway to calculate x based on given specifications or data?

And is it correct to say, that if b < a, full well capacity = b, and if a < b, full well capacity = a?

Many thanks!

Improve this question
\$\endgroup\$
2
  • \$\begingroup\$ Bit depth has nothing to do with full well capacity. Low bit depth just forces a wider range of analog charges into each digital step. \$\endgroup\$
    – Michael C
    Nov 16, 2020 at 7:46
  • \$\begingroup\$ Bill has the A6000 listed at 14bit. It could be that the A6000 is actually 14bit for single shot images and 12bit for everything else like the A6300 is; and Sony just never made that known. Or there could be an error. You could contact Bill and ask; he is quite helpful, and if there is an error I'm sure he would like to know. \$\endgroup\$
    – Steven Kersting
    Nov 16, 2020 at 15:26

1 Answer 1

Reset to default
0
\$\begingroup\$

Just because a camera is using a higher accuracy (bit depth) ADC doesn't mean the sensor is generating an equivalent accuracy (bit depth) of data. E.g. AFAIK there is no digital camera that exceeds 8bit/color accuracy. And processing 8bit color through a 14bit ADC and stuffing it in a 16bit file doesn't change anything... it's still 8bit accuracy.

Pixel size affects the engineering dynamic range. And a larger pixel has a greater potential because it has a larger capacity and difference between min/max. However, it also occupies more area. What matters in terms of recorded DR is light/area capability; and when you have more pixels of smaller size the light is simply divided among more of them. I.e. a sensor doesn't clip to overexposure earlier due to being of a higher or lower resolution.

There are some potential differences... more smaller pixels means more boundaries between them which could reduce fill efficiency; but this is an extremely small factor for modern sensors. And smaller pixels could have lower sensor read noise; and they have a lower gain requirement which works in their favor. In general you will find sensors of higher resolution, and of the same size, have equal or greater dynamic range for those reasons. In fact, that is why the latest sensors are going to dual gain designs... a smaller gain/full well capacity for lower light conditions and a larger gain/FWC for brighter light conditions.

Improve this answer
\$\endgroup\$
7
  • \$\begingroup\$ Thanks for your response. Yes, I understand the concept fine. What I am interested in is the maths and available data. I know pixel size affects dynamic range, which in simple terms = log2 (full well capacity / noise). What I wish to know is how to calculate full well capacity accurately. Can it be done based on given data/sources before camera purchase, or can it only be done via measurements after? AND if DxO are simply giving data based on adc and gain, which doesn't tell the whole story, is there any source out there giving full well capacity based on pixel pitch and absorption? \$\endgroup\$
    – ItHertz
    Nov 17, 2020 at 9:54
  • \$\begingroup\$ IDK of any source that gives FWC based on pixel pitch, only manufacturer sensor spec sheets which are hard to find for digital cameras. But IDKW you care about pixel level FWC, it is a non-factor. What you care about is image DR capability, and that information is available. FWIW, a pixel doesn't absorb photons; the photodiode is acting as a capacitor and it collects electrons which are already present (photoconductive). \$\endgroup\$
    – Steven Kersting
    Nov 17, 2020 at 17:12
  • \$\begingroup\$ YDKW? I told you why, and gave the equation. I'll say it again. "dynamic range, in simple terms = log2 (full well capacity / noise)." That is max potential DR of a sensor. So you can't say FWC is a "non-factor" then claim DR is important, when one relies on the other. I also showed how given DR info is problematic, and in response have been told bit depth adc isn't useful. Yet this is what certain given DR information seems to be based on. So on the one hand you tell me to look at the available information, and on the other hand I'm told the available information is useless. A contradiction. \$\endgroup\$
    – ItHertz
    Nov 19, 2020 at 2:01
  • \$\begingroup\$ @ItHertz, you are trying to calculate the engineering DR based on FWC. AFAIK, the only way to get the FWC is from the MDS, which are proprietary, or testing after the fact. You would also need the photosite/sensor read noise measure and ADC conversion noise measure (which can be variable). The FWC cannot be greater than ADC bit depth x gain (ADC clipping) but it can be something less. \$\endgroup\$
    – Steven Kersting
    Nov 19, 2020 at 14:23
  • \$\begingroup\$ You are also focusing on pixel level DR which is affected by photosite size/capacity. But that only tells you the difference between the min and max that can be discerned by one PS; it does not tell you the DR capability of the sensor overall. Smaller photosites have a lower fill/gain requirement and (can have) lower readout noise. This extends their DR farther into darks. They also have a lower FWC, but that doesn't matter, because they receive less light due to their smaller size. The light the PS doesn't receive falls on an adjacent PS; the DR of the sensor overall is (may be) expanded. \$\endgroup\$
    – Steven Kersting
    Nov 19, 2020 at 14:36

Your Answer

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

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