Pixel saturation

[Guest]

As I understand it the iso setting on a camera adjusts the gain on the amplifier which reads the voltages from the sensor pixels. From what I've read CCD sensors have a single shared amplifier but CMOS sensors have an amplifier for each pixel. So I was wondering if changing the ISO setting would affect the saturation of a pixel. I think a sensor pixel can only store a certain amount of voltage so there is a saturation point when it has reached it's capacity. But is it possible to cause a saturated value from the amplifier by having too high an ISO value which would cause the ADC to saturate even if the pixel being read was not saturated.

On my last session I realised that some of the pixels were saturated. So I thought I could either reduce the exposure time or reduce the iso but then I thought perhaps reducing the iso setting might not help.

My camera is a canon eos 450d and I think it has a cmos sensor.

Cheers

Steve

 

[vlaiv]

You can cause pixel to saturate if you use high enough gain, but that saturation has nothing to do with electronics.

It has to do with the fact that ADC has fixed output range.

Say you have 14bit ADC in your camera (most DSLR cameras have either 12bit or 14bit). This means that you can have 0-16383 as a result of ADC conversion. Resulting number is referred as ADU number (short for analog-digital unit I think). Value is obtained by measuring number of electrons in pixel potential well and dividing with e/ADU gain factor.

Say you have 0.1 e/ADU gain selected (which is high gain) and you have 6000e in your pixel well. Resulting ADU will be 6000 / 0.1 = 60000 but you can't write it down in 0-16383 range so it gets clipped to 16383ADU. You have saturated pixel.

Same scenario of 6000e and gain setting of 1 e/ADU will result in 6000ADU - without clipping since it falls into 0-16383 range.

Gain is just a multiplier - it does not contribute to sensitivity of the camera as many people believe (although, because of construction of CMOS sensors, it does have impact on read noise of camera - higher gain means slightly lower read noise).

[Guest]

Thanks Vlaiv. I understand about the ADC saturation but can you also get saturation because the pixel well cannot accept any more electrons. Can you get the situation where the pixel well is full and can't accept any more electrons but you have a very low gain on the amplifier (very low ISO) so the output from the ADC is not saturated and you might not see this as a saturated pixel.  I don't know what the figures are but lets say the pixel well cannot accept more that 20000e. If you have a very low gain, say 20 e/adu then the output would be 10000 which would not saturate the ADU.

 

[vlaiv]

1 minute ago, woodblock said:

Thanks Vlaiv. I understand about the ADC saturation but can you also get saturation because the pixel well cannot accept any more electrons. Can you get the situation where the pixel well is full and can't accept any more electrons but you have a very low gain on the amplifier (very low ISO) so the output from the ADC is not saturated and you might not see this as a saturated pixel.  I don't know what the figures are but lets say the pixel well cannot accept more that 20000e. If you have a very low gain, say 20 e/adu then the output would be 10000 which would not saturate the ADU.

 

Indeed, however, engineers designing the chip usually match ADC bit count and full well capacity together with read noise.

Minimum gain that you can set is usually derived as FWC / max_bit_count. In situation where you have 20000e FWC and 12bit ADC (with max value of 4095), resulting min gain would be around 20000/4095 = ~4.915.

In fact that is the way you measure FWC and for all intents and purposes - that is maximum FWC that you can expect. Actual FWC could be say 22000 - but you'll always see it as 20000 since that is saturation value that you can get.

Read noise is matched with this to mask quantization effects. Since ADC numbers are whole numbers and when you divide with a number larger than 1 you get rounding off error. Say you gathered 12e and your gain value is 7e/ADU. Number that you end up getting is 12/7 = 1ADU (can't have decimal numbers here).

When you try to get number of electrons back you need to multiply with gain value so it would be 1 ADU * 7e/ADU = 7e. There is 5e of error due to this rounding in this example. In order to minimize this masking technique is often used - random noise is added (this is in signal processing). Sensors have read noise so you don't have to add it back in, you just need to match it against FWC and ADC (gain setting) so it masks quantization properly.

Large value of minimum gain requires larger read noise. Here are some examples:

ASI174 - FWC 32K, min Gain ~8 e/ADU, read noise ~5.8e

ASI1600 - FWC 20K, min Gain ~ 5 e/ADU, read noise ~3.6e

Both cameras have 12bit ADC (max 4095)

ASI071 - FWC 48.6K, min Gain ~2.95 e/ADU, read noise ~3.25e

This one has 14bit ADC so max value is 16383 ( * 2.97 = 48657.51). Read noise here is larger than it needs to be but I guess they hit the limit and could not make it smaller at this gain setting.

[Guest]

Just been reading the AAVSO guide to dslr variable star observation. There is a table of full well capacity against ISO for a canon 1100D camera which I think might have the same sensor as my canon EOS 450D

ISO    Full well capacity

100    46376

200    25603

400    12787

800    6088

 

 

[Scooot]

Does this link help. I bookmarked it ages ago as I thought it was a good explanation. From what I’ve read the most appropriate ISO for the 450D which is what I use is 400.

http://astrofriend.eu/astronomy/tutorials/tutorial-iso-dslr/tutorial-iso-dslr.html

http://dslr-astrophotography.com/iso-values-canon-cameras/

Edited December 16, 2020 by Scooot

[vlaiv]

Here is good resource on DSLR model statistics:

https://www.photonstophotos.net/

It says that 450D has:

14bit ADC, 30555 FWC,  ISO 212.4 as unity gain (at ISO 212.4 - gain is 1 e/ADU, while at ISO 100 it will be 30555 / 16383 = ~ 1.865 e/ADU)

Btw, you can measure these values yourself with a test protocol. Let me find that for you. For example this:

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html

[Guest]

Thanks, I've been using ISO 400 so far but the AAVSO guide recommends 100-300 

[Scooot]

3 minutes ago, woodblock said:

Thanks, I've been using ISO 400 so far but the AAVSO guide recommends 100-300 

They’ll be fine but the reduction to dynamic range by increasing it to 400 is so minor, that most of the time it’s worth using 400. I think that’s the gist of the argument. 

[Scooot]

There’s some good info about the 450D in this link as well.

https://www.the-digital-picture.com/Reviews/Canon-EOS-Rebel-XSi-450D-Digital-SLR-Camera-Review.aspx