1Dx M2 Sensor Resolution - Back of envelop estimate

[raptor3x]

Because of the discussion that has arisen here, see my answer 50, I asked Prof. Eric Fossum The Inventor of the active CMOS
if noise increases with higher iso as Neuro states in this thread and I got this answer.

Hello Eric
I have a discussion with a person who said that NOISE increase with higher iso, I try to explain that noise does not increase with higher iso, the noise is lower with less light, but the noise-to-signal ratio gets higher.

Comment please?

the answer:


Referring to photon shot noise, this is exactly correct. When considering shot noise plus read noise, as there is less light, at some point the photon shot noise becomes smaller than the read noise so the total noise (sqrt of the sum of the squares) becomes constant at the read noise level.

Best regards,
Eric

Ask him if output-referred noise in ADU increases with higher gain settings.

 

[tpatana]

Because of the discussion that has arisen here, see my answer 25.

 

[raptor3x]

Just for reference, the actual read noise generated by the electronics does not change with ISO

Are you sure? The component coming from the ADC itself does not change, but the amplification adds some more noise and I would also count that as read noise. It's no great contribution at low ISO values but becomes more important at higher ISO values. Also in the table which filluppa has now posted about twenty times the input referred noise does not decrease inversely proportional with ISO (it's almost constant at high ISO), so the output referred noise increases (although slower than proportional to ISO).

Yeah I'm sure, what's happening is that you can break the read noise down into two components: upstream of the amplifier and downstream of the amplifier (it should also be noted that some cameras use 2-stage amplifiers but the same approach is easily extensible to those as well). The resulting output referred noise becomes something like

No^2 = C*(N_us)^2 + (N_ds)^2

Where No is the output referred noise, N_us is the actual read noise of the upstream components, N_ds is the contribution of the downstream components, and C is the gain. To get the input referred noise we simply use the relationship*

Ni = (1/C)*No

where Ni is the input referred noise, which gives us

Ni^2 = (N_us)^2 + (1/C)*(N_ds)^2

Looking at both of these forms, you can see that as the gain becomes large, the relative contribution of the downstream noise becomes small; however, looking at the input referred noise should make it clear why you get an essentially constant read noise at high ISO in that table. The noise coming from the electronics doesn't change at all with ISO, but both their relative and absolute contributions to the final image absolutely changes.

*There's a little bit more to it than this but I honestly don't remember all the details; the scaling holds up though.

The C in equations 1 and 3 has to be squared. But then it is pretty much what I said, when we identify my "noise coming from the ADC itself" with N_ds and " the amplification adds some more noise" with C*N_us. And because this amplification is done by electronics, I would definitely see C*N_us as increasing "read noise generated by the electronics", but that is mainly a question of definition.

You're absolutely correct about that it should be C^2, sloppy typing on my part. My point was more just that the N_us and N_ds terms are constant and the apparent change in read noise is coming as a direct consequence of the signal amplification.

 

[Loswr]

Asking a person much smarter than oneself an inappropriately worded question out of context to the relevant discussion, and getting exactly the answer for which one thinks one is looking...and then misinterpreting the answer to support one's point even though it does not... I think that's about the worst thing I've ever heard. How marvelous!

I think everyone since long before answer #84 and with a little knowledge of Mikael understands:

...he still believes he is correct, and will persist in doing so regardless of all evidence to the contrary.

 

[takesome1]

In one of Eric Fossum's videos on his website one of the topics is Societal issues that have come out of sensor and digital technology.The loss of privacy, the dangers of criminal activity and more.

He didn't mention the rise of the sensor internet troll, of which he may have been a victim.

He does discuss Noise at about mark 22.

https://www.youtube.com/watch?v=JkBh71zZKrM

 

[3kramd5]

Asking a person much smarter than oneself an inappropriately worded question out of context to the relevant discussion, and getting exactly the answer for which one thinks one is looking...and then misinterpreting the answer to support one's point even though it does not... I think that's about the worst thing I've ever heard. How marvelous!

I think everyone since long before answer #84 and with a little knowledge of Mikael understands:

...he still believes he is correct, and will persist in doing so regardless of all evidence to the contrary.

are noise increasing with higher iso NEURO
yes or no.

All else being equal (i.e. same exposure time and aperture), yes. Right?

 

[bdunbar79]

Asking a person much smarter than oneself an inappropriately worded question out of context to the relevant discussion, and getting exactly the answer for which one thinks one is looking...and then misinterpreting the answer to support one's point even though it does not... I think that's about the worst thing I've ever heard. How marvelous!

I think everyone since long before answer #84 and with a little knowledge of Mikael understands:

...he still believes he is correct, and will persist in doing so regardless of all evidence to the contrary.

are noise increasing with higher iso NEURO
yes or no.

All else being equal (i.e. same exposure time and aperture), yes. Right?

what do you mean with same exposure? yes right

This isn't that difficult of a thing to grasp. Set your camera to 1/125s, f/4, ISO 100. Don't move, don't do anything except flip your ISO dial to 200. Which photo has more noise? There is a noise difference between the two and the ONLY thing done was raise ISO. For some reason you can't or won't understand that, and nobody can figure out why.

 

[3kramd5]

Asking a person much smarter than oneself an inappropriately worded question out of context to the relevant discussion, and getting exactly the answer for which one thinks one is looking...and then misinterpreting the answer to support one's point even though it does not... I think that's about the worst thing I've ever heard. How marvelous!

I think everyone since long before answer #84 and with a little knowledge of Mikael understands:

...he still believes he is correct, and will persist in doing so regardless of all evidence to the contrary.

are noise increasing with higher iso NEURO
yes or no.

All else being equal (i.e. same exposure time and aperture), yes. Right?

what do you mean with same exposure? yes right

No, I'm ISOlating (badum ch) the variable. Not the same exposure. Same exposure time and aperture diameter; different ISO. How does noise vary? More, less, or equal?

Alternately stated, keep the input signal the same, but double ISO. Did you increase, decrease, or maintain your SNR?

 

[takesome1]

In one of Eric Fossum's videos on his website one of the topics is Societal issues that have come out of sensor and digital technology.The loss of privacy, the dangers of criminal activity and more.

He didn't mention the rise of the sensor internet troll, of which he may have been a victim.

He does discuss Noise at about mark 22.

https://www.youtube.com/watch?v=JkBh71zZKrM

please explain what you are writing or telling me
are you one of them who think noise is increasing with higher iso as Neuro does?

What I am saying is your trolling not to prove any point about noise and ISO, you are simply trolling to attack one particular person, which is Neuro. You think if someone agrees with you then it proves Neuro wrong, which will be an incorrect assumption.

To the point of what I think, I believe what Eric Fossum says. I also believe Neuro is referring to a different point in the process stream and his assessment is correct in the context of what he was saying. If you read Neuro's posts he carefully answered any of your questions in the context of the original discussion where you challenged what he was saying.

To sum up what I think of the conversation in regard to this subject. Eric Fossum is correct, Neuro is correct and your inability to look beyond a single point has left you with an incorrect assumption.

 

[3kramd5]

To sum up what I think of the conversation in regard to this subject. Eric Fossum is correct, Neuro is correct and your inability to look beyond a single point has left you with an incorrect assumption.

Speaking of assumptions, here is one, and it goes to my question above.

I assume that most people who deliberately set exposure on a camera typically choose exposure time and aperture based on the subject and their creative intent, and set sensitivity according to the availability of light and their desired brightness. That may be a poor assumption, but I'm self-centered and that's how I shoot.

So say I'm shooting a hummingbird in flight and want to stop motion. I want 1/4000 or faster exposure time. Say I'm using a lens combination which allows me f/5.6 aperture at the maximum, but I want f/8 for a little wider DOF given my distance to subject and a desire to have the entire bird in focus. That right there is the subject/creative intent side of the equation. Next comes light. I can't change or augment ambient - it is what it is. I'm at ISO1000, but a "chimped" glance at the histogram shows that I'm about 2 stops under where I want it to be in the first shot. Accordingly, I change to ISO4000 and shoot again. Can I expect more noise, less noise, or the same noise in the second shot?

 

[takesome1]

To sum up what I think of the conversation in regard to this subject. Eric Fossum is correct, Neuro is correct and your inability to look beyond a single point has left you with an incorrect assumption.

Speaking of assumptions, here is one, and it goes to my question above.

I assume that most people who deliberately set exposure on a camera typically choose exposure time and aperture based on the subject and their creative intent, and set sensitivity according to the availability of light and their desired brightness. That may be a poor assumption, but I'm self-centered and that's how I shoot.

So say I'm shooting a hummingbird in flight and want to stop motion. I want 1/4000 or faster exposure time. Say I'm using a lens combination which allows me f/5.6 aperture at the maximum, but I want f/8 for a little wider DOF given my distance to subject and a desire to have the entire bird in focus. That right there is the subject/creative intent side of the equation. Next comes light. I can't change or augment ambient - it is what it is. I'm at ISO1000, but a "chimped" glance at the histogram shows that I'm about 2 stops under where I want it to be in the first shot. Accordingly, I change to ISO4000 and shoot again. Can I expect more noise, less noise, or the same noise in the second shot?

We'll, I'm doubt you needed to go in such detail for the question.
But from your question you say "shot" so we would be doing a visual comparison of the finished product.

 

[3kramd5]

We'll, I'm doubt you needed to go in such detail for the question.

Ordinarily I'd agree, but then came this thread. Without exhaustive detail, I could expect a screenshot of http://www.sensorgen.info/CanonEOS-1DX.html in reply.

Go ahead and address both capture (i.e. sensor level) and final product (i.e. raw file).

 

[tpatana]

Asking a person much smarter than oneself an inappropriately worded question out of context to the relevant discussion, and getting exactly the answer for which one thinks one is looking...and then misinterpreting the answer to support one's point even though it does not... I think that's about the worst thing I've ever heard. How marvelous!

I think everyone since long before answer #84 and with a little knowledge of Mikael understands:

...he still believes he is correct, and will persist in doing so regardless of all evidence to the contrary.

are noise increasing with higher iso NEURO
yes or no.

Are you sure you want to do this?

Yes or no.

Aces.

 

[No Mayo]

A tribute, an oversimplification, an observation, and an apology

My grandfather owned and operated a small bakery for several decades. He worked 12 hour days, six days a week until he finally retired. As a kid, I worked with him on many Saturdays and for one Summer. He taught me card games and he was a great living example of how to get along with every kind of person. He was our Poppa and we all loved him dearly. He was a people person and he valued each and every person and what they had to say. He was tough (a professional boxer in his 20s) and he was fair and kind. He was not an educated man, but he was a wise man. I believe he might say... "Filluppa... if we change light to muffins... you are saying that the small muffin has less salt, but tastes saltier because of the salt to batter ratio? I agree with you. And Neuro... you are saying that the small muffins are saltier because they taste saltier? I agree with you. Now let's play cards..."

Neuro, I personally value your contribution to this forum much more than you may realize. I have learned much from you and have never thanked you for sharing your knowledge and insight... Thank you.

Filluppa, Some might say that it is just tech semantics that noise is not greater in high ISO, but due to it's increased signal to noise ratio it appears greater, but I did not know this and I found it interesting. I value your knowledge and hope that you will continue to share it on this forum. I also offer this from my heart... please do not take personal offense from differing opinions. We are all human and we may not always receive each others viewpoints on things as well as you might hope. Do not measure your value based on a person's ability to appreciate it.

My apologies if this is offensive to anyone and/or if you feel that my thoughts should have been kept to myself.

 

[midluk]

To sum up what I think of the conversation in regard to this subject. Eric Fossum is correct, Neuro is correct and your inability to look beyond a single point has left you with an incorrect assumption.

Speaking of assumptions, here is one, and it goes to my question above.

I assume that most people who deliberately set exposure on a camera typically choose exposure time and aperture based on the subject and their creative intent, and set sensitivity according to the availability of light and their desired brightness. That may be a poor assumption, but I'm self-centered and that's how I shoot.

So say I'm shooting a hummingbird in flight and want to stop motion. I want 1/4000 or faster exposure time. Say I'm using a lens combination which allows me f/5.6 aperture at the maximum, but I want f/8 for a little wider DOF given my distance to subject and a desire to have the entire bird in focus. That right there is the subject/creative intent side of the equation. Next comes light. I can't change or augment ambient - it is what it is. I'm at ISO1000, but a "chimped" glance at the histogram shows that I'm about 2 stops under where I want it to be in the first shot. Accordingly, I change to ISO4000 and shoot again. Can I expect more noise, less noise, or the same noise in the second shot?

You forgot to specify one detail: Do you lift the exposure of the first shot in post in the raw converter or do you use the shots directly as coming from the camera?
If you don't lift the exposure you will have more noise (in ADU) in the final image with higher ISO. But the noise will increase by a factor of less than four (how much less depends on the contribution of downstream read noise compared to upstream read noise and Poisson noise), while you signal (also in ADU) raises by a factor of 4, so your signal to noise ratio will be better.
If you lift the exposure in post, this will increase the signal and the noise by the same factor four, so the final (lifted) image originally recorded with ISO1000 will have more noise compared to the image that was shot with ISO4000 in the first place.
Take home message: Do not deliberately under expose to avoid the noisy high ISO settings, lifting the exposure in post will give you even more noise.

 

[Loswr]

To sum up what I think of the conversation in regard to this subject. Eric Fossum is correct, Neuro is correct and your inability to look beyond a single point has left you with an incorrect assumption.

That pretty much sums it up. We all know the outcome (and in many cases, the root cause) of an incorrect ASSumption.

It's worth noting that repeating that incorrect assumption over and over and putting it in bold-face type does not make it correct, or even less wrong. The only thing that increases with that useless repetition is exemplified by the first three letters of the word 'assumption'.

 

[rs]

To sum up what I think of the conversation in regard to this subject. Eric Fossum is correct, Neuro is correct and your inability to look beyond a single point has left you with an incorrect assumption.

Speaking of assumptions, here is one, and it goes to my question above.

I assume that most people who deliberately set exposure on a camera typically choose exposure time and aperture based on the subject and their creative intent, and set sensitivity according to the availability of light and their desired brightness. That may be a poor assumption, but I'm self-centered and that's how I shoot.

So say I'm shooting a hummingbird in flight and want to stop motion. I want 1/4000 or faster exposure time. Say I'm using a lens combination which allows me f/5.6 aperture at the maximum, but I want f/8 for a little wider DOF given my distance to subject and a desire to have the entire bird in focus. That right there is the subject/creative intent side of the equation. Next comes light. I can't change or augment ambient - it is what it is. I'm at ISO1000, but a "chimped" glance at the histogram shows that I'm about 2 stops under where I want it to be in the first shot. Accordingly, I change to ISO4000 and shoot again. Can I expect more noise, less noise, or the same noise in the second shot?

You forgot to specify one detail: Do you lift the exposure of the first shot in post in the raw converter or do you use the shots directly as coming from the camera?
If you don't lift the exposure you will have more noise (in ADU) in the final image with higher ISO. But the noise will increase by a factor of less than four (how much less depends on the contribution of downstream read noise compared to upstream read noise and Poisson noise), while you signal (also in ADU) raises by a factor of 4, so your signal to noise ratio will be better.
If you lift the exposure in post, this will increase the signal and the noise by the same factor four, so the final (lifted) image originally recorded with ISO1000 will have more noise compared to the image that was shot with ISO4000 in the first place.
Take home message: Do not deliberately under expose to avoid the noisy high ISO settings, lifting the exposure in post will give you even more noise.

Neuro are not correct and please stick to the topic of noise increases with higher ISO as Neuro claim.
it's just a simple point we discuss where Neuro is wrong, noise Does not Increase with higher iso

Mikael - please stick to the topic of 1Dx mk II resolution estimate, as stated in reply 36 section 3 part 2.

 

[mb66energy]

Inspired by this thread and doing some photography while making a break:

 

[takesome1]

We'll, I'm doubt you needed to go in such detail for the question.

Ordinarily I'd agree, but then came this thread. Without exhaustive detail, I could expect a screenshot of http://www.sensorgen.info/CanonEOS-1DX.html in reply.

Go ahead and address both capture (i.e. sensor level) and final product (i.e. raw file).

The finished product if it were in my workflow the ISO 1000 shot would have 0 noise and 0 data as it would have been deleted when I chimped it. This would be a comparison on recovering an underexposed shot.

 

[scyrene]

This isn't that difficult of a thing to grasp. Set your camera to 1/125s, f/4, ISO 100. Don't move, don't do anything except flip your ISO dial to 200. Which photo has more noise? There is a noise difference between the two and the ONLY thing done was raise ISO. For some reason you can't or won't understand that, and nobody can figure out why.

Oh I have a good idea why. He wants to get one over on Neuro, of whom he seems to have a particular dislike.

A lot of people seem to have trouble admitting they're wrong, they think it's a sign of weakness or something.