Equivalence and dynamic range

[bclaff]

It's not a hypothetical exercise. PDR has been measured for over 230 cameras and those are the measurements that matter. You can attempt to correlate it with whatever you like; but that isn't the purpose of the measurement and it doesn't add clarity to what PDR is.

Bill, I don't discount your data and your measurements. However, I have a conundrum and please help me out here.

On one hand your measurements suggest that PDR has no dependence on pixel size. However, on the other hand, as I have said a number of times, that can make sense only if CoC_Areas --> Single_Pixel_Area gains are exactly cancelled by more noisy-ness of smaller pixels. And, that implies a scaling of read noise of pixel areas. However, Prof. Fossum (quoted earlier) and some other experts maintain that read noise may not necessarily scale with pixel size shrinkage.

So there is a disconnect or discrepancy.

How to resolve that?

For one thing it would appear that you might be operating in a range on SNR curve that photon shot noise could tend to be dominant over read noise. And, photon shot noise would scale with pixel size. Perhaps, that possibly offers an outward solution to this situation that makes the conundrum go away. But, does that mean then that PDR would effectively be a measure of photon noise - and little dependence on read noise?

it doesn't add clarity to what PDR is.

If PDR is a more realistic measure than EDR then how are optical nuisances such as flare / glare incorporated? That could be a different line of reasoning but still important regarding the merits of PDR vis a vis what is the relationship of flare / glare to CoC size and how that gets affected by reduction CoC_Area --> Single_Pixel_Area?

It doesn't seem that I am going to be able to lead you out of your confusion.

It is not a confusion per se. It is a requirement imposed by your own equation, which I have written below in a slightly modified form:

(Coc_Area --> Single_pixel_Area_gain) * raw_SNR_per_pixel = 20 PDR_per_pixel.

In order for the RHS of the above equation to stay balanced at 20, any change in Coc_Area --> Single_pixel_Area_gain must be undone by the corresponding change in SNR_per_pixel.

At this stage it appears to me that only two conditions, in isolation or conjunction, would do that:

1. Pixel read noise scales with pixel size. And / or,
2. We are operating in photon shot noise dominated regime.

Experts, such as Prof. Fossum, don't necessarily agree with 1. That leaves us with 2. But, you don't think that 2 is happening. I guess that means that the only choice left is 1. But, then it might put you at odds with experts. I shall let you sort that out with them.

I'm saying this doesn't matter. It's a mind exercise and little else.
The results are the results.
It does appear that pixel noise does approximately scale and even thought the measure is not photon noise dominated it is pretty pixel size independent.
You're also disregarding that, particularly at low ISO settings, PDR is a measurement of downstream noise such as from the Analog to Digital Converter (ADC), and that is independent of pixel size.
This is why on chip ADC and off chip ADC have differently shaped PDR curves .

For example, if you had looked at the PTCs I referenced earlier you'd know that PDR is generally near the intersection of read noise and photon noise; definitely not photon noise dominated.

I think we should revisit 2 above. Your own worked example shows a raw_SNR_per_pixel = 5.64. Is that near 'intersection of read noise and photon noise', in your estimation?

Not my "estimation" but an objective statement given where PDR is on the Photon Transfer Curve (PTC); here it is for the D300 , the camera in that write-up:



The red line indicates where the Read Noise dominated region ends and the Photon Noise region begins; they don't switch instantly, it is a transition.
The black dot, PDR, is not far from that transition.
Note that PDR falls on the blue SNR = 5.64 line (log2 is 2.49).

For fun I added a red dot where DxOMark Landscape score is located.
This is solidly in the Read Noise region and a good reason not to consider it a useful photographic measure.

BTW, we should be celebrating new year instead of arguing about PDR.

I didn't know we were arguing ... so I do agree.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[J A C S]

He just computes the noise curve if the image is downsized to a reference size and takes SNR=20. That does not mean that two sensors with the same PDR would have the same curve. Actually, comparing larger and smaller sensors with close PDR, they would have quite different curves.

 

[Joofa]

It's not a hypothetical exercise. PDR has been measured for over 230 cameras and those are the measurements that matter. You can attempt to correlate it with whatever you like; but that isn't the purpose of the measurement and it doesn't add clarity to what PDR is.

Bill, I don't discount your data and your measurements. However, I have a conundrum and please help me out here.

On one hand your measurements suggest that PDR has no dependence on pixel size. However, on the other hand, as I have said a number of times, that can make sense only if CoC_Areas --> Single_Pixel_Area gains are exactly cancelled by more noisy-ness of smaller pixels. And, that implies a scaling of read noise of pixel areas. However, Prof. Fossum (quoted earlier) and some other experts maintain that read noise may not necessarily scale with pixel size shrinkage.

So there is a disconnect or discrepancy.

How to resolve that?

For one thing it would appear that you might be operating in a range on SNR curve that photon shot noise could tend to be dominant over read noise. And, photon shot noise would scale with pixel size. Perhaps, that possibly offers an outward solution to this situation that makes the conundrum go away. But, does that mean then that PDR would effectively be a measure of photon noise - and little dependence on read noise?

it doesn't add clarity to what PDR is.

If PDR is a more realistic measure than EDR then how are optical nuisances such as flare / glare incorporated? That could be a different line of reasoning but still important regarding the merits of PDR vis a vis what is the relationship of flare / glare to CoC size and how that gets affected by reduction CoC_Area --> Single_Pixel_Area?

It doesn't seem that I am going to be able to lead you out of your confusion.

It is not a confusion per se. It is a requirement imposed by your own equation, which I have written below in a slightly modified form:

(Coc_Area --> Single_pixel_Area_gain) * raw_SNR_per_pixel = 20 PDR_per_pixel.

In order for the RHS of the above equation to stay balanced at 20, any change in Coc_Area --> Single_pixel_Area_gain must be undone by the corresponding change in SNR_per_pixel.

At this stage it appears to me that only two conditions, in isolation or conjunction, would do that:

1. Pixel read noise scales with pixel size. And / or,
2. We are operating in photon shot noise dominated regime.

Experts, such as Prof. Fossum, don't necessarily agree with 1. That leaves us with 2. But, you don't think that 2 is happening. I guess that means that the only choice left is 1. But, then it might put you at odds with experts. I shall let you sort that out with them.

I'm saying this doesn't matter. It's a mind exercise and little else.
The results are the results.
It does appear that pixel noise does approximately scale and even thought the measure is not photon noise dominated it is pretty pixel size independent.
You're also disregarding that, particularly at low ISO settings, PDR is a measurement of downstream noise such as from the Analog to Digital Converter (ADC), and that is independent of pixel size.
This is why on chip ADC and off chip ADC have differently shaped PDR curves .

For example, if you had looked at the PTCs I referenced earlier you'd know that PDR is generally near the intersection of read noise and photon noise; definitely not photon noise dominated.

I think we should revisit 2 above. Your own worked example shows a raw_SNR_per_pixel = 5.64. Is that near 'intersection of read noise and photon noise', in your estimation?

Not my "estimation" but an objective statement given where PDR is on the Photon Transfer Curve (PTC); here it is for the D300 , the camera in that write-up:



The red line indicates where the Read Noise dominated region ends and the Photon Noise region begins; they don't switch instantly, it is a transition.
The black dot, PDR, is not far from that transition.
Note that PDR falls on the blue SNR = 5.64 line (log2 is 2.49).

For fun I added a red dot where DxOMark Landscape score is located.
This is solidly in the Read Noise region and a good reason not to consider it a useful photographic measure.

I like the graph. Lets inject a little more 'objectivity' into the analysis. What follows is just my eye balling a couple of your graphs including the one above, instead of using explicit formulas, so the actual numbers could be slightly off.

From your own estimates PDR is about 9 EV. -- That is photon noise + read noise effect. (And, for comparison EDR from your estimates is about 12 EV.)

Now, it appears that SNR just from photon shot noise is about 6.5. Eye balling from another of your graph it appears that PDR is about 8.4 EV.

I shall let you decide if this number is kind of correct, and if so, does that mean that the PDR here is photon shot noise dominated?

--
Dj Joofa
http://www.djjoofa.com

 

[J A C S]

I wonder exactly the opposite - how much PDR is read noise determined. Since there is no much difference (?) between photon noise for the same format, isn't PDR just a function of the read noise/EDR?

 

[bclaff]

It's not a hypothetical exercise. PDR has been measured for over 230 cameras and those are the measurements that matter. You can attempt to correlate it with whatever you like; but that isn't the purpose of the measurement and it doesn't add clarity to what PDR is.

Bill, I don't discount your data and your measurements. However, I have a conundrum and please help me out here.

On one hand your measurements suggest that PDR has no dependence on pixel size. However, on the other hand, as I have said a number of times, that can make sense only if CoC_Areas --> Single_Pixel_Area gains are exactly cancelled by more noisy-ness of smaller pixels. And, that implies a scaling of read noise of pixel areas. However, Prof. Fossum (quoted earlier) and some other experts maintain that read noise may not necessarily scale with pixel size shrinkage.

So there is a disconnect or discrepancy.

How to resolve that?

For one thing it would appear that you might be operating in a range on SNR curve that photon shot noise could tend to be dominant over read noise. And, photon shot noise would scale with pixel size. Perhaps, that possibly offers an outward solution to this situation that makes the conundrum go away. But, does that mean then that PDR would effectively be a measure of photon noise - and little dependence on read noise?

it doesn't add clarity to what PDR is.

If PDR is a more realistic measure than EDR then how are optical nuisances such as flare / glare incorporated? That could be a different line of reasoning but still important regarding the merits of PDR vis a vis what is the relationship of flare / glare to CoC size and how that gets affected by reduction CoC_Area --> Single_Pixel_Area?

It doesn't seem that I am going to be able to lead you out of your confusion.

It is not a confusion per se. It is a requirement imposed by your own equation, which I have written below in a slightly modified form:

(Coc_Area --> Single_pixel_Area_gain) * raw_SNR_per_pixel = 20 PDR_per_pixel.

In order for the RHS of the above equation to stay balanced at 20, any change in Coc_Area --> Single_pixel_Area_gain must be undone by the corresponding change in SNR_per_pixel.

At this stage it appears to me that only two conditions, in isolation or conjunction, would do that:

1. Pixel read noise scales with pixel size. And / or,
2. We are operating in photon shot noise dominated regime.

Experts, such as Prof. Fossum, don't necessarily agree with 1. That leaves us with 2. But, you don't think that 2 is happening. I guess that means that the only choice left is 1. But, then it might put you at odds with experts. I shall let you sort that out with them.

I'm saying this doesn't matter. It's a mind exercise and little else.
The results are the results.
It does appear that pixel noise does approximately scale and even thought the measure is not photon noise dominated it is pretty pixel size independent.
You're also disregarding that, particularly at low ISO settings, PDR is a measurement of downstream noise such as from the Analog to Digital Converter (ADC), and that is independent of pixel size.
This is why on chip ADC and off chip ADC have differently shaped PDR curves .

For example, if you had looked at the PTCs I referenced earlier you'd know that PDR is generally near the intersection of read noise and photon noise; definitely not photon noise dominated.

I think we should revisit 2 above. Your own worked example shows a raw_SNR_per_pixel = 5.64. Is that near 'intersection of read noise and photon noise', in your estimation?

Not my "estimation" but an objective statement given where PDR is on the Photon Transfer Curve (PTC); here it is for the D300 , the camera in that write-up:



The red line indicates where the Read Noise dominated region ends and the Photon Noise region begins; they don't switch instantly, it is a transition.
The black dot, PDR, is not far from that transition.
Note that PDR falls on the blue SNR = 5.64 line (log2 is 2.49).

For fun I added a red dot where DxOMark Landscape score is located.
This is solidly in the Read Noise region and a good reason not to consider it a useful photographic measure.

I like the graph. Lets inject a little more 'objectivity' into the analysis. What follows is just my eye balling a couple of your graphs including the one above, instead of using explicit formulas, so the actual numbers could be slightly off.

From your own estimates PDR is about 9 EV. -- That is photon noise + read noise effect. (And, for comparison EDR from your estimates is about 12 EV.)

Now, it appears that SNR just from photon shot noise is about 6.5. Eye balling from another of your graph it appears that PDR is about 8.4 EV.

I shall let you decide if this number is kind of correct, and if so, does that mean that the PDR here is photon shot noise dominated?

We can get some pretty good estimates.
For the example given at PDR Signal is about 34.8DN, total noise is about 6.2DN, Read Noise is about 3.6DN, and other (mostly Photon Noise) is about 5.1DN.
Remember sqrt(3.6DN^2+5.1DN^2) = 6.2DN
And log2(34.8DN/6.2DN) = 2.49 (log2 of 5.64)

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[Joofa]

I wonder exactly the opposite - how much PDR is read noise determined. Since there is no much difference (?) between photon noise for the same format, isn't PDR just a function of the read noise/EDR?

The CoC_Area ---> Single_Pixel_Area collapse results in lowering the raw_SNR_per_pixel and the issue is how much of it can be attributed to photon noise.

 

[Joofa]

It is not a confusion per se. It is a requirement imposed by your own equation, which I have written below in a slightly modified form:

(Coc_Area --> Single_pixel_Area_gain) * raw_SNR_per_pixel = 20 PDR_per_pixel.

In order for the RHS of the above equation to stay balanced at 20, any change in Coc_Area --> Single_pixel_Area_gain must be undone by the corresponding change in SNR_per_pixel.

At this stage it appears to me that only two conditions, in isolation or conjunction, would do that:

1. Pixel read noise scales with pixel size. And / or,
2. We are operating in photon shot noise dominated regime.

Experts, such as Prof. Fossum, don't necessarily agree with 1. That leaves us with 2. But, you don't think that 2 is happening. I guess that means that the only choice left is 1. But, then it might put you at odds with experts. I shall let you sort that out with them.

For example, if you had looked at the PTCs I referenced earlier you'd know that PDR is generally near the intersection of read noise and photon noise; definitely not photon noise dominated.

I think we should revisit 2 above. Your own worked example shows a raw_SNR_per_pixel = 5.64. Is that near 'intersection of read noise and photon noise', in your estimation?

Not my "estimation" but an objective statement given where PDR is on the Photon Transfer Curve (PTC); here it is for the D300 , the camera in that write-up:



The red line indicates where the Read Noise dominated region ends and the Photon Noise region begins; they don't switch instantly, it is a transition.
The black dot, PDR, is not far from that transition.
Note that PDR falls on the blue SNR = 5.64 line (log2 is 2.49).

For fun I added a red dot where DxOMark Landscape score is located.
This is solidly in the Read Noise region and a good reason not to consider it a useful photographic measure.

I like the graph. Lets inject a little more 'objectivity' into the analysis. What follows is just my eye balling a couple of your graphs including the one above, instead of using explicit formulas, so the actual numbers could be slightly off.

From your own estimates PDR is about 9 EV. -- That is photon noise + read noise effect. (And, for comparison EDR from your estimates is about 12 EV.)

Now, it appears that SNR just from photon shot noise is about 6.5. Eye balling from another of your graph it appears that PDR is about 8.4 EV.

I shall let you decide if this number is kind of correct, and if so, does that mean that the PDR here is photon shot noise dominated?

We can get some pretty good estimates.
For the example given at PDR Signal is about 34.8DN, total noise is about 6.2DN, Read Noise is about 3.6DN, and other (mostly Photon Noise) is about 5.1DN.
Remember sqrt(3.6DN^2+5.1DN^2) = 6.2DN
And log2(34.8DN/6.2DN) = 2.49 (log2 of 5.64)

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

It seems to appear that the PDRs calculated with photon shot noise + read noise and just photon shot noise are relatively close to each other.

Also, using your own numbers, the numbers which actually go in the calculation of SNR (5.1, 6.2) -- i.e., photon shot noise = 5.1 and total noise ( read noise + shot noise) = 6.2, seem closer to each other than (3.6, 5.1).

Based upon the above facts, is it safe for one to say that a major source of raw_SNR_per_pixel scaling is photonic - resulting in PDR being relatively independent of pixel size?

--
Dj Joofa
http://www.djjoofa.com

 

[bclaff]

...

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

It seems to appear that the PDRs calculated with photon shot noise + read noise and just photon shot noise are relatively close to each other.

Also, using your own numbers, the numbers which actually go in the calculation of SNR (5.1, 6.2) -- i.e., photon shot noise = 5.1 and total noise ( read noise + shot noise) = 6.2, seem closer to each other than (3.6, 5.1).

Based upon the above facts, is it safe for one to say that a major source of raw_SNR_per_pixel scaling is photonic - resulting in PDR being relatively independent of pixel size?

No.
First you should compare the 5.1DN to the 3.6DN; not to the 6.2DN.
Second you should not make such a broad statement based on one of over 230 cameras.


In any case. It doesn't matter to me.
As I have already said, my 10+ years of experience tells me PDR is pretty insensitive to pixel size.

 

[Joofa]

...

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

It seems to appear that the PDRs calculated with photon shot noise + read noise and just photon shot noise are relatively close to each other.

Also, using your own numbers, the numbers which actually go in the calculation of SNR (5.1, 6.2) -- i.e., photon shot noise = 5.1 and total noise ( read noise + shot noise) = 6.2, seem closer to each other than (3.6, 5.1).

Based upon the above facts, is it safe for one to say that a major source of raw_SNR_per_pixel scaling is photonic - resulting in PDR being relatively independent of pixel size?

No.
First you should compare the 5.1DN to the 3.6DN; not to the 6.2DN.

IMHO, 6.2 and 5.1 should be compared to each other (and not 5.1 and 3.6) as they go in SNR calculations in two cases being considered - photon noise + read noise together, and just photon noise only.

In any case, you have not answered that why PDRs calculated with and without photo noise are close to each other?

Second you should not make such a broad statement based on one of over 230 cameras.

I'm just working your example.

In any case. It doesn't matter to me.
As I have already said, my 10+ years of experience tells me PDR is pretty insensitive to pixel size.

And, we are trying to figure out why that is so.

--
Dj Joofa
http://www.djjoofa.com

 

[bclaff]

...

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

It seems to appear that the PDRs calculated with photon shot noise + read noise and just photon shot noise are relatively close to each other.

Also, using your own numbers, the numbers which actually go in the calculation of SNR (5.1, 6.2) -- i.e., photon shot noise = 5.1 and total noise ( read noise + shot noise) = 6.2, seem closer to each other than (3.6, 5.1).

Based upon the above facts, is it safe for one to say that a major source of raw_SNR_per_pixel scaling is photonic - resulting in PDR being relatively independent of pixel size?

No.
First you should compare the 5.1DN to the 3.6DN; not to the 6.2DN.

IMHO, 6.2 and 5.1 should be compared to each other (and not 5.1 and 3.6) as they go in SNR calculations in two cases being considered - photon noise + read noise together, and just photon noise only.

I disagree.

In any case, you have not answered convincingly that why PDRs calculated with and without photo noise are close to each other?

Huh? PDR is only measured one way.

Second you should not make such a broad statement based on one of over 230 cameras.

I'm just working your example.

Still a bad idea to generalize.

In any case. It doesn't matter to me.
As I have already said, my 10+ years of experience tells me PDR is pretty insensitive to pixel size.

And, we are trying to figure out why that is so.

No we, you; I accept that this is what I see.

--
Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

 

[Joofa]

...

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

It seems to appear that the PDRs calculated with photon shot noise + read noise and just photon shot noise are relatively close to each other.

Also, using your own numbers, the numbers which actually go in the calculation of SNR (5.1, 6.2) -- i.e., photon shot noise = 5.1 and total noise ( read noise + shot noise) = 6.2, seem closer to each other than (3.6, 5.1).

Based upon the above facts, is it safe for one to say that a major source of raw_SNR_per_pixel scaling is photonic - resulting in PDR being relatively independent of pixel size?

No.
First you should compare the 5.1DN to the 3.6DN; not to the 6.2DN.

IMHO, 6.2 and 5.1 should be compared to each other (and not 5.1 and 3.6) as they go in SNR calculations in two cases being considered - photon noise + read noise together, and just photon noise only.

I disagree.

Kindly show me how you calculate SNR with photon noise only, and with photon noise + read noise together, then.

In any case, what is more important, as said a number of times before, is that PDRs with photon noise only and photon noise + read noise are suspiciously close to each other.

In any case, you have not answered convincingly that why PDRs calculated with and without photo noise are close to each other?

Huh? PDR is only measured one way.

Is that because that enables you to make certain claims.

Second you should not make such a broad statement based on one of over 230 cameras.

I'm just working your example.

Still a bad idea to generalize.

You have the data. Why don't you go over it and see how PDRs with and without photon noises are holding up against each other for all the sensors?

In any case. It doesn't matter to me.
As I have already said, my 10+ years of experience tells me PDR is pretty insensitive to pixel size.

And, we are trying to figure out why that is so.

No we, you; I accept that this is what I see.

You are seeing what you want to see, I guess.

--
Dj Joofa
http://www.djjoofa.com

 

[bclaff]

...

So I don't think 5.1DN photon noise is very much compared to 3.6DN read noise.
Remember, the axes are logarithmic; so we're not very far at all into the Photon Noise region.

It seems to appear that the PDRs calculated with photon shot noise + read noise and just photon shot noise are relatively close to each other.

Also, using your own numbers, the numbers which actually go in the calculation of SNR (5.1, 6.2) -- i.e., photon shot noise = 5.1 and total noise ( read noise + shot noise) = 6.2, seem closer to each other than (3.6, 5.1).

Based upon the above facts, is it safe for one to say that a major source of raw_SNR_per_pixel scaling is photonic - resulting in PDR being relatively independent of pixel size?

No.
First you should compare the 5.1DN to the 3.6DN; not to the 6.2DN.

IMHO, 6.2 and 5.1 should be compared to each other (and not 5.1 and 3.6) as they go in SNR calculations in two cases being considered - photon noise + read noise together, and just photon noise only.

I disagree.

Kindly show me how you calculate SNR with photon noise only, and with photon noise + read noise together, then.

In any case, what is more important, as said a number of times before, is that PDRs with photon noise only and photon noise + read noise are suspiciously close to each other.

PDR is only measured with as an SNR threshold where Noise is all noise sources; so there is no such thing as "photon noise only" PDR.

This discussion has run it's course; I have other priorities so this is my last post.

 

[Iuvenis]

Well, this discussion certainly took some twists and turns, albeit with some lingering uncertainty over the effect of shining lasers on smartphones.

Thanks everyone for their contributions. I was of course taking Bill's PDR as a starting point, as my musings were based on looking at his graphs.

In summary, it seems to me that dynamic range (or PDR if you prefer) is too dependent on sensor technology for it to be incorporated into any discussions of equivalence.

However, I still think it is worth bearing in mind that when cameras of different sensor size have equivalent settings, even larger sensor cameras with excellent dynamic range may well lose that advantage as a result of the higher iso setting (or post processing) required to maintain the same brightness in the final image.

In fact, having played around more with the graphs, it seems extremely uncommon for a larger sensor camera to have a dynamic range advantage at all in that scenario, whatever the technical reasons for that may be.

Of course, when light is abundant, or long exposures are an option, larger sensor cameras will have a DR advantage, as smaller sensor cameras rarely have lower iso settings than larger sensor cameras (though I note that smartphones normally do, so that is maybe something for smaller sensor camera manufacturers to consider)

 

[Great Bustard]

Well, this discussion certainly took some twists and turns, albeit with some lingering uncertainty over the effect of shining lasers on smartphones.

Thanks everyone for their contributions. I was of course taking Bill's PDR as a starting point, as my musings were based on looking at his graphs.

In summary, it seems to me that dynamic range (or PDR if you prefer) is too dependent on sensor technology for it to be incorporated into any discussions of equivalence.

However, I still think it is worth bearing in mind that when cameras of different sensor size have equivalent settings, even larger sensor cameras with excellent dynamic range may well lose that advantage as a result of the higher iso setting (or post processing) required to maintain the same brightness in the final image.

Actually, the higher ISO setting will help, since higher ISO settings typically have lower electronic noise. What hurts is the larger pixel count which results in greater electronic noise for a given proportion of the photo.

On the other hand, more pixels, while intrinsically more noisy (as a general rule), result in more detail, and this detail can allow stronger noise filtering to be used for a given level of detail, the end result of which is a lower noise floor and thus greater DR.

In fact, having played around more with the graphs, it seems extremely uncommon for a larger sensor camera to have a dynamic range advantage at all in that scenario, whatever the technical reasons for that may be.

Not knowing exactly how BC computes his PDR, I cannot answer. What I can say is that *if* a FF sensor and mFT sensor had the same pixel count and each FF pixel at two stops higher ISO had the same electronic noise as each mFT pixel, then the DR would be identical for equivalent photos.

Of course, when light is abundant, or long exposures are an option, larger sensor cameras will have a DR advantage, as smaller sensor cameras rarely have lower iso settings than larger sensor cameras (though I note that smartphones normally do, so that is maybe something for smaller sensor camera manufacturers to consider)

Smartphones, while having lower ISO options, do not come close to FF ISO equivalent settings.

 

[bclaff]

...

In summary, it seems to me that dynamic range (or PDR if you prefer) is too dependent on sensor technology for it to be incorporated into any discussions of equivalence.

...

Interesting. I would think that any measure of dynamic range would need to be adjusted for "equivalence"; provided "equivalence" is meaningful for the intended photographic situation.

 

[AiryDiscus]

In the sciences, you should generally demonstrate not only your technique, but also how it compares to reasonably related techniques. PDR w/ // w/o photon noise should be compared to figure out which is superior. Otherwise, how do we know PDR is worth paying attention to? Or its specific formulation is ideal?

 

[Iuvenis]

Well, this discussion certainly took some twists and turns, albeit with some lingering uncertainty over the effect of shining lasers on smartphones.

Thanks everyone for their contributions. I was of course taking Bill's PDR as a starting point, as my musings were based on looking at his graphs.

In summary, it seems to me that dynamic range (or PDR if you prefer) is too dependent on sensor technology for it to be incorporated into any discussions of equivalence.

However, I still think it is worth bearing in mind that when cameras of different sensor size have equivalent settings, even larger sensor cameras with excellent dynamic range may well lose that advantage as a result of the higher iso setting (or post processing) required to maintain the same brightness in the final image.

Actually, the higher ISO setting will help, since higher ISO settings typically have lower electronic noise. What hurts is the larger pixel count which results in greater electronic noise for a given proportion of the photo.

On the other hand, more pixels, while intrinsically more noisy (as a general rule), result in more detail, and this detail can allow stronger noise filtering to be used for a given level of detail, the end result of which is a lower noise floor and thus greater DR.

Using Bill's figures for PDR, pixel count has little impact. I think he said that was his experience earlier in the thread. You can play around with the graphs and see the same, however.

My understanding was that this was because PDR is influenced by photon noise as well as electronic noise, and photon noise is equalised at equivalent settings. Accordingly, even though electronic noise may not be equalised at those settings, it is often the case (as demonstrated by the graphs) that dynamic range does end up similar at equivalent settings.

In fact, having played around more with the graphs, it seems extremely uncommon for a larger sensor camera to have a dynamic range advantage at all in that scenario, whatever the technical reasons for that may be.

Not knowing exactly how BC computes his PDR, I cannot answer. What I can say is that *if* a FF sensor and mFT sensor had the same pixel count and each FF pixel at two stops higher ISO had the same electronic noise as each mFT pixel, then the DR would be identical for equivalent photos.

Of course, when light is abundant, or long exposures are an option, larger sensor cameras will have a DR advantage, as smaller sensor cameras rarely have lower iso settings than larger sensor cameras (though I note that smartphones normally do, so that is maybe something for smaller sensor camera manufacturers to consider)

Smartphones, while having lower ISO options, do not come close to FF ISO equivalent settings.

Yes, I didn't mean to say that they came close to FF equivalent settings. But they are closer in dynamic range than they would be if their base iso was 100.

 

[Great Bustard]

Well, this discussion certainly took some twists and turns, albeit with some lingering uncertainty over the effect of shining lasers on smartphones.

Thanks everyone for their contributions. I was of course taking Bill's PDR as a starting point, as my musings were based on looking at his graphs.

In summary, it seems to me that dynamic range (or PDR if you prefer) is too dependent on sensor technology for it to be incorporated into any discussions of equivalence.

However, I still think it is worth bearing in mind that when cameras of different sensor size have equivalent settings, even larger sensor cameras with excellent dynamic range may well lose that advantage as a result of the higher iso setting (or post processing) required to maintain the same brightness in the final image.

Actually, the higher ISO setting will help, since higher ISO settings typically have lower electronic noise. What hurts is the larger pixel count which results in greater electronic noise for a given proportion of the photo.

On the other hand, more pixels, while intrinsically more noisy (as a general rule), result in more detail, and this detail can allow stronger noise filtering to be used for a given level of detail, the end result of which is a lower noise floor and thus greater DR.

Using Bill's figures for PDR, pixel count has little impact. I think he said that was his experience earlier in the thread. You can play around with the graphs and see the same, however.

If pixel count has little impact, that means either that the noise floor used for his PDR calculation is dominated by photon noise and/or the aggregate electronic noise over the area he computes the PDR doesn't vary by much as a function of pixel count, on average.

My understanding was that this was because PDR is influenced by photon noise as well as electronic noise, and photon noise is equalised at equivalent settings. Accordingly, even though electronic noise may not be equalised at those settings, it is often the case (as demonstrated by the graphs) that dynamic range does end up similar at equivalent settings.

I wish BC would spell out how he computes the noise floor, saturation limit, and area over which he computes the PDR. Furthermore, to be a useful measure in cross-format comparisons, then he would have to have the same exposure (light per area on the sensor) for each corresponding ISO setting.

For example, if he used f/4 (t/4.5) 1/100 ISO 200 on one camera to compute the PDR at ISO 200, then, for a meaningful cross-format comparison, he would have needed to use f/4 (t/4.5) 1/100 ISO 200 with the same scene luminance on all cameras (or any other settings that give the same exposure at the same ISO setting).

In fact, having played around more with the graphs, it seems extremely uncommon for a larger sensor camera to have a dynamic range advantage at all in that scenario, whatever the technical reasons for that may be.

Not knowing exactly how BC computes his PDR, I cannot answer. What I can say is that *if* a FF sensor and mFT sensor had the same pixel count and each FF pixel at two stops higher ISO had the same electronic noise as each mFT pixel, then the DR would be identical for equivalent photos.

Of course, when light is abundant, or long exposures are an option, larger sensor cameras will have a DR advantage, as smaller sensor cameras rarely have lower iso settings than larger sensor cameras (though I note that smartphones normally do, so that is maybe something for smaller sensor camera manufacturers to consider)

Smartphones, while having lower ISO options, do not come close to FF ISO equivalent settings.

Yes, I didn't mean to say that they came close to FF equivalent settings. But they are closer in dynamic range than they would be if their base iso was 100.

Honestly, not knowing how BC does his PDR computations, I can't say one way or another. It would be very much appreciated if he spelled it out clearly in a way that all could understand and used a method that made sense with respect to cross-format comparisons.