[Gooshin]

Hint.. ok. Steal? NOT smart nor clever IMO.

the worlds greatest thieves are the most brilliant of people.

perhaps i chose the wrong verbs, in any case the point is that even schematics and details of OLD technology is being kept secret, and i am simply surprised that nothing ever surfaces to the enthusiast level.

[Pål Jensen]

A 34 Mp FF camera with the DR and noise characteristics similar to a K20D is usable only (in studio) at low ISO and with very expensive lenses. Radu

Are you telling us the K20D is only usable in studio, and at low ISO with very expensive lenses? The FF sensor you mention will have 50% lower noise and 50% less demand from lenses than the K20D sensor.That should be a real recipe for sucess....

[bdery]

Hi, Bdery!

I am interested in your take about using old regular lenses on a over 25 Mo FF sensor. Regardless of their heavy CA and lack of "digital coatings" how do you think this lenses will stand in terms of resolution (center, border, extreme border), distorsion and vignetting - last 2 very model dependant for sure?

Thank you,
Radu

Hi,

you are of course correct that the lens model is important, and no generalization can be made regarding quality. Older lenses were often very well made, however, and their qualities are generally well known.

That being said, older lenses have a flaw, generally. They are not telecentric (wikipedia will help you here if needed). The main difference between a CCD/CMOS and film is that the digital sensor has an angle of acceptance that's much smaller than that of film. Microlenses help with that, but there is a difference. Lenses made for digital are usually telecentric and help with this. So this could lead to a potential loss in resolution.

coatings are also different, and current coatings are, honestly, better than what was dreamed about 10 years ago. But coatings do not really affect resolution, at least not directly.

One thing that's good about APS is that you don't use the border of your lenses, so corner sharpness is not as much an issueas with FF (the "sweet spot" reviewers talk about).

Regarding your initial question : how would oldere lenses perform on new FF sensors? It will depend on the lens, but generally speaking, the answer would be "not as well as with film". Now, this does not mean that the difference will be visible. That's a key point. If the lens was really amazing it might still outperform the sensor.

One thing to consider is that FF, as currently presented on the market, will have a pixel pitch comparable to current APS systems. So the results you get with the current systems will be transferable to FF, but your corners will be worse (the telecentric efffect I mentionned, as well as vigneting, will be the biggest problems).

I have read (but cannot confirm) that good ASA 100 film had a resolution comparable to a 22 MP FF sensor. So you can expect ses to have been targeted at this. More resolution will put more demands on lens. Some ple have comlained that the 14 MP of the K20D is already showing the limits of some lenses previously unseen. regarding area an APs sensor has abouut half the surface of a 35 mm film, so quick and dirty math yields a "28 MP 35 mm sensor equivalence" (that is not a rigorous calculation, but serves the purpose). you could suppose that a FF sensor would behave like the K20D in the center.

Does that help?

Disclaimer : I've studied lens design as part of my studies, but I an NOT a lens designer. I might be overlooking things here, but what I tell you I am confortable with.

P.S. My personal opinion is that if you want FF for wide lenses, you're better off purchasing an APS system with a digital wide lens, than investing a fortune in FF to be able to use older lenses. If you want higher resolution, that's your thing, but many lenses will not be up to the task (some lenses were soft on film too!) If you want larger pixels, you're out of luck, that's not gonna happen...

[Cambo]

Hi,

you are of course correct that the lens model is important, and no generalization can be made regarding quality. Older lenses were often very well made, however, and their qualities are generally well known.

That being said, older lenses have a flaw, generally. They are not telecentric (wikipedia will help you here if needed). The main difference between a CCD/CMOS and film is that the digital sensor has an angle of acceptance that's much smaller than that of film. Microlenses help with that, but there is a difference. Lenses made for digital are usually telecentric and help with this. So this could lead to a potential loss in resolution.

coatings are also different, and current coatings are, honestly, better than what was dreamed about 10 years ago. But coatings do not really affect resolution, at least not directly.

One thing that's good about APS is that you don't use the border of your lenses, so corner sharpness is not as much an issueas with FF (the "sweet spot" reviewers talk about).

Regarding your initial question : how would oldere lenses perform on new FF sensors? It will depend on the lens, but generally speaking, the answer would be "not as well as with film". Now, this does not mean that the difference will be visible. That's a key point. If the lens was really amazing it might still outperform the sensor.

One thing to consider is that FF, as currently presented on the market, will have a pixel pitch comparable to current APS systems. So the results you get with the current systems will be transferable to FF, but your corners will be worse (the telecentric efffect I mentionned, as well as vigneting, will be the biggest problems).

I have read (but cannot confirm) that good ASA 100 film had a resolution comparable to a 22 MP FF sensor. So you can expect ses to have been targeted at this. More resolution will put more demands on lens. Some ple have comlained that the 14 MP of the K20D is already showing the limits of some lenses previously unseen. regarding area an APs sensor has abouut half the surface of a 35 mm film, so quick and dirty math yields a "28 MP 35 mm sensor equivalence" (that is not a rigorous calculation, but serves the purpose). you could suppose that a FF sensor would behave like the K20D in the center.

Does that help?

Disclaimer : I've studied lens design as part of my studies, but I an NOT a lens designer. I might be overlooking things here, but what I tell you I am confortable with.

P.S. My personal opinion is that if you want FF for wide lenses, you're better off purchasing an APS system with a digital wide lens, than investing a fortune in FF to be able to use older lenses. If you want higher resolution, that's your thing, but many lenses will not be up to the task (some lenses were soft on film too!) If you want larger pixels, you're out of luck, that's not gonna happen...

that if it's full frame (in the 35mm sense), that would only be using the central part of the older medium format lenses, right in the sweet spot.

And I'm sure microlens technology has increased to the point where they are able to overcome most of the edge deficiencies inherent in the older designs.

Really looking forward to this mega-camera, whatever it turns out to be.

Surprized Wendy (nee Ned) hasn't got in on this discussion.



Cheers,
Cameron

Hi,

you are of course correct that the lens model is important, and no generalization can be made regarding quality. Older lenses were often very well made, however, and their qualities are generally well known.

That being said, older lenses have a flaw, generally. They are not telecentric (wikipedia will help you here if needed). The main difference between a CCD/CMOS and film is that the digital sensor has an angle of acceptance that's much smaller than that of film. Microlenses help with that, but there is a difference. Lenses made for digital are usually telecentric and help with this. So this could lead to a potential loss in resolution.

coatings are also different, and current coatings are, honestly, better than what was dreamed about 10 years ago. But coatings do not really affect resolution, at least not directly.

One thing that's good about APS is that you don't use the border of your lenses, so corner sharpness is not as much an issueas with FF (the "sweet spot" reviewers talk about).

Regarding your initial question : how would oldere lenses perform on new FF sensors? It will depend on the lens, but generally speaking, the answer would be "not as well as with film". Now, this does not mean that the difference will be visible. That's a key point. If the lens was really amazing it might still outperform the sensor.

One thing to consider is that FF, as currently presented on the market, will have a pixel pitch comparable to current APS systems. So the results you get with the current systems will be transferable to FF, but your corners will be worse (the telecentric efffect I mentionned, as well as vigneting, will be the biggest problems).

I have read (but cannot confirm) that good ASA 100 film had a resolution comparable to a 22 MP FF sensor. So you can expect ses to have been targeted at this. More resolution will put more demands on lens. Some ple have comlained that the 14 MP of the K20D is already showing the limits of some lenses previously unseen. regarding area an APs sensor has abouut half the surface of a 35 mm film, so quick and dirty math yields a "28 MP 35 mm sensor equivalence" (that is not a rigorous calculation, but serves the purpose). you could suppose that a FF sensor would behave like the K20D in the center.

Does that help?

Disclaimer : I've studied lens design as part of my studies, but I an NOT a lens designer. I might be overlooking things here, but what I tell you I am confortable with.

P.S. My personal opinion is that if you want FF for wide lenses, you're better off purchasing an APS system with a digital wide lens, than investing a fortune in FF to be able to use older lenses. If you want higher resolution, that's your thing, but many lenses will not be up to the task (some lenses were soft on film too!) If you want larger pixels, you're out of luck, that's not gonna happen...

Bdery,

I really apreciated your response. I'll be glad to exchange some opinions with you in the future!

Thanks again!
Radu

[jeffkrol]

Jeff, interesting read ... but!

The claim that the photon collecting efficiency per area is the same for 2 um P&S sensor cells and 8 um DSLR cells is too important to be made w/o a source.

I understand that this may be the case. Or that it may not.

All those threads give no source. They are circle referencing themselves. Can you provide a scientific source to back up this claim?

You do not see the problem in the right way. There are many issues with smaller pixels : smaller quantum wells means that the pixels saturate more quickly, have lower signal-to-noise ratios (thus more noise, all thins being equal).

Plus, as pixels get smaller, they approach a size near that of the wavelengths they measure (red is around 0,65 microns, while many current sensors have pixels less than 2 icrons in size). And when your sensor is near the size of your "target" you get problems. It's the same as trying to measure a car's length with another car as your ONLY ruler and reference. You see why that could leads to problem? In this analogy, you could say nothing more than "this car looks to be about the same size as that other car". Not very precise. Same thing with too small pixels.

Your calculations about photons collected are incomplete, and do not make sense to me. Perhaps youcould explain more, but you will never convince me (or any other physicists) that smaller pixels are a good thing for data collection.

To both of you. I can only say that there are some individuals that, from researching their posts, checking biographies, and seeing the company they keep that I trust implicitly. That does not imply that I trust everything they say.
People like John Sheey, GordonBGood, Joseph Wisniewski,Thom Hogan, Illya Borg, the guy that developed the CMOS sensor (he hasn't posted at dpreview for awhile) have a greatr track record and most have EXTREMELY deep backgrounds in the primary subjects.
I'd love to see their private data but would likely understand little of it
I'm sure you can request it from them. Most except for Mr. Sheehy, who keeps a low profile, are more than glad to share. My purpose of posting the links was for you to dig deeper for your own satisfaction. I can't defend their claims, but have yet to see anything that was in error. Writing to Gordon and having him contact John has always been the easiest for me, short of just asking him at dpreview.
Much of the discussion is a bit theoretical and based on "best case" scenarios. Best ADC's Best De-Mosaic algorithems ect.
In my tests, a sensor like the FZ50's with its 1.97u pixel pitch outperforms any equal-sized crop from any DSLR. It's not a question "if" more, smaller pixels are better. The question is when and how they will be implemented. There is nothing wrong with small pixels, per se; the ideal camera is one that records a list of individual photons, their positions striking the focal plane, and their wavelengths. There is no reason to package photons into large bins, except for convenience.

Who knows, maybe Canon will break down and admit that their big-pixel hype is a big fat lie!
Sorry lost the link for the above:
Here is another one though, more for fun.
Luminous Landscape Forum -> D3 High Iso Comparison Shots
Re: Calculation of snr
Some "pictures"
tests Photo Gallery by John Sheehy at pbase.com
Emil Martinec is new to me, not that that means anything:
Noise, Dynamic Range and Bit Depth in Digital SLRs -- page 3
An amusing extrapolation of the analysis of the effect of pixel size results from consideration of digicam raw data. The Panasonic FZ50 is a 10MP superzoom digicam that shoots raw files. An analysis by John Sheehy, as well as the author's rough measurements on this FZ50 raw file using the noise vs. exposure graphical method outlined on page 2, yields a gain g of roughly .29 photons/12-bit ADU at ISO 400. Dividing by the square of the 2 micron pixel size yields .072 photons per ADU per square micron, comfortably in the middle of the efficiency table above. However, just as the Canon ISO calibration was off for its earlier models, digicam ISO calibrations differ from those of DSLR's. Typical DSLR's leave about 3.5 stops between metered middle grey and raw saturation; digicams put middle gray about 0.5-1.0 stop closer to raw saturation, due to their lower dynamic range. This means that their ISO calibration is about 0.5-1.0 stops understated in relation to DSLR ISO calibration, and so the FZ50 efficiency figure is actually higher than .072 by a factor 1.4-2.0, making the FZ50 sensor about the most efficient per unit area in capturing photons of any digital camera sensor currently available! Of course the sensor is only about 5.5mm x 7.3mm in size, so the photon noise referred to the frame size is rather poor, as it is for any digicam; but the photon noise at fixed spatial scale rivals or betters the 1D3, 1Ds3 and D3. If the FZ50 sensor could be scaled up to the size of full frame, it would indeed rival these cameras for photon shot noise performance; and the resolution -- the 2 micron pixels translate into a 216MP (!) full frame camera (of course, for many applications the actual resolution will be limited by diffraction and lens aberrations). It is currently unclear whether this performance could be maintained as the sensor is scaled up by a factor of nearly five in linear dimension -- practicalities of supporting electronics for the sensels, speed in readout and processing, etc, may make it difficult to maintain the FZ50's performance in a scaled-up version.

So a digicam sensor can rival or even exceed the performance of the best current DSLR's in photon capture efficiency, and thus shot noise performance. What about read noise? Naively one might expect that the read noise is a fixed cost per pixel, so more pixels mean more read noise. However, the binning argument above suggests that if the noise cost per pixel remains relatively constant, the overall cost goes down on a per area basis since when pixels are combined to refer the noise to a fixed spatial scale, their noise fluctuations tend to average out. John Sheehy reports a read noise for the FZ50 of 2.7 ADU at ISO 100; my rough measurements on the raw file linked to above yielded 4.5 ADU of read noise. Let's compare this to the Canon 1D3 by referring to fixed scale; the proper scaling is to multiply by the pixel spacing. The 1D3 read noise is 1.2 12-bit ADU at ISO 100, and the pixel spacing is 7.2 microns, so its read noise figure of merit is 1.2 x 7.2 = 8.6; the FZ50 read noise figure of merit is somewhere between 2.7 x 2.0 = 5.4 (Sheehy's measurement) and 4.5 x 2.0 = 9.0 (my measurement). So again the FZ50 is as good or better than the best DSLR's in read noise (fairly compared by referring to fixed spatial scale), at the lowest ISO's. The comparison gets even better if we again factor in the relative calibration of ISO between the two cameras. At high ISO, things are rather different -- the 1D3 read noise at ISO 1600 is 3.4 12-bit ADU, only about a factor of three higher than ISO 100 on a per pixel basis, while the FZ50's is about 16 times higher per pixel according to Sheehy. Now the comparison substantially favors the big-pixel DSLR, even when the figures are referred to fixed spatial scale, and the relative ISO calibration is accounted for.
The above DSLR/digicam comparison outlines the extremes of what may be possible with current or near-term technology, if digicam pixel densities were scaled up to full-frame sensors. The fact that a digicam's performance is in the same ballpark as the best DSLR's when referred to fixed spatial scale, suggests that the problems with noise in digicams is not due to their ever smaller pixels, but rather it is due to their small sensors.

EDIT: Found the CMOS gentleman who's name I forgot above:
Welcome to Dr. Eric R. Fossum's Web Address
Re: Image Sensors past present and future: News Discussion Forum: Digital Photography Review
http://ericfossum.com/Presentations/CMOS%20Image%20Sensors%20Past%20Present%...Jan%202008.pdf

[falconeye]

There are many issues with smaller pixels : smaller quantum wells means that the pixels saturate more quickly, have lower signal-to-noise ratios (thus more noise, all thins being equal).

Plus, as pixels get smaller, they approach a size near that of the wavelengths they measure

@bdery,

welcome on board! Having another physicist in this board is a nice thing for me. I hold a PhD in Theoretical Physics myself

Esp. with respect to Jeff's post #126, we should get out of the way a minor ommission from your statement above:

- You are right, smaller pixels have the issues you describe. But in a first order approximation, these issues cancel out exactly against the better statistics they provide. We have to look at second order effects which you ommitted to do.

- Wavelength. I am not the expert. But from what I understand, pixel sites in the order of 1 micron would be ideal for single photon detection, a prerequisite for max. quantum efficiency. Because you have to avoid coinciding events where two photons interfere when being counted. I know, this is still quite a way to go. But ultimately, this will become a photon counter's sensitive area. Not that I would expect as many pixels in the raw format then


It would be interesting to study more of Jeff's sources in order to understand the second order effects. But I guess, we better consult scientific papers rather than the internet here...


EDIT:
Let me add a comment about max. numbers of pixels to expect...

The best possible optics resolves at the diffraction limit

x = 1.22 lamda (f/d)

Where x is the Airy disk size, i.e., smallest useful pixel size in a RAW format, lambda the light's wavelength and (f/d) the f-stop number.
We shouldn't expect f/1.0 optics to be diffraction-limited Let's assume that f/2 is feasible (I think f/2.8 is more realistic, though). Then 2.5 lambda is the smallest meaningful pixel size in any image raw format. This is 1 micron for blue light. BTW, this is the photoreceptor distance in the human eye

---

Last edited by falconeye; 07-12-2008 at 05:11 AM.

[Gooshin]

you guys are awesome.

[jeffkrol]

@bdery,

welcome on board! Having another physicist in this board is a nice thing for me. I hold a PhD in Theoretical Physics myself




It would be interesting to study more of Jeff's sources in order to understand the second order effects. But I guess, we better consult scientific papers rather than the internet here...

Emil's credentials don't seem too shabby.....
Emil Martinec
I am a Professor of Physics in the Physics Department, the Enrico Fermi Institute, and the College of the University of Chicago and a member of the particle theory group. My research focuses on string theory and particle physics. For online overviews of string theory, see the article in Science , or follow links from google or yahoo . For further information on my research, see my CV, or a research blurb. You can also check out my papers since 1991 from arXiv.org e-Print archive, or all papers from Spires.
Emil Martinec

[falconeye]

Emil's credentials don't seem too shabby..... [...] the particle theory group.

Oh yes, we are numerous

I should say, however, that Emil like myself, seems to do photography as a hobby and we both can be wrong when it comes to the nifty details

bdery is closer to the subject Optics, not necessarily sensor design, though. Speaking for myself, I am now in the software industry. When it comes to single particle detection, experimental particle physicists and astro physicists are the people who are doing it already now. And when it comes to statements about stuff in actual products, don't trust us. You better trust the engineers in here


Having said this, I found two papers of a Stanford group on topic, which are available w/o having to subscribe to a scientific library service:
http://www.invensense.com/shared/pdf/ResolutionandLightSensitivityTradeoffWithPixelSize.pdf
http://isl.stanford.edu/~abbas/group/papers_and_pub/pixelsize.pdf
The second one is more recent.

IEEE Xplore - Login

[bdery]

welcome on board! Having another physicist in this board is a nice thing for me. I hold a PhD in Theoretical Physics myself

Great! Two of us

You are right, smaller pixels have the issues you describe. But in a first order approximation, these issues cancel out exactly against the better statistics they provide. We have to look at second order effects which you ommitted to do.

I didn't think it was the place to do so... there are really many parameters that enter into this (one of the most important, which I didn't mention, being analog/digital conversion).

- Wavelength. I am not the expert. But from what I understand, pixel sites in the order of 1 micron would be ideal for single photon detection, a prerequisite for max. quantum efficiency. Because you have to avoid coinciding events where two photons interfere when being counted. I know, this is still quite a way to go. But ultimately, this will become a photon counter's sensitive area. Not that I would expect as many pixels in the raw format then.

If we used our sensors as photon counting devices (thus, digital) then I would agree, as a photon counting devices is basically an zero vs non zero states comparator. But our sensors are used as analog devices, building up charge with each new photon, so a better well gives you a better SNR.

Then 2.5 lambda is the smallest meaningful pixel size in any image raw format. This is 1 micron for blue light. BTW, this is the photoreceptor distance in the human eye

Human eyes are completely different in the way they work, however, and cannot be compared with sensors Regarding the 2,5 lambda estimation, I agree with the calculation but that would mean 1650 nm (1,65 microns) for red (depending on the "red") and that,s already half as large. And that's a theoretical sensor...

Interesting discussion.

[MikePerham]

Look. Pentax was the ipso facto camera for professionals for two decades in the 67 and 645 format. Yes, the pro's would love to be able to use Pentax glass on a pro spec camera. Pentax has a pedigree in the pro milieu and this would be a great way for them to come back to the fold so to speak.

Ben

Ben, I think you and Pal are correct on this point. Also, I think given the current interest in FF that is the way Pentax will have to go to stay in the game. However, as mentioned in another thread, I don't believe Pentax can try to compete in the FF market and suggest there are 24 million lenses out there.... you will have to offer current technology (digitaly optimized and SDM etc.,) FF lenses along with a FF body.

I sincerly hope they have the manufacturing capacity to do this and continue the DA* APS sized lenses as well as they can't abandon that market either.

[jeffkrol]

Oh yes, we are numerous

I should say, however, that Emil like myself, seems to do photography as a hobby and we both can be wrong when it comes to the nifty details

bdery is closer to the subject Optics, not necessarily sensor design, though. Speaking for myself, I am now in the software industry. When it comes to single particle detection, experimental particle physicists and astro physicists are the people who are doing it already now. And when it comes to statements about stuff in actual products, don't trust us. You better trust the engineers in here


Having said this, I found two papers of a Stanford group on topic, which are available w/o having to subscribe to a scientific library service:
http://www.invensense.com/shared/pdf/ResolutionandLightSensitivityTradeoffWithPixelSize.pdf
http://isl.stanford.edu/~abbas/group/papers_and_pub/pixelsize.pdf
The second one is more recent.

IEEE Xplore - Login

Thanks, unfortunately both seem to downplay the role of microlenses, and lens efficiency....
Oh well, there is always a gap between "theoretical" and "practical"....
As to the whole discussion, like dynamic range, the conclusion to me is that the differences (be it DR, pixel size, noise) are much smaller on the sensor side, and much of the difference is in the "engineering". K20 is a good example. From the "theoretical" point of view it should be a lot noiser than it is. And from a theoretical view, if Pentax used a higher quality ADC it would be even a cleaner sensor.
So much for small vs large pixels.
One of the reasons I like the "fringe experimenters" better is that most of their work does not rely on past workers, which have to be verified as well as the current "hypothesis". One must be careful as to who's house of cards you build yours on
A final Pentax related quote.........
Graystar wrote:
> And for all your mockery, there STILL are no examples posted of small
> pixels performing as well or better than large pixels of equal sensor
> area.


Pentax K10d and K20d have the same area. K20d owners seem very happy by and large, some commenting that they employ a higher ISO sensitivity, routinely, than with the K10d. The "official view" suggests that the reverse should be true.

Of course the two sensors are of different types, different ADCs are used, all kinds of other relevant or irrelevant changes. Two sensor packages with different specifications are going to have different strengths and weaknesses, even if the pixel pitch were the *same*.

Still - so long as the per-pixel / gestalt / upsizing vs downsizing methodology debates continue, it seems impossible to demonstrate ANY parity, in a way that would satisfy absolutely everyone.
RP
Re: Here's your 10+ stops of DR: Open Talk Forum: Digital Photography Review

[kenyee]

K20 is a good example. From the "theoretical" point of view it should be a lot noiser than it is.

IIRC, the reason it works so well is Samsung shrunk the distance *between* photosites instead of the photosites themselves which are close to the K10D's photosite size. Anyone remember the actual numbers?

[Cambo]

An amusing extrapolation of the analysis of the effect of pixel size results from consideration of digicam raw data. The Panasonic FZ50 is a 10MP superzoom digicam that shoots raw files. An analysis by John Sheehy, as well as the author's rough measurements on this FZ50 raw file using the noise vs. exposure graphical method outlined on page 2, yields a gain g of roughly .29 photons/12-bit ADU at ISO 400. Dividing by the square of the 2 micron pixel size yields .072 photons per ADU per square micron, comfortably in the middle of the efficiency table above. However, just as the Canon ISO calibration was off for its earlier models, digicam ISO calibrations differ from those of DSLR's. Typical DSLR's leave about 3.5 stops between metered middle grey and raw saturation; digicams put middle gray about 0.5-1.0 stop closer to raw saturation, due to their lower dynamic range. This means that their ISO calibration is about 0.5-1.0 stops understated in relation to DSLR ISO calibration, and so the FZ50 efficiency figure is actually higher than .072 by a factor 1.4-2.0, making the FZ50 sensor about the most efficient per unit area in capturing photons of any digital camera sensor currently available! Of course the sensor is only about 5.5mm x 7.3mm in size, so the photon noise referred to the frame size is rather poor, as it is for any digicam; but the photon noise at fixed spatial scale rivals or betters the 1D3, 1Ds3 and D3. If the FZ50 sensor could be scaled up to the size of full frame, it would indeed rival these cameras for photon shot noise performance; and the resolution -- the 2 micron pixels translate into a 216MP (!) full frame camera (of course, for many applications the actual resolution will be limited by diffraction and lens aberrations). It is currently unclear whether this performance could be maintained as the sensor is scaled up by a factor of nearly five in linear dimension -- practicalities of supporting electronics for the sensels, speed in readout and processing, etc, may make it difficult to maintain the FZ50's performance in a scaled-up version.

So a digicam sensor can rival or even exceed the performance of the best current DSLR's in photon capture efficiency, and thus shot noise performance. What about read noise? Naively one might expect that the read noise is a fixed cost per pixel, so more pixels mean more read noise. However, the binning argument above suggests that if the noise cost per pixel remains relatively constant, the overall cost goes down on a per area basis since when pixels are combined to refer the noise to a fixed spatial scale, their noise fluctuations tend to average out. John Sheehy reports a read noise for the FZ50 of 2.7 ADU at ISO 100; my rough measurements on the raw file linked to above yielded 4.5 ADU of read noise. Let's compare this to the Canon 1D3 by referring to fixed scale; the proper scaling is to multiply by the pixel spacing. The 1D3 read noise is 1.2 12-bit ADU at ISO 100, and the pixel spacing is 7.2 microns, so its read noise figure of merit is 1.2 x 7.2 = 8.6; the FZ50 read noise figure of merit is somewhere between 2.7 x 2.0 = 5.4 (Sheehy's measurement) and 4.5 x 2.0 = 9.0 (my measurement). So again the FZ50 is as good or better than the best DSLR's in read noise (fairly compared by referring to fixed spatial scale), at the lowest ISO's. The comparison gets even better if we again factor in the relative calibration of ISO between the two cameras. At high ISO, things are rather different -- the 1D3 read noise at ISO 1600 is 3.4 12-bit ADU, only about a factor of three higher than ISO 100 on a per pixel basis, while the FZ50's is about 16 times higher per pixel according to Sheehy. Now the comparison substantially favors the big-pixel DSLR, even when the figures are referred to fixed spatial scale, and the relative ISO calibration is accounted for.
The above DSLR/digicam comparison outlines the extremes of what may be possible with current or near-term technology, if digicam pixel densities were scaled up to full-frame sensors. The fact that a digicam's performance is in the same ballpark as the best DSLR's when referred to fixed spatial scale, suggests that the problems with noise in digicams is not due to their ever smaller pixels, but rather it is due to their small sensors.

And you call youself physicists! Too bad he didn't add any algebra in there to make me feel like a total doob like they usually do... I actually understood most of that.



Well done. Didn't know you guys could speak English!

@bdery,

welcome on board! Having another physicist in this board is a nice thing for me. I hold a PhD in Theoretical Physics myself

>>>snip

Let me add a comment about max. numbers of pixels to expect...

The best possible optics resolves at the diffraction limit

x = 1.22 lamda (f/d)

Where x is the Airy disk size, i.e., smallest useful pixel size in a RAW format, lambda the light's wavelength and (f/d) the f-stop number.
We shouldn't expect f/1.0 optics to be diffraction-limited Let's assume that f/2 is feasible (I think f/2.8 is more realistic, though). Then 2.5 lambda is the smallest meaningful pixel size in any image raw format. This is 1 micron for blue light. BTW, this is the photoreceptor distance in the human eye

Ah, that's better. Personally, I thought it was caused by the hypotunuse of the retina being divided by the square sum of the total inductance in picofarads of the sensor minus the capacitance across the battery terminals.



Ben, how your GI tract this morning? Still glutenous?

Cheers,
Cameron