 |
 | Home | Contact Us | FAQ | Search & Site Map | Link to Us |
 | Sign In | Join | Other 45 Sites in Network |
| PhotoKB HomeDiscussion Groups Digital Photography Digital PhotoDSLR CamerasZLR CamerasPoint & Shoot CamerasFilm Photography 35 mmLarge FormatMedium formatDarkroomFilm and LabsOther EquipmentPhoto Technique Nature PhotographyPeople PhotographyTechnique GeneralGeneral Photo Topics General TopicsAustralian PhotographyUK PhotographyDirectoryPhoto Clubs |
Photo Forum / Digital Photography / DSLR Cameras / February 2006Dinosaurs
The large form factor of the sauropod was eventually discarded in favor of models that were more efficient at energy exchange, locomotion and reproduction. There are still lizards but they are not the dominant terrestrial quadripeds. Sic semper film. Hopefully sic semper the ancient 35mm SLR form factor, the slavish devotion to the idea that a larger imaging sensor is necessarily better and the many other knuckleheaded concepts constantly paraded through this newsgroup. Evolution may proceed imperceptibly but the extremely intelligent designers of digital imaging equipment have the ability to move far beyond ancient paradigms of camera design but for the inability to move the market away from anicient, preconceived notions of what constitutes a high quality photographic instrument. I hope I live long enough to see the digital revolution play out in the directions it ought to go. >The large form factor of the sauropod was eventually discarded in favor of >models that were more efficient at energy exchange, locomotion and [quoted text clipped - 12 lines] >I hope I live long enough to see the digital revolution play out in the >directions it ought to go. A large 8MPix sensor can always have less noise than a small 8MPix sensor; for equal noise a larger sensor can always have more resolution than a small sensor. A large lens can always have less resolution degradation due to defecation than a small lens; precession in manufacturing large versus small lenses. Or do you think that the laws of physics will be magically inverted?  SignatureIan G8ILZ >>The large form factor of the sauropod was eventually discarded in favor of >>models that were more efficient at energy exchange, locomotion and [quoted text clipped - 22 lines] > small lenses. Or do you think that the laws of physics will be magically > inverted? Ian, I think you miss some of the point here. An 8MP sensor in the smaller size tomorrow will not have the same noise problems as todays smaller size sensors, good enough also comes into the equation somewhat. Yes the larger sensor is usually more desirable but economics and the rules of diminishing returns also dictates that many people will be satisfied with the cheaper, smaller sensor because it will be good enough for the job, have a look at the current 12MP cameras such as the 5D and the D200 or D2X, the smaller sensor gives away little and in some areas exceeds the performance of the larger sensor, for quite a large group the smaller sensor is far more desireable. Grey has many shades. Take to a logically conclusion, eventual we will have infinitely small sensors with no noise! Then we can all just have camera phones! > Take to a logically conclusion, eventual we will have infinitely small > sensors with no noise! Then we can all just have camera phones! That no one has small enough hamds to use?? Some people like big cars, some like small. Some people need big cars, some need small. Big cars have advantages over small cars, small cars have advantages over big cars. The entire world would not be suited by a single car, yet so many photographers seem to perceive that 'big' is better than 'small' or 'small is better than 'big'. NO they're simply DIFFERENT. Get over it...and argue which religion is right! The argument will be just as fruitless!!! > Some people like big cars, some like small. Some people need big cars, > some need small. [quoted text clipped - 5 lines] > it...and argue which religion is right! The argument will be just as > fruitless!!! Man will occasionally stumble over the truth, but most of the time he will pick himself up and continue on. -- Winston Churchill >Some people like big cars, some like small. Some people need big cars, >some need small. [quoted text clipped - 5 lines] >it...and argue which religion is right! The argument will be just as >fruitless!!! Analogies are sometimes useful to demonstrate how something works, but they are useless as a proof that it is in fact how it works. Whatever, the above analogy has a serious fault. It does *not* match up two things that work in a similar way! It is invalid. Now, if it was focal length of the lense rather than the size of the sensor, it would be dead on. Long lenses are useful to some, short lenses are useful to others. Just like big cars and little cars. However, if the analogy to relate how we view sensor size, it isn't big or small *cars* that we need to compare to, but big or small penises. Everyone claims bigger ones are better, but in fact the smaller ones accomplish the job well enough, except for just a few specialized needs. Hence, if you are one of those high end "professionals" with a need that only the big ones can satisfy, go for a Canon! You might be just getting screwed though, while the rest of us are working on the next generation! SignatureFloyd L. Davidson http://www.apaflo.com/floyd_davidson Ukpeagvik (Barrow, Alaska) floyd@apaflo.com >>Some people like big cars, some like small. Some people need big cars, >>some need small. [quoted text clipped - 28 lines] > might be just getting screwed though, while the rest of us are > working on the next generation! Yes! > Some people like big cars, some like small. Some people need big cars, > some need small. [quoted text clipped - 5 lines] > it...and argue which religion is right! The argument will be just as > fruitless!!! Agree! >>>The large form factor of the sauropod was eventually discarded in favor of >>>models that were more efficient at energy exchange, locomotion and [quoted text clipped - 33 lines] >larger sensor, for quite a large group the smaller sensor is far more >desireable. The 5D is a full frame sensor, the D200 is ~1mm larger than APS-C and has worse noise than the 5D; it should be compared to the Canon 20D. >Grey has many shades. But less shades with a small sensor than a large one. SignatureIan G8ILZ >>>>The large form factor of the sauropod was eventually discarded in favor >>>>of [quoted text clipped - 41 lines] > The 5D is a full frame sensor, the D200 is ~1mm larger than APS-C and has > worse noise than the 5D; it should be compared to the Canon 20D. LOL, the difference is negligable. Compared to the 20D, okay!! >>Grey has many shades. > > But less shades with a small sensor than a large one. How many do you need? >>>The large form factor of the sauropod was eventually discarded in favor >>>of [quoted text clipped - 38 lines] > > Grey has many shades. The same technology that may lower noise in a small 8mp sensor can be used to lower noise in a large 8mp sensor, too. Nes pas? There is no area that the smaller 12mp sensor exceeds the performance of the larger, in real terms. SignatureSkip Middleton http://www.shadowcatcherimagery.com >> Ian, >> [quoted text clipped - 13 lines] > The same technology that may lower noise in a small 8mp sensor can be used > to lower noise in a large 8mp sensor, too. Nes pas? Of course, see above, the smaller sensor may still give me the performance that I need from smaller lenses, smaller body, cheaper to produce, etc. > There is no area that the smaller 12mp sensor exceeds the performance of > the larger, in real terms. Maybe but in some areas the difference is negligable. You have made your choice, you need wide in one shot, cool, I can do the same in two shots when I need too. >>>Ian, >>> [quoted text clipped - 16 lines] > Of course, see above, the smaller sensor may still give me the performance > that I need from smaller lenses, smaller body, cheaper to produce, etc. The problem with this whole thread is that physics is being ignored. Sensor noise in better digital cameras, both DSLRs and P&S, are already photon noise limited at their quantum efficiency (QE). QE's run >20%, typically 30%. So there is only a factor of 3 to 5 improvement that is possible. Basically noise in digital cameras are limited by photon statistics, and the number of photons incident on the Earth from the sun is finite. Think of a photon "rain" and the pixels are buckets left out in the rain. Make the buckets smaller, and you get less rain drops. For example, a canon 10D DSLR acquires about 7900 photons on an 18% gray card, iso100, for a typical exposure, with its 7.4 micron pixels. Increase the pixel count, e.g. double it in the same format size, and the pixel size must drop to 3.7 microns, and the photons per pixel drops by 4x, to 1975 photons. Decrease the pixel size by another factor of 2 to 1.85 microns, and the photon count drops to 494 photons. Noise is the square root of the number of photons counted, so the noise with only 494 photons is 22, for a signal to noise ratio of only 22 compared to the 10D DSLR signal to noise on the same target with the same exposure time and f/ratio lens is 88. The small pixels lose detail in the shadows. Note too the dynamic range decreases too. With a very good 3-electron read noise, the 494 photons on 18% gray give a dynamic range of 494/(3*0.18) = 915. But the DSLR has a dynamic range of 7900*(3/.18)= 14,600! >>There is no area that the smaller 12mp sensor exceeds the performance of >>the larger, in real terms. > > Maybe but in some areas the difference is negligable. You have made your > choice, you need wide in one shot, cool, I can do the same in two shots when > I need too. For bright subjects, the smaller pixels and lower pixel counts are not always a problem. That is why the small pixel P&S cameras do as well as they do. But small pixels is also why P&S cameras have reduced dynamic range and higher noise. It is unlikely that the physics will change. References: The Signal-to-Noise of Digital Camera images and Comparison to Film http://www.clarkvision.com/imagedetail/digital.photons.and.qe Digital Cameras: Does Pixel Size Matter? Factors in Choosing a Digital Camera http://www.clarkvision.com/imagedetail/does.pixel.size.matter Digital Cameras: Counting Photons, Photometry, and Quantum Efficiency http://www.clarkvision.com/imagedetail/digital.photons.and.qe Roger > The problem with this whole thread is that physics is being ignored. > Sensor noise in better digital cameras, both DSLRs and P&S, are already [quoted text clipped - 3 lines] > by photon statistics, and the number of photons incident on the > Earth from the sun is finite. ... > For example, a canon 10D DSLR acquires about 7900 photons on an > 18% gray card, iso100, for a typical exposure, with its 7.4 micron pixels. [quoted text clipped - 3 lines] > another factor of 2 to 1.85 microns, and the photon count drops > to 494 photons. There is another good reason why nobody is going to rush a 1.85 micron pixel camera to market, which is that 1.85 micron pixels critically sample the diffraction limit of visible light at f/3.3 or so. By f/4 the diffraction limit would already be more than two pixels across, so the small pixels aren't buying you any more detail. (IOW, this is just another example of physics being ignored.) Most 35mm/APS camera lenses are not of diffraction limited quality at f/4 anyway, so pixels that small would outstrip the image the lens can deliver. But this is kind of a straw man example - I could maybe see a use for 4 micron pixels in 18x24mm or 24x36mm (27 or 54 MP), but 2 micron pixels would be excessive. If you need that much detail, you need a larger detector, larger than either 35mm or APS. All other things being equal, larger detectors are better. The problem is that all other things usually aren't equal. It's a matter of personal choice AFAICT. > Noise is the square root of the number of photons > counted, so the noise with only 494 photons is 22, for a signal [quoted text clipped - 4 lines] > the 494 photons on 18% gray give a dynamic range of 494/(3*0.18) = > 915. But the DSLR has a dynamic range of 7900*(3/.18)= 14,600! This calculation was a little confusing (and I did read the photon counting link below). I think you're saying that with 7900 photons detected from 18% reflectance, pure white 100% is at 7900/0.18 = 43,900 photons. Assuming the camera sets this perfectly to maximum response, and the minimum noise level is the 3 electron read noise, then you calculate the dynamic range as 43,900/3 = 14,600. However, for practical purposes I don't think a device with a 12 bit A/D converter can record a dynamic range of 14,600 (since 2^12 = 4096). For a linear conversion of photons to camera DN, the dynamic range should be limited to ~4096; with this many photons the gain is such that the minimum level is the digitization noise, not the read noise. Although the DSLR still has a dynamic range advantage in this example it is 4096/915, not 14,600/915. If I take the number you computed of 113 photons/DN at ISO 100, the dynamic range is actually only 43,900/113 = 390. This seems low. Is there a mixup between gain in (electrons/16-bit-DN) and (electrons/12-bit-DN) somewhere? Anyway, it seems to me that you've shown that for the SLRs, gain at ISO 100 is high enough that the read noise is undersampled. > Digital Cameras: Counting Photons, Photometry, > and Quantum Efficiency > http://www.clarkvision.com/imagedetail/digital.photons.and.qe >>The problem with this whole thread is that physics is being ignored. >>Sensor noise in better digital cameras, both DSLRs and P&S, are already [quoted text clipped - 24 lines] > pixels would be excessive. If you need that much detail, you need > a larger detector, larger than either 35mm or APS. Yes, I agree. My "does pixel size matter?" page discusses diffraction limits. > All other things being equal, larger detectors are better. The problem > is that all other things usually aren't equal. It's a matter of [quoted text clipped - 22 lines] > Although the DSLR still has a dynamic range advantage in this > example it is 4096/915, not 14,600/915. The numbers I gave were the sensor specifications. Of course 12-bit digitization limits the dynamic range to 4095. I predict future cameras will use 14 or 16-bit A/Ds (hopefully Canon will announce one this month at the annual photo show). > If I take the number you computed of 113 photons/DN at ISO 100, the > dynamic range is actually only 43,900/113 = 390. This seems low. > Is there a mixup between gain in (electrons/16-bit-DN) and > (electrons/12-bit-DN) somewhere? Anyway, it seems to me that you've > shown that for the SLRs, gain at ISO 100 is high enough that the read > noise is undersampled. The 113 photons/DN are the photons incident on the device. The transmission of the Bayer filter, T, and the quantum efficiency, QE, of the sensor, T*QE =0.095 so of those 113 photons, only 113 * 0.095 = 10.7 photons/DN are detected. The full well capacity of the Canon 10D is ~44,200 electrons, so the dynamic range at ISO 100 is 44200/10.7 ~ 4100, or within roundoff, 4095. Yes, at ISO 100 the Canon 10D noise and dynamic range is limited by the 12-bit A/D. So are other Canon DSLRs. I'm hoping new Canon DSLRs will have 14 or 16-bit A/Ds. Roger >> Digital Cameras: Counting Photons, Photometry, >> and Quantum Efficiency >> http://www.clarkvision.com/imagedetail/digital.photons.and.qe > For example, a canon 10D DSLR acquires about 7900 photons on an > 18% gray card, iso100, for a typical exposure, with its 7.4 micron pixels. > Increase the pixel count, e.g. double it in the same format size, > and the pixel size must drop to 3.7 microns, and the photons per > pixel drops by 4x, to 1975 photons. Ummm. You might want to check your math and a few other assumptions... starting with the assumption that doubling the number of pixels necessarily halves their size. Your model also ignores changes in pixel sensitivity, etc. The bottom line is that while sensor sizes for different cameras have remained basically unchanged for some time now, each new model manages to put more pixels on those same size sensors with less noise than the model before it. You are arguing that it is impossible for the manufacturers to do what we see they are doing every day. Sure, you cannot change the laws of physics, but you cannot change the facts, either. >>For example, a canon 10D DSLR acquires about 7900 photons on an >>18% gray card, iso100, for a typical exposure, with its 7.4 micron pixels. [quoted text clipped - 14 lines] > Sure, you cannot change the laws of physics, but you cannot change the > facts, either. If you read the links and references to many other studies, you would find the data shows the physical limits have been reached. This has only occurred in the last couple of years, so if you are comparing older cameras, what you are observing is correct, and what I am saying is correct. You will not see that same trend in the future, and in fact comparing the trend in cameras over the last year shows the noise is proportional to pixel size. Newer cameras also use microlenses to make the fill factor effectively 100%. The only room left for improvement is to improve quantum efficiency, lower read noise, and lower dark current. Read noise and dark current are negligible in high light conditions (pretty much all but night photography). Roger > >>For example, a canon 10D DSLR acquires about 7900 photons on an > >>18% gray card, iso100, for a typical exposure, with its 7.4 micron pixels. [quoted text clipped - 22 lines] > in the future, and in fact comparing the trend in cameras over > the last year shows the noise is proportional to pixel size. Gobbledygook. The Canon 20D is less noisy than the 10D. The Nikon D2x and the D200 are less noisy than the D70 and D100. In fact, some reviews have the D200 as being less noisy than the Canon 5D. "And yet, it moves." And yet, the manufacturers continue to produce cameras where each generation has lower noise than the previous one, despite the fact that they have the same size sensor. I am not saying there is anything wrong with your reasoning, but when the demonstrated facts don't match your conclusions then it is time to check your assumptions. >>>>For example, a canon 10D DSLR acquires about 7900 photons on an >>>>18% gray card, iso100, for a typical exposure, with its 7.4 micron pixels. [quoted text clipped - 26 lines] > and the D200 are less noisy than the D70 and D100. In fact, some > reviews have the D200 as being less noisy than the Canon 5D. These are small effects. For example, canon improved the micro lenses with the 20D to make an effective 100% fill factor. The read noise of the 20D is lower than the 10D, but that does not effect sensitivity or signal to noise in high signal regions. The full well of the 20D is about 51000 electrons for a maximum signal-to-noise of 225. The full well of the 10D is about 44000 electrons for a maximum signal to noise of 210. At the low photon end, the 10D has about 10 electron read noise and the 20D about 7 electrons. The 20Da has about 3. Once you reach 100% fill factor, the only way to improve maximum signal to noise is to improve quantum efficiency. I did say there is room for quantum efficiency improvements of 3 to 5x. But all good cameras are already photon noise limited for their quantum efficiency. If you don't want to believe that, that is your problem. But if you would take the time to read the research you would see it is correct. > "And yet, it moves." > [quoted text clipped - 5 lines] > the demonstrated facts don't match your conclusions then it is time to > check your assumptions. They are not assumptions. It is hard data on sensor performance. Try reading. Roger > > I am not saying there is anything wrong with your reasoning, but when > > the demonstrated facts don't match your conclusions then it is time to > > check your assumptions. > > They are not assumptions. It is hard data on sensor performance. > Try reading. Not hard data at all -- yet another list of unproven theories with almost no supporting data. Those articles fail to explain how manufacturers keep coming out with lower noise cameras having more pixels on the same size sensors. Like you, they deny the facts because those facts don't fit their theories. Neither do those articles show how those theories apply to real-world photography. It is great to talk about photon rain and draw a graph, but the graphs make the same mistakes I pointed out earlier: they assume that sensors will not become more efficient and they assume that a smaller sensor size means a smaller pixel size. Neither of those assumptions are valid. In fact, the author of "Does Pixel Size Matter" seems to be confused. Nowhere does he tell us pixel size; he only talks about pixel spacing and then uses pixel size and pixel spacing interchangeably, which they are not. You can talk about quantum efficiency all you want. Those articles throw the term around a lot. I suspect that the author of those articles has about as clear an understanding of quantum physics as the Ramtha group. Great for making New Age movies. Poor for understanding physics. You claim that you are an image tester, yet it has been shown over and over that you have not got the faintest idea of how your theories apply to real photographs. Great, if all you want to shoot are test patterns and the inside of lens caps. Poor, if you want to take pictures in a rain forest. If you are really an image tester, you know that digital noise has been reduced despite increasing the number of pixels. Yet you ignore that to promote a pet theory without any hard data that supports it. A little honesty would be greatly appreciated here. If you really have data that shows the quantum limit has been reached, lets see it. >>>I am not saying there is anything wrong with your reasoning, but when >>>the demonstrated facts don't match your conclusions then it is time to [quoted text clipped - 8 lines] > pixels on the same size sensors. Like you, they deny the facts because > those facts don't fit their theories. If you read and check references at the end of the article, you would find many other studies that point to the same conclusion. The data are clear and tell the correct story. You and Stacey, on the other hand sound like those film fanatics in the film versus digital wars. You wave your arms and say it isn't so and this or that proves me wrong. Well, show a real study that shows real data that proves your point. Why don't you take data from dpreview.com and plot it up and see if you can actually prove your point? > Neither do those articles show how those theories apply to real-world > photography. It is great to talk about photon rain and draw a graph, [quoted text clipped - 5 lines] > about pixel spacing and then uses pixel size and pixel spacing > interchangeably, which they are not. The manufacturers use microlenses over the pixels to make effectively 100% fill factors. Then effectively pixel pitch = pixel size. With CMOS, actual active area of a pixel may be only 25%, so with microlenese, actual pixel size is irrelevant. > You can talk about quantum efficiency all you want. Those articles > throw the term around a lot. I suspect that the author of those > articles has about as clear an understanding of quantum physics as the > Ramtha group. Great for making New Age movies. Poor for understanding > physics. All you have to do is follow the references and you will see many other studies that come to the same conclusions. E.g. see: http://www.astrosurf.org/buil/20d/20dvs10d.htm http://spiff.rit.edu/classes/phys559/lectures/gain/gain.html http://www.britastro.org/vss/ccdtable.html http://www.kodak.com/global/plugins/acrobat/en/digital/ccd/papersArticles/Photog raphyWithAn11-megapixel35mmFormatCCD.pdf See figure 1 for quantum efficiencies of sensors: http://huhepl.harvard.edu/~LSST/general/Janesick_paper_2003.pdf Then a simple google search will show you more sites and manufacturer's data sheets. > You claim that you are an image tester, yet it has been shown over and > over that you have not got the faintest idea of how your theories apply [quoted text clipped - 4 lines] > ignore that to promote a pet theory without any hard data that supports > it. Just what I expected: attack personal integrity if you don't like the data. Here is my bio: http://www.clarkvision.com/rnc and my publications: http://www.clarkvision.com/rnc/publist.html Many of the publications evaluate sensors, including NASA spacecraft. So what are your qualifications to evaluate quantum effects? I apply the technical knowledge of sensors to produce better photographs. My photos are at: http://www.clarkvision.com and include international contest winners. I currently have photos hanging in galleries. Do you? Where are your photos? > A little honesty would be greatly appreciated here. If you really have > data that shows the quantum limit has been reached, lets see it. All of the above links were referenced from my pages, and prove the point. How about a reference that shows real data that supports your contention? I think what you and Stacey are seeing is that manufactures have improved their electronics over the last 5 or so years to get their cameras up to the photon limits the CCDs and CMOS sensors had all along. They added microlenses to make the fill factors 100%. The sensors were always capable of producing at the photon noise limit, and have been so for years before consumer digital cameras came out. You also seem to ignore the fact that I said that quantum efficiencies of the sensors in consumer cameras could be improved by 3 to 5x, so there is room for improvement. Roger > > A little honesty would be greatly appreciated here. If you really have > > data that shows the quantum limit has been reached, lets see it. [quoted text clipped - 4 lines] > How about a reference that shows real data that supports > your contention? What contention? I am asking how the manufacturers are producing cameras with more pixels with less noise without increasing the size of the sensor, despite your saying that they cannot. But then, you say later on: > I think what you and Stacey are seeing is that manufactures have > improved their electronics over the last 5 or so years to get [quoted text clipped - 7 lines] > of the sensors in consumer cameras could be improved by 3 to 5x, > so there is room for improvement. So you say that there is room for improvement. Perhaps I misunderstood you when you said that the large sensors of today are necessarily better than the small ones because the limit of quantum efficiency had been reached. I hope you will understand that when someone starts throwing around the equations of light fall-off from Arcturus in order to prove something about photons striking pixels that my baloney meter is likely to peg. Irrelevant techno-babble followed by sweeping statements without supporting data does not sit well with me. Neither does it convince me if someone must resort to falling back on credentials instead of clearly explaining the concept. I will allow that you are a better photographer than I am; I have never claimed to be anything better than a terrible amateur, so I will take your word for it. It is not much of an accomplishment to be a better photographer than I am, but please, go ahead and bask in that glory all you like. :-) Neither do I have the expertise or the time to create a web site of my own. I do post a few photos on http://photography-cafe.com/ from time to time if you are all that interested. >>>A little honesty would be greatly appreciated here. If you really have >>>data that shows the quantum limit has been reached, lets see it. [quoted text clipped - 8 lines] > cameras with more pixels with less noise without increasing the size of > the sensor, despite your saying that they cannot. See: http://www.clarkvision.com/imagedetail/digital.signal.to.noise See figures 1, 2, tables 1, 2, 3, 4, 5. The square root dependence of the plot proves photon noise (Poisson statistics) limit has been reached. There are many other references at the end of the page that show the same results. Then see: http://www.astrosurf.org/buil/20d/20dvs10d.htm The plots at the end of the page prove the D70, 10D and 20D are photon noise limited (the straight line proves that). >But then, you say > later on: [quoted text clipped - 15 lines] > better than the small ones because the limit of quantum efficiency had > been reached. My first post in this thread read: "The problem with this whole thread is that physics is being ignored. Sensor noise in better digital cameras, both DSLRs and P&S, are already photon noise limited at their quantum efficiency (QE). QE's run >20%, typically 30%. So there is only a factor of 3 to 5 improvement that is possible. Basically noise in digital cameras are limited by photon statistics, and the number of photons incident on the Earth from the sun is finite." > I hope you will understand that when someone starts > throwing around the equations of light fall-off from Arcturus in order > to prove something about photons striking pixels that my baloney meter > is likely to peg. It was Alpha Lyra, and it is the reference standard used in Astronomy. It ia a super calibrated light source that is free for all to use. You have made the assertion: "And yet, the manufacturers continue to produce cameras where each generation has lower noise than the previous one, despite the fact that they have the same size sensor." What data proves this? While manufacturers have improved read noise, the high signal regime is controlled by how many photons can be collected, and this is basically controlled by pixel size and fill factor. Roger > I am not saying there is anything wrong with your reasoning, but when > the demonstrated facts don't match your conclusions then it is time to > check your assumptions. Nah, he's too convinced this can't change and technology can't improve. There can't be anyone smarter than he is that can find a solution.. SignatureStacey >>I am not saying there is anything wrong with your reasoning, but when >>the demonstrated facts don't match your conclusions then it is time to >>check your assumptions. > > Nah, he's too convinced this can't change and technology can't improve. > There can't be anyone smarter than he is that can find a solution.. More shoot the messenger and name calling. Show me a reference that says you can get better than 100% quantum efficiency. Do you know what quantum efficiency is? http://en.wikipedia.org/wiki/Quantum_efficiency "the percentage of photons hitting the photoreactive surface that will produce an electron-hole pair. It is an accurate measurement of the device's sensitivity." I suppose you can create photons out of thin air in your world. But the real world we live on is illuminated by the sun, which sends a limited (read not infinite) number of photons per square meter to the earth's surface. (If it was a higher number of photons, we would all be cooked.) It is simple college freshman physics, perhaps even high school physics, to compute the number of photons received from our sun and assuming perfect detectors (100% QE) one can see where the ultimate limits really are. You can also look up the definition of ISO and how it relates to the incident photon flux. If you did this, you would see that current digital cameras are within a factor of about 10 of being perfect, whether they be top end DSLRs or good P&S cameras, assuming you could get 100% optics transmission, 100% color filter transmission, 100% microlens transmission, 100% infrared filter transmission, and 100% quantum efficiency. That sets your ultimate improvement, unless you can invent photons in your world. ;-) Roger >>>I am not saying there is anything wrong with your reasoning, but when >>>the demonstrated facts don't match your conclusions then it is time to [quoted text clipped - 35 lines] > >Roger How does the quantum efficiency of the CMOS sensors used in consumer DSLRs compared with back-thinned monochromatic sensors which have about 90% efficiency? -Rich > How does the quantum efficiency of the CMOS sensors used > in consumer DSLRs compared with back-thinned monochromatic > sensors which have about 90% efficiency? > -Rich CMOS seems to be in the 20 to 30% range. See Figure 1 in this paper: http://huhepl.harvard.edu/~LSST/general/Janesick_paper_2003.pdf It shows 8 different sensors. Roger >>>I am not saying there is anything wrong with your reasoning, but when >>>the demonstrated facts don't match your conclusions then it is time to [quoted text clipped - 4 lines] >> > More shoot the messenger and name calling. No tired of you acting like you are the last word on this subject. > Show me a reference that says you can get better than 100% > quantum efficiency. Do you know what quantum efficiency is? Do you understand what unit sensitivity is? > http://en.wikipedia.org/wiki/Quantum_efficiency > "the percentage of photons hitting the photoreactive > surface that will produce an electron-hole pair. It is an > accurate measurement of the device's sensitivity." So? Just like you can turn up the ISO on the present sensors, what is stopping them from increasing the sensitivity even further and'or reducing the noise at present levels? They've already been doing this and I doubt it's going to stop. According to your logic, there is no way to push B&W film to 6400ISO and beyond because of "Limited photons" but people do it . I agree there is a limit to the MP on a given sensor size but I don't buy present densities can't be improved. > If you did this, you would see that current digital cameras are > within a factor of about 10 of being perfect, By your math.. You've already been called on a couple of mistakes. What if they come up with a sensor with no microlenses and remove that loss? Or a 'good' sensor that doesn't need color filters or a infrared filter? You assume that todays technology is all they will be able to come up with. Looking at the past of how electronics are, I'm sure not going to bet on that! Then again I guess you would have made predictions on what could be done when vac tubes were in use and there were no transistors! SignatureStacey > No tired of you acting like you are the last word on this subject. Then why don't you just killefile me? >>Show me a reference that says you can get better than 100% >>quantum efficiency. Do you know what quantum efficiency is? > Do you understand what unit sensitivity is? You didn't answer the question. >>http://en.wikipedia.org/wiki/Quantum_efficiency >>"the percentage of photons hitting the photoreactive [quoted text clipped - 8 lines] > I agree there is a limit to the MP on a given sensor size but I don't buy > present densities can't be improved. ISO has nothing to do with quantum efficiency. Your statements also indicate you do not understand photon statistics. The noise in a photon stream is the square root of the number of photons detected. If you electronics are good enough then your sensor is photon noise limited. That is true whether you collect 100 photons (noise =10, S/N = 10) at 10% QE or 1000 photons (noise =32, S/N = 32) at 100% QE. You can up the gain as much as you want, e.g. ISO 50,000. Noise will be either dominated by electronics, or photon statistics. Photons statistics is the best you can do. Sensors collect a maximum number of photons as converted electrons given by the full-well of the device. Typical full wells in large pixel DSLRs are in the 50,000 pixel range, and gets full at around ISO 100. At ISO 3200, the range of electrons digitized is 50,000/32 ~ 0 to 1560. At ISO 100 the maximum S/N = 50,000/sqrt(50,000) = 223. With only 1560 electrons, the maximum S/N = 1560/sqrt(1560) = 39. One could easily design more gain, e.g. ISO 6400 and digitize 50000/64 ~ 0 to 780 electrons, giving a maximum S/N - 780/sqrt(780) = 28. But now increase the QE from, say 30% to 90%. Now that same ISO 6400 gain will get 780 * 90/30 = 2340 electrons, and the S/N increases to 48. Regardless, there is lower S/N as you go up in ISO, just as observed in digital cameras images. So you see, counting photons and quantum efficiency is the key. QE sets sensitivity and S/N achievable in an an image acquisition. You can't do better than photon statistics. >>If you did this, you would see that current digital cameras are >>within a factor of about 10 of being perfect, > By your math.. You've already been called on a couple of mistakes. By many people's math. See, for example data for the 10D, 20D and D70: http://www.astrosurf.org/buil/20d/20dvs10d.htm My results agree with others, and there is a consistent picture: current good digital cameras are photon noise limited. > What if they come up with a sensor with no microlenses and remove that loss? Assuming all optical surfaces are multicoated, each surface loses about 1%, so 2% loss for front and rear surface. Big deal. > Or a 'good' sensor that doesn't need color filters or a infrared filter? You still need to synthesize the color response function of the eye, so you can't simply ignore color, unless you want black and white. > You assume that todays technology is all they will be able to come up with. No, I never said that. In fact I said just that opposite: that improvements of 3 to 5x are possible. The fact that a photon limit has been reached only means that further improvements in electronics will not help the high end S/N does not mean another thing could not be improved, and that is QE. > Looking at the past of how electronics are, I'm sure not going to bet on > that! Then again I guess you would have made predictions on what could be > done when vac tubes were in use and there were no transistors! See, there you go again: making personal attacks. Try discussing facts and data. If you don't agree with something presented, show data that proves you point. Personal attacks only bring you down. Roger >> I am not saying there is anything wrong with your reasoning, but when >> the demonstrated facts don't match your conclusions then it is time to >> check your assumptions. > >Nah, he's too convinced this can't change and technology can't improve. >There can't be anyone smarter than he is that can find a solution.. You can believe that sensors will improve significantly. But discussions about believes are rarely fruitful. I think there are areas where sensor may improve besides QE and read noise. But it requires explicitly and accurately stating your assumptions to have a meaningful discussion. Otherwise, we can just as well assume that 200 mpg cars will be build in the future. SignatureThat was it. Done. The faulty Monk was turned out into the desert where it could believe what it liked, including the idea that it had been hard done by. It was allowed to keep its horse, since horses were so cheap to make. -- Douglas Adams in Dirk Gently's Holistic Detective Agency > If you read the links and references to many other studies, you > would find the data shows the physical limits have been reached. [quoted text clipped - 6 lines] > only room left for improvement is to improve quantum efficiency, > lower read noise, and lower dark current. Higher well capacity is also possible: there's no reason to suppose the current capacity limits will be maintained forever. In particular, there's no reason in principle why more layers couldn't be fabricated beneath the sensor array to hold more charge. And if you did that you could increase pixel density and therefore resolution while maintaining SNR. Andrew. > Higher well capacity is also possible: there's no reason to suppose > the current capacity limits will be maintained forever. [quoted text clipped - 5 lines] > > Andrew. I agree that well capacities could be improved. But that is not the ultimate limit. The ultimate limit is the number of photons coming from the sun (or other light sources which are all less intense than our sun). See my other post this morning. At f/8, a sunny bright day with a perfect optical system (100% transmission) would deliver about 910,000 photons/second per square micron in the focal plane in the green passband (the green response of your eye) from an 18% gray card with the sun straight overhead and the lens pointing straight down at the gray card (not a very interesting photo). If you use ISO 100 speed and the sunny f/16 rule of 1/400 second, you get only about 2300 photons/pixel per square micron (a 1 micron pixel would have signal to noise ratio of 47. Current real cameras lose light due to reflection and absorption in the optics and lower quantum efficiency, and get only about 230 photons/pixel per square micron for a signal to noise ratio of 15. For interesting photos with shadows and the sun not straight overhead, the photon flux drops, factors of 10 or more. Even so, yes, there is room for improvement, but there are real and easily computable upper limits due to the finite photon flux from the sun. But obviously, with small pixels, one gets fewer photons, and a larger pixel will collect more. It is up to the user to decide how large a pixel they want (or can afford). see: Digital Cameras: Counting Photons, Photometry, and Quantum Efficiency http://www.clarkvision.com/imagedetail/digital.photons.and.qe Roger >> Higher well capacity is also possible: there's no reason to suppose >> the current capacity limits will be maintained forever. In >> particular, there's no reason in principle of a photo sensor, and >> hy more layers couldn't be fabricated beneath the sensor array to >> hold more charge. And if you did that you could increase pixel >> density and therefore resolution while maintaining SNR. > I agree that well capacities could be improved. But that is not > the ultimate limit. The ultimate limit is the number of > photons coming from the sun (or other light sources which > are all less intense than our sun). Of course, but that's a totally different limit. If you can increase the well capacity, and I'm sure you can, then the limits you implied go away. And there's nothing at all to prevent you capturing more light with a larger lens and focusing it on that high-capacity sensor. > At f/8, a sunny bright day with a perfect optical system (100% transmission) > would deliver about 910,000 photons/second per square micron in the > focal plane There's nothing to stop you, if you wish, focusing the image from a metre-wide lens on a sensor a few millimetres wide. Apart from the fact that the sensor would melt, of course. Andrew. > Of course, but that's a totally different limit. If you can increase > the well capacity, and I'm sure you can, then the limits you implied [quoted text clipped - 9 lines] > metre-wide lens on a sensor a few millimetres wide. Apart from the > fact that the sensor would melt, of course. Any lens, no matter how big, pointing at a given scene and operating at f/8 will cast the same surface brightness of photons per time per area onto the focal plane. This is why when you meter a scene at f/8, 1/125 second at ISO 100, the meter doesn't need to know whether you are using an 8x10 camera, 35mm, or an ISO 100 electronic sensor. If you use a bigger (longer focal length) lens at f/8, more light enters the lens but the longer f.l. spreads the light from a given patch of the scene over a wider area on the sensor. If you want more photons from that piece of the scene to fall into a pixel, you need either a faster lens, or a bigger pixel (which then implies a longer lens to keep the same field of view per pixel). I don't have a dog in the DSLR format fight. From a neutral point of view, Roger's points about physical limitations on CCD pixel size, signal-to-noise and quantum efficiency are correct. Manufacturers will continue to make improvements, of course, but I expect the rate of increase to be smaller than it has been. Just as with film, at some point the way to increase the level of detail is to go to a larger sensor[*]. That doesn't mean one should always choose a larger sensor. In film-speak, 6x7 gives more detail than 35mm but there are perfectly good reasons why 35mm was used more commonly than 6x7 for most applications, where 35mm was good enough. The same could be true of 18x24 and 24x36 in the long run. The ratio between these two formats is about the same as between 6x4.5 and 6x7 (both "medium format"), it's less than the difference between 35mm and 6x4.5. So the whole fight is not really about that much. [*] There is an exception to the advantage of large formats, which is that if your limit is set by signal-to-noise rather than film MTF, and it's easier to make faster lenses for the smaller format, the advantage of the larger format is lessened, although there are other tradeoffs. >> Of course, but that's a totally different limit. If you can increase >> the well capacity, and I'm sure you can, then the limits you implied [quoted text clipped - 9 lines] >> metre-wide lens on a sensor a few millimetres wide. Apart from the >> fact that the sensor would melt, of course. > Any lens, no matter how big, pointing at a given scene and operating > at f/8 will cast the same surface brightness of photons per time per > area onto the focal plane. Of course, but that's to define away the problem. For example, let's say you have a 50mm lens that throws an image on a 35mm sensor. You can construct an optical relay that focuses all that light on a sensor half the diameter. The image on the smaller sensor is exactly the same as the image on the 35mm sensor, just smaller and brighter. (Ignoring, for the sake of the thought experiment, losses in the relay.) The notion that there is some physical limit to the surface brightness at the focal plane only applies if you assume relative aperture is fixed. But relative aperture is not fixed because you can make an image circle smaller without reducing the physical diameter of the aperture. Andrew. >>>Of course, but that's a totally different limit. If you can increase >>>the well capacity, and I'm sure you can, then the limits you implied [quoted text clipped - 30 lines] > > Andrew. But if you keep the same spatial resolution, assuming a perfect lens, then the brightness per pixel stays the same. Roger >>>> "Roger N. Clark (change username to rnclark)" <username@qwest.net> >>>> wrote: [quoted text clipped - 36 lines] > lens, then the brightness per pixel stays the same. > Roger I must correct this. I'm surprised someone didn't jump sooner about this. The rate of arrival of photons per unit area per unit time is proportional to the square of the f-ratio. In other words, if you keep f/ratio constant, then the photons per unit area in the focal plane is constant. The problem is that if you scale a camera down, say 2x, the aperture drops by 2x, the focal length drops by 2x (to give the same field of view), and the pixel size drops by 2x (to give the same spatial resolution on the subject). It should be obvious by this point that per unit time the aperture has collected only 1/4 the number of photons. Also, the smaller pixels each collect 1/4 less photons since their area is divided by 4 to keep spatial resolution constant. Another way to look at the problem is aperture collects light, the focal length spreads out the light, and the pixels are buckets that collect the light in the focal plane. But the total number of photons delivered to the focal plane is ONLY dependent on aperture (ignoring transmission losses of the optics). Back to the camera example: scale a camera down by 2x keeping f/ratio and spatial resolution constant. You lose 4x the photons entering the lens with the smaller camera, and since you you must use 2x smaller pixels, the area is 4x less, so you lose 4x more photons/pixel. Thus photons delivered to a pixel for a given resolution on the subject goes as the 4th power of the aperture (and camera size)! Decreasing your camera by 2x means 16x less photons per pixel! And this is what we observe with small cameras: their smaller sensors have smaller full well capacities, that get filled for a given exposure time with a lower number of photons. That in turn means higher noise because there are fewer photons. Check out this web page for more info on this subject: http://home.earthlink.net/~stanleymm/f_ratio_myth.htm Perhaps this should be a new thread? I'm going to post this in rec.photo.digital. Roger > The rate of arrival of photons per unit area per unit time is > proportional to the square of the f-ratio. In other words, if you [quoted text clipped - 36 lines] > Perhaps this should be a new thread? I'm going to post > this in rec.photo.digital. Scott W over in my thread in rec.photo.digital pointed out an error: Scott, You are correct. I forgot to include that for the smaller lens at half the focal length, the area on the subject doubles, canceling one of the squared terms. So, yes, I agree that the relation scales as the square not as the 4th power. So halving the camera size means pixels get 1/4 the photons. A good example is the Canon 20D with 6.4 micron pixels and a maximum signal at ISO 100 of 50,000 electrons, compared to the Canon S60 with 2.8 micron pixels with a maximum signal of about 11,000 electrons at ISO 100. The pixel size is 6.4^2 / 2.8^2 = 5.2x scaling, similar to the 50000/11000 = 4.5 scaling of maximum recorded signal. Then, for photon noise limited systems, signal-to-noise ratio scales as the square root of the camera size. Roger > Then, for photon noise limited systems, signal-to-noise ratio > scales as the square root of the camera size. Oops--signal-to-noise ratio scales linearly not square root with camera size. Roger >>>>> "Roger N. Clark (change username to rnclark)" <username@qwest.net> >>>>> wrote: [quoted text clipped - 30 lines] >>> image circle smaller without reducing the physical diameter of the >>> aperture. > The rate of arrival of photons per unit area per unit time is > proportional to the square of the f-ratio. In other words, if you > keep f/ratio constant, then the photons per unit area in the focal > plane is constant. > The problem is that if you scale a camera down, say 2x, the aperture > drops by 2x, the focal length drops by 2x (to give the same field of [quoted text clipped - 4 lines] > since their area is divided by 4 to keep spatial resolution > constant. When sensor sizes get smaller there is no need to keep the f/ratio constant. So don't do that! If you can make a 50mm f/8 lens with a 40mm image circle, you can make a 25mm f/4 lens with a 20mm image circle. > Another way to look at the problem is aperture collects light, the > focal length spreads out the light, and the pixels are buckets that > collect the light in the focal plane. But the total number of photons > delivered to the focal plane is ONLY dependent on aperture > (ignoring transmission losses of the optics). Of course, but the smaller sensor is collecting exactly the same number of photons as the large one. That's as long as you keep the diameter of the aperture constant, not the f/ratio. > Back to the camera example: scale a camera down by 2x keeping > f/ratio and spatial resolution constant. No, that's not the idea. You're the one saying keep the f/ratio constant, and I'm saying no, don't keep the f/ratio constant. There's no need. After all, how many medium format cameras have f/1.8 lenses? How many large format cameras have f/4 lenses? Andrew. > When sensor sizes get smaller there is no need to keep the f/ratio > constant. So don't do that! If you can make a 50mm f/8 lens with a > 40mm image circle, you can make a 25mm f/4 lens with a 20mm image > circle. > Of course, but the smaller sensor is collecting exactly the same > number of photons as the large one. That's as long as you keep the > diameter of the aperture constant, not the f/ratio. > No, that's not the idea. You're the one saying keep the f/ratio > constant, and I'm saying no, don't keep the f/ratio constant. There's > no need. After all, how many medium format cameras have f/1.8 lenses? > How many large format cameras have f/4 lenses? But 35mm lenses do come fast. Like a 50mm comes in f/1.8, f/1.4, and f/1.2. There has even been an f/0.95. Are you saying you could do an f/0.47 and still have performance? At the long end, like 500 mm f/4 on 35mm on a 1/2 size camera would need a 250 mm f/2. Not cheap. If you keep your aperture constant, you are not scaling the lens size much (same diameter, a little shorter in length), but you are making more demands on the optical system, and may actually drive up cost without saving much bulk and weight. That was not the point of going smaller. Roger >> When sensor sizes get smaller there is no need to keep the f/ratio >> constant. So don't do that! If you can make a 50mm f/8 lens with a >> 40mm image circle, you can make a 25mm f/4 lens with a 20mm image >> circle. >> Of course, but the smaller sensor is collecting exactly the same >> number of photons as the large one. That's as long as you keep the >> diameter of the aperture constant, not the f/ratio. >> No, that's not the idea. You're the one saying keep the f/ratio >> constant, and I'm saying no, don't keep the f/ratio constant. There's >> no need. After all, how many medium format cameras have f/1.8 lenses? >> How many large format cameras have f/4 lenses? > But 35mm lenses do come fast. Like a 50mm comes in f/1.8, f/1.4, > and f/1.2. There has even been an f/0.95. Are you saying you > could do an f/0.47 and still have performance? I imagine the performance of the f/0.95 lens didn't have much performance wide open! But yes, that's precisely what I'm saying. Take, say, a typical 50mm f/1.8 lens with a 40mm image circle, put an optical relay behind it, and you have a 25mm f/0.9 lens with a 20mm image circle. This will lose quality because of the extra elements that make up the relay, true. However, if you were to design for a 20mm image circle from the get go it'd be much better. > At the long end, like 500 mm f/4 on 35mm on a 1/2 size camera > would need a 250 mm f/2. Not cheap. But not necessarily much more expensive than the 500 mm f/4. > If you keep your aperture constant, you are not scaling the lens size > much (same diameter, a little shorter in length), but you are > making more demands on the optical system, and may actually drive > up cost without saving much bulk and weight. > That was not the point of going smaller. That was not the point of my answer, either. Your claim was that high quality inherently requires a large sensor due to fundamental physics. But it doesn't: it requires big lenses that collect a lot of light, and this is largely independent of sensor size. Andrew. >>>When sensor sizes get smaller there is no need to keep the f/ratio >>>constant. So don't do that! If you can make a 50mm f/8 lens with a [quoted text clipped - 39 lines] > But it doesn't: it requires big lenses that collect a lot of light, > and this is largely independent of sensor size. But you are advocating halving the f/ratio while at the same time, to maintain equivalent spatial resolution, you would need to DOUBLE the spatial resolution of the lens. A 50mm f/1.4 lens is alread not that sharp, and you want to make a 25mm f/0.7 lens that has double the resolution? That will certainly drive up costs, if it is possible at any price! Another example: Canon's 600 mm f/4 L IS is $7000 at B&H, and the 300 f/2.8 L IS is $3850. But it is not f/2, and it is only similar in sharpness, NOT DOUBLE. Your goal is unrealistic to impossible and may actually cost more. There is, however, a small regime of focal lengths and apertures where small can perform quite well (not necessarily better). Roger >>>If you keep your aperture constant, you are not scaling the lens size >>>much (same diameter, a little shorter in length), but you are [quoted text clipped - 6 lines] >> But it doesn't: it requires big lenses that collect a lot of light, >> and this is largely independent of sensor size. > But you are advocating halving the f/ratio while at the same time, > to maintain equivalent spatial resolution, you would need to > DOUBLE the spatial resolution of the lens. While halving the diameter of the image circle, yes. This is much easier than trying to double the spatial resolution over the same image circle. After all, I'm not trying to get any more information out of the lens: I'm just focusing on a smaller sensor. Of course there are engineering challenges to doing this sort of thing! > A 50mm f/1.4 lens is alread not that sharp, and you want to make a > 25mm f/0.7 lens that has double the resolution? That will certainly > drive up costs, if it is possible at any price! > Another example: Canon's 600 mm f/4 L IS is $7000 at B&H, > and the 300 f/2.8 L IS is $3850. But it is not f/2, and it is > only similar in sharpness, NOT DOUBLE. That is a totally different problem. Andrew. >>But you are advocating halving the f/ratio while at the same time, >>to maintain equivalent spatial resolution, you would need to >>DOUBLE the spatial resolution of the lens. > While halving the diameter of the image circle, yes. This is much > easier than trying to double the spatial resolution over the same > image circle. After all, I'm not trying to get any more information > out of the lens: I'm just focusing on a smaller sensor. Of course > there are engineering challenges to doing this sort of thing! I think the engineering challenge is the key. If what you say is true, then one could keep scaling, not just 2x. For example, scale the canon 5D 35.8 x 23.9 mm sensor with 8.2 micron pixels down to an S70 class sensor with 2.3 micron pixels (it would be larger than the S70). That is a 3.57x change in scale. Let's start with a 50mm f/1.4 lens on the 5D, and you need a 14 mm f/0.39 lens that has the same field of view as the larger lens with the larger sensor, but the smaller lens must have 3.57 times the spatial resolution at f/0.39. Not realistic. Now try matching something like Canon 70-200 f/2.8 IS L lens. You would need a 19.6-56mm f/0.79 lens with 3.57 times the spatial resolution. Again, not possible for any reasonable price, and probably not possible at all. >>Another example: Canon's 600 mm f/4 L IS is $7000 at B&H, >>and the 300 f/2.8 L IS is $3850. But it is not f/2, and it is >>only similar in sharpness, NOT DOUBLE. > > That is a totally different problem. Why is it a different problem? People are buying the FZx P&S cameras that have 400+mm equivalent and think they have great telephoto lenses. Roger >>>But you are advocating halving the f/ratio while at the same time, >>>to maintain equivalent spatial resolution, you would need to >>>DOUBLE the spatial resolution of the lens. >> While halving the diameter of the image circle, yes. This is much >> easier than trying to double the spatial resolution over the same >> image circle. After all, I'm not trying to get any more information >> out of the lens: I'm just focusing on a smaller sensor. Of course >> there are engineering challenges to doing this sort of thing! > I think the engineering challenge is the key. If what you > say is true, then one could keep scaling, not just 2x. Well, nothing is for free: there are inevitably some losses, and eventually you'd end up with pixels so small that the finite wavelength of light would come into play. > For example, scale the canon 5D 35.8 x 23.9 mm sensor with > 8.2 micron pixels down to an S70 class sensor with 2.3 micron [quoted text clipped - 3 lines] > as the larger lens with the larger sensor, but the smaller lens > must have 3.57 times the spatial resolution at f/0.39. There are people using relay lenses to match wide-angle lenses to smaller DSLR sensors. Such relays are complicated, heavy and lossy, partly because they're made from a pair of regular lenses fastened back-to-back, but they work. Could a lens manufacturer make a relay simpler, with fewer elements, perform better? But of course if the prime lens were from the beginning designed with this application in mind, it would be better. > Not realistic. Maybe. The question in my mind is just how bad the losses from a well-designed 2x relay would be. Astrophotographers use these things all the time, so presumably the data are available. Andrew. >There are people using relay lenses to match wide-angle lenses to >smaller DSLR sensors. Such relays are complicated, heavy and lossy, >partly because they're made from a pair of regular lenses fastened >back-to-back, but they work. Could a lens manufacturer make a relay >simpler, with fewer elements, perform better? The answer is probably: yes. Can they make one that is good enough to be useful in practice: probably not. The Nikon E2/Fuji DS-505 used optics to project the image of Nikon lenses onto a 3/2" CCD (i.e. while keeping the same field of view as on 35mm film). There is however one problem: the widest aperture supported is f/6.7 (and f/4.8 in the next model). I doubt they have chosen such small apertures just for fun. SignatureThat was it. Done. The faulty Monk was turned out into the desert where it could believe what it liked, including the idea that it had been hard done by. It was allowed to keep its horse, since horses were so cheap to make. -- Douglas Adams in Dirk Gently's Holistic Detective Agency > >There are people using relay lenses to match wide-angle lenses to > >smaller DSLR sensors. Such relays are complicated, heavy and lossy, [quoted text clipped - 12 lines] > > I doubt they have chosen such small apertures just for fun. I have no experience with that camera. However, a problem with optical relays of that sort - more or less reimagers with demagnification - is that if you don't get the exit pupil of the primary lens in the same position as the entrance pupil of the relay, you will get vignetting, potentially severe. The reason is that some of the rays from the primary lens are exiting it an an angle which does not make it through the stop of the relay. In practice, the entrance pupil of the relay is fixed, but the exit pupils of different lenses can be in different places, so it may not be possible to make a relay that works well with all of them, or any. Faster apertures make the problem worse and that's probably why there was an aperture limit. Pupil mismatch is also why teleconverter X may vignette with lenses of focal length Y (or even have different vignetting with two different lenses of the same focal length, because they have different pupil locations). An example of a reimaging system with demagnification is mounting a 50mm lens on your camera and reverse mounting a 100mm lens thread-to-thread in front of that. Sometimes this works, sometimes you get bad vignetting. It depends on the lenses. If you move the front lens forward and back the vignetting may change. In general, this kind of problem is not well solved by throwing more glass at it. >> >There are people using relay lenses to match wide-angle lenses to >> >smaller DSLR sensors. Such relays are complicated, heavy and lossy, [quoted text clipped - 12 lines] >> >> I doubt they have chosen such small apertures just for fun. > I have no experience with that camera. However, a problem with > optical relays of that sort - more or less reimagers with > demagnification - is that if you don't get the exit pupil of the > primary lens in the same position as the entrance pupil of the > relay, you will get vignetting, potentially severe. Sure, but I can't see that makes any difference to the general argument, which is whether or not it's reasonably possible. All it implies is that the lens and the relay need to be designed as a system. Andrew. >Sure, but I can't see that makes any difference to the general >argument, which is whether or not it's reasonably possible. All it >implies is that the lens and the relay need to be designed as a >system. You are basically saying that if I want a 50/0.7 for 35mm full frame, then I can just start with 100/1.4 and use a matching relay lens. I'm not a lens designer. But my guess is that making the relay lens is proably as hard as making the 50/0.7 directly. SignatureThat was it. Done. The faulty Monk was turned out into the desert where it could believe what it liked, including the idea that it had been hard done by. It was allowed to keep its horse, since horses were so cheap to make. -- Douglas Adams in Dirk Gently's Holistic Detective Agency > > I have no experience with that camera. However, a problem with > > optical relays of that sort - more or less reimagers with [quoted text clipped - 6 lines] > implies is that the lens and the relay need to be designed as a > system. I was explaining a likely reason why the camera and reimaging system Phillip described was limited to small apertures. This discussion is becoming vacuous. I've given a few reasons why reimaging systems of the sort you describe are difficult to make (let alone to mate with several different lenses) and why very fast lenses are difficult to make even for small formats. You're coming back with the argument that it could be reasonably possible. There's no way to immediately disprove that (it's hard to prove something impossible), except to point out that lenses of say f/1.2 or faster, even cine lenses, have generally been rare, expensive, and/or compromised. When an optical designer says something is "difficult," he or she doesn't usually mean that it's impossible, but that it's going to cost a lot of money. Also the more difficult something gets, the harder it is to translate from design into as-built and from as-built into mass production. > lenses of > say f/1.2 or faster, even cine lenses, have generally been rare, [quoted text clipped - 3 lines] > the more difficult something gets, the harder it is to translate > from design into as-built and from as-built into mass production. Maybe some interesting old cinema lenses out there suitable for 35mm still shooting? I guess movies are limited to 1/24 sec or whatever frame rate (24 fps?) so they need that but also it can produce some interesting effects and extraordinary bokeh it would seem. > Maybe. The question in my mind is just how bad the losses from a > well-designed 2x relay would be. Astrophotographers use these things > all the time, so presumably the data are available. You mean a Barlow lens? Barlow lenses are negative - they increase the focal length of the telescope, like a teleconverter. This is somewhat easier to do than the opposite, which is what you want. > >>Another example: Canon's 600 mm f/4 L IS is $7000 at B&H, > >>and the 300 f/2.8 L IS is $3850. But it is not f/2, and it is [quoted text clipped - 5 lines] > P&S cameras that have 400+mm equivalent and think they > have great telephoto lenses. I agree on your main point. Miniaturization definitely isn't a vademecum for image quality. But it is for a couple of other parameters. I keep my FZ20 because it provides the best telephoto opportunity _that is readily portable on an everyday basis_. If I'd start again today, I would even have an FZ5 instead. Smaller, lighter, faster autofocus, less expensive. There is no telephoto lens-camera combo that I know of which rivals it in therms of image quality _relative to price_, or image quality _relative to weight_. This said, of course you'll get better image quality with a Rebel XT plus a 17-85 IS and a 70-300 IS, which gives a reasonably similar FOV coverage (a bit better on the wide end, insignificantly better on the tele end). But then you have a combo that not only costs between four and five times more than the FZ5, it weighs five times more as well. (Amusingly enough, price per gramme of the two setups is astonishingly similar, at least in my neck of the woods) Plus that you need a place to store the lens that you don't use at the moment. At least for me, this is an important difference, which distinguishes between purposely going somewhere with the intention to shoot pictures, and keeping a reasonable camera at hand just in case something turns up. Jan Böhme > I agree on your main point. Miniaturization definitely isn't a > vademecum for image quality. But it is for a couple of other [quoted text clipped - 20 lines] > > Jan Böhme Jan, I agree with you completely. The same argument can be made from 8x10 to 4x5, or 4x5 to medium format, or medium to 35mm, etc. Each format has its advantages at some cost (and going down in size is not always a negative as you say). But there is always some compromise in different sized systems, whether it be resolution, light gathering ability, depth of field, bulk and weight. Roger > > Any lens, no matter how big, pointing at a given scene and operating > > at f/8 will cast the same surface brightness of photons per time per [quoted text clipped - 8 lines] > smaller and brighter. (Ignoring, for the sake of the thought > experiment, losses in the relay.) That's basically the opposite of a teleconverter. It would be a positive lens - adding a positive lens decreases focal length, so instead of a 50mmf/2 lens you'd have a 25mm f/1 lens. (The focal length was halved but the aperture stayed the same.) As a supplementary lens, it would be difficult to make such a thing without degrading image quality. Of course, the manufacturer could design a 25mm f/1 lens from the ground up, to cover the smaller sensor. This is the exception I noted at the end of my post: smaller formats get some gain back because it is easier to make fast lenses for small formats without making them obscenely large or hard to manufacture. However, the gain seems to scale slower than format size. For example, 35mm normal 50mm lenses are f/1.7-2 for the average lens and f/1.4 for the fast ones, while f/1.2 or f/1 are exotic. In 6x7, which is double the size of 35mm, normal lenses ~100mm are f/2.8 for the average and f/2.5 to 2 for the occasional fast ones, so they've doubled the focal length and only lost about one stop. In practice, optical systems get very difficult to make around f/1.something. The other issue is that the diffraction limit spot size depends only on f-number. So as you make pixel size smaller, even if you can make the lens faster, the image becomes diffraction-limited at a lower f-number. For 4 micron pixels, the limit is about f/8 in visible light, so if you use an aperture smaller than that, the image is degraded beyond the quality the pixel sampling allows. That's a PITA. > The notion that there is some physical limit to the surface brightness > at the focal plane only applies if you assume relative aperture is > fixed. But relative aperture is not fixed because you can make an > image circle smaller without reducing the physical diameter of the > aperture. > Of course, the manufacturer could design a 25mm f/1 lens from the > ground up, to cover the smaller sensor. Right. > This is the exception I noted at the end of my post: smaller formats > get some gain back because it is easier to make fast lenses for [quoted text clipped - 6 lines] > fast ones, so they've doubled the focal length and only lost about > one stop. Mmm. I'm not sure it's quite so clear-cut as that, but okay. > In practice, optical systems get very difficult to make around > f/1.something. The other issue is that the diffraction limit spot [quoted text clipped - 4 lines] > smaller than that, the image is degraded beyond the quality the > pixel sampling allows. But you don't need to use an aperture smaller than that, because the smaller sensor (with shorter lens) has greater depth of field. So, on a hypothetical 2* sensor you can use f/4 instead of f/8 for optimum sharpness. Andrew. >> Higher well capacity is also possible: there's no reason to suppose >> the current capacity limits will be maintained forever. [quoted text clipped - 30 lines] >a larger pixel will collect more. It is up to the user >to decide how large a pixel they want (or can afford). Don't optical coatings exist that absorb in the UV and re-radiate in the visible? I don't know the exact wavelenghts they re-radiate in. It might be possible to apply these to the windows of CCDs/CMOS in order to "boost" incoming light. -Rich > Don't optical coatings exist that absorb in the UV and re-radiate in > the visible? I don't know the exact wavelenghts they re-radiate in. > It might be possible to apply these to the windows of CCDs/CMOS > in order to "boost" incoming light. > -Rich Well, if they don't exist yet, it would probably be simple to make them. There are two issues that I see. 1) UV would not help get the best color, and in fact could really mess up color accuracy as UV is outside the normal range of human color vision. 2) This is called fluorescence. Fluorescence efficiency is generally very low, often less than a few %, so it really would not help much. Roger Andrew Haley skrev: > > The only room left for improvement is to improve quantum > > efficiency, lower read noise, and lower dark current. > Higher well capacity is also possible: there's no reason to suppose > the current capacity limits will be maintained forever. Everything else being equal, this will lead to higher signal-to noise ratio, and thus a larger dynamic range, but only at the expense of ISO sensitivity. Thus, higher well capacity will, in and by itself, lead to to an ISO50 or ISO25 with higher signal-to noise ratio, but it will do absolutely nothing for the S/N ratio at ISO200. There won't be enough photons to exploit the difference there. And such a dynamic range would have to be matched with a corresponding precision in the RAW recording. 16-bit probably wouldn't be enough, seeing that the 12-bit standard of today falls rather short of the capacity already of today's sensors. Jan Böhme |
Enable EMail Alerts
Start New Thread
Reply to this Message