If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below. |
|
|
Thread Tools | Display Modes |
#51
|
|||
|
|||
25 MP sensor of Sony
Hi
In article , "David J. Littleboy" wrote: This is where 3-color perpixel sensors are problematic: the raw files are three times larger. harr be ye meaning the Foveon thinggy? Thought they had seperate per pixel bits internally I did. See Ya (when bandwidth gets better ;-) Chris Eastwood Photographer, Programmer Motorcyclist and dingbat please remove undies for reply |
#52
|
|||
|
|||
25 MP sensor of Sony
Floyd L. Davidson wrote:
"David J Taylor" wrote: Chris Malcolm wrote: [] You're quite right, the non-linearity of gamma correction does increase the resolutions required. The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between. Chris, It affects things by how you define dynamic range, whether the lowest bit is just on-off, or whether there is a finite signal-to-noise ratio at the bottom end of the "dynamic range". Some people use the latter definition, and it may relate to how film dynamic range is defined. By either definition, his statement is not true. Were I making your assumptions about range representation, which I'm not. I suspect that the most important thing we disagree about is whether we disagree :-) -- Chris Malcolm DoD #205 IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK [http://www.dai.ed.ac.uk/homes/cam/] |
#53
|
|||
|
|||
25 MP sensor of Sony
Chris Malcolm wrote:
Floyd L. Davidson wrote: "David J Taylor" wrote: Chris Malcolm wrote: [] You're quite right, the non-linearity of gamma correction does increase the resolutions required. The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between. Chris, It affects things by how you define dynamic range, whether the lowest bit is just on-off, or whether there is a finite signal-to-noise ratio at the bottom end of the "dynamic range". Some people use the latter definition, and it may relate to how film dynamic range is defined. By either definition, his statement is not true. Were I making your assumptions about range representation, which I'm not. I suspect that the most important thing we disagree about is whether we disagree :-) To you perhaps. But what you said was: "The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between." And that, as virtually *any* tutorial or text on the subject will explain the math for you, is wrong. The standard formula is: Dynamic_Range_in_dB = (n * 6.02) + 1.72 Where 'n' is the number of bits. The number of bits of course does in fact set the maximum number of steps, and therefore the minimum size of those steps. And *that* is what sets the dynamic range that can be encoded. You can claim all you like that any range can be endocded with one bit, but the *fact* is that 1 bit is limited to 6.7 dB of dynamic range. Digital encoding wil *never* make sense until the reason for that limit is clearly understood. -- Floyd L. Davidson http://www.apaflo.com/floyd_davidson Ukpeagvik (Barrow, Alaska) |
#54
|
|||
|
|||
25 MP sensor of Sony
|
#55
|
|||
|
|||
25 MP sensor of Sony
Kevin McMurtrie wrote:
In article , (Floyd L. Davidson) wrote: To you perhaps. But what you said was: "The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between." And that, as virtually *any* tutorial or text on the subject will explain the math for you, is wrong. The standard formula is: Dynamic_Range_in_dB = (n * 6.02) + 1.72 That formula calculates the theoretical dynamic range for a single sample of data. That is not correct. It calculates the dynamic range of the resulting digital data set, not from 1 sample but over the average of many (i.e., all possible) samples. The quantization distortion is determined by the integral of the error signal spread across the entire range of each step (as opposed to one sample that has only a single value of error). It is not correct to apply this to a set of data, such as an audio signal, an image, or a video signal. As *any* half decent tutorial or text on the subject will explain, that is *exactly* what it applies to. Here are three examples, two books on audio signals and one URL that discusses digital imaging. "Digital Telephony", 3rd Ed., John C. Bellamy, 2000, published by John Wiley & Sons. pp. 99-101 The above text shows the mathematical derivation, as well as examples of its use. "Telecommunications System Engineering", 3rd Ed., Roger L. Freeman, 1996, published by John Wiley & Sons. p 353. This text does not show the derivation, but states the formula and give examples. http://www.cis.rit.edu/class/simg712...antization.pdf See pages 12 and 13. It doesn't make any more sense than it would to say that a camera's performance is pi*r^2 of the front lens. .... Digital encoding wil *never* make sense until the reason for that limit is clearly understood. That is still a very valid statement, obviously. Until people understand the theory involved in digitizing analog data with a linear ADC... the rest of this discussion is meaningless wandering in the dark. -- Floyd L. Davidson http://www.apaflo.com/floyd_davidson Ukpeagvik (Barrow, Alaska) |
#56
|
|||
|
|||
25 MP sensor of Sony
Floyd L. Davidson wrote:
Chris Malcolm wrote: Floyd L. Davidson wrote: "David J Taylor" wrote: Chris Malcolm wrote: [] You're quite right, the non-linearity of gamma correction does increase the resolutions required. The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between. Chris, It affects things by how you define dynamic range, whether the lowest bit is just on-off, or whether there is a finite signal-to-noise ratio at the bottom end of the "dynamic range". Some people use the latter definition, and it may relate to how film dynamic range is defined. By either definition, his statement is not true. Were I making your assumptions about range representation, which I'm not. I suspect that the most important thing we disagree about is whether we disagree :-) To you perhaps. But what you said was: "The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between." And that, as virtually *any* tutorial or text on the subject will explain the math for you, is wrong. The standard formula is: Dynamic_Range_in_dB = (n * 6.02) + 1.72 Where 'n' is the number of bits. The number of bits of course does in fact set the maximum number of steps, and therefore the minimum size of those steps. And *that* is what sets the dynamic range that can be encoded. You can claim all you like that any range can be endocded with one bit, but the *fact* is that 1 bit is limited to 6.7 dB of dynamic range. If you can't see the assumption in that formula even though you have gone so far as to describe it there's no point in continuing this argument. (There's nothing wrong with it being based on assumptions -- nearly all standard engineering formulas are. But it matters when you're trying to discuss the principles behind the engineering assumptions. And why should we be doing that? Because we were discsussing the differences between two different kinds of engineering implementation, which raised the question of the relevance of the usual assumptions which are based on the usual technology. At least, that's what I was trying to do :-) -- Chris Malcolm DoD #205 IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK [http://www.dai.ed.ac.uk/homes/cam/] |
#57
|
|||
|
|||
25 MP sensor of Sony
Chris Malcolm wrote:
Floyd L. Davidson wrote: Chris Malcolm wrote: Floyd L. Davidson wrote: "David J Taylor" wrote: Chris Malcolm wrote: [] You're quite right, the non-linearity of gamma correction does increase the resolutions required. The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between. Chris, It affects things by how you define dynamic range, whether the lowest bit is just on-off, or whether there is a finite signal-to-noise ratio at the bottom end of the "dynamic range". Some people use the latter definition, and it may relate to how film dynamic range is defined. By either definition, his statement is not true. Were I making your assumptions about range representation, which I'm not. I suspect that the most important thing we disagree about is whether we disagree :-) To you perhaps. But what you said was: "The important point remains that the number of bits in the encoding doesn't affect the dynamic range between darkest and lightest, just the number of steps in between." And that, as virtually *any* tutorial or text on the subject will explain the math for you, is wrong. The standard formula is: Dynamic_Range_in_dB = (n * 6.02) + 1.72 Where 'n' is the number of bits. The number of bits of course does in fact set the maximum number of steps, and therefore the minimum size of those steps. And *that* is what sets the dynamic range that can be encoded. You can claim all you like that any range can be endocded with one bit, but the *fact* is that 1 bit is limited to 6.7 dB of dynamic range. If you can't see the assumption in that formula even though you have gone so far as to describe it there's no point in continuing this argument. There are a number of assumptions. Your vague reference to such doesn't make much sense. (It is assumed, for example, that the error signal is evenly distributed across the entire range of a single step. That may not be true for any given signal, but it has been demonstrated to be a close enough approximation.) (There's nothing wrong with it being based on assumptions -- nearly all standard engineering formulas are. But it matters when you're trying to discuss the principles behind the engineering assumptions. And why should we be doing that? Because we want *accurate* answers that allow further understanding of that and other processes. Because we were discsussing the differences between two different kinds of engineering implementation, which raised the question of the relevance of the usual assumptions which are based on the usual technology. Trying to baffle someone with bull****? Appeal to foggy confusion? (The "two different" implementations are the figment of someone's imagination to start with, but it would make no difference because they are both based on precisely the same engineering "implementation".) At least, that's what I was trying to do :-) You've make a very basic error, and repeated is several times, in relations to bit depth and dynamic range. As I've said repeatedly, until *that* part is understood, the rest of it is going to remain a confused mystery. Let me repeat, the dynamic range which can be represented by a linear ADC when an analog signal is converted to digital data, is *well* *known* to be represented by the formula: Dynamic_Range_in_dB = (Bit_Depth * 6.02) + 1.76 You can, as noted, find the derivation of that formula in almost any good engineering text. You can find the formula stated in everything from simple tutorials to the manufacturer's data sheets and application notes for ADC's. For example, from http://datasheets.maxim-ic.com/en/ds/MAX1183.pdf "Signal-to-Noise Ratio (SNR) For a waveform perfectly reconstructed from digital samples, the theoretical maximum SNR is the ratio of the fullscale analog input (RMS value) to the RMS quantization error (residual error). The ideal, theoretical minimum analog-to-digital noise is caused by quantization error only and results directly from the ADC's resolution (N-Bits): SNRdB[max] = 6.02 N + 1.76 In reality, there are other noise sources besides quantization noise (thermal noise, reference noise, clock jitter, etc.). SNR is computed by taking the ratio of the RMS signal to the RMS noise, which includes all spectral components minus the fundamental, the first five harmonics, and the DC offset. What you *cannot* find is *anything* which will support your repeated statements that bit depth does not restrict the dynamic range. And until you understand why, all of this is just going to continue to confuse you. -- Floyd L. Davidson http://www.apaflo.com/floyd_davidson Ukpeagvik (Barrow, Alaska) |
#59
|
|||
|
|||
25 MP sensor of Sony
Kevin McMurtrie wrote:
(Floyd L. Davidson) wrote: Kevin McMurtrie wrote: (Floyd L. Davidson) wrote: The standard formula is: Dynamic_Range_in_dB = (n * 6.02) + 1.72 That formula calculates the theoretical dynamic range for a single sample of data. That is not correct. It calculates the dynamic range of the resulting digital data set, not from 1 sample but .... It is not correct to apply this to a set of data, such as an audio signal, an image, or a video signal. As *any* half decent tutorial or text on the subject will explain, that is *exactly* what it applies to. Here are .... This text does not show the derivation, but states the formula and give examples. http://www.cis.rit.edu/class/simg712...antization.pdf See pages 12 and 13. Excellent. Please read the rest of it. It says that virtually everything you stated is not true, as noted in the quoted text above. It is *not* from 1 sample. It is used for audio data, image data, and video signals. And I'll grant that you are probably confusing the issue with dithering... :-) -- Floyd L. Davidson http://www.apaflo.com/floyd_davidson Ukpeagvik (Barrow, Alaska) |
#60
|
|||
|
|||
25 MP sensor of Sony
Floyd L. Davidson wrote:
What you *cannot* find is *anything* which will support your repeated statements that bit depth does not restrict the dynamic range. And until you understand why, all of this is just going to continue to confuse you. It doesn't confuse me at all. You've given a number of reasons why bit depth has to restrict dynamic range because the current accepted relationships between bit depth and dynamic range can't be changed without making other important things a lot worse. I don't disagree with any of that. What I'm pointing out is that it's the optimal point in an engineering compromise, not a fundamental scientific principle. It's the difference between practical and theoretical possibilities. If the technology changes, the optimal compromise might change. -- Chris Malcolm DoD #205 IPAB, Informatics, JCMB, King's Buildings, Edinburgh, EH9 3JZ, UK [http://www.dai.ed.ac.uk/homes/cam/] |
Thread Tools | |
Display Modes | |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Sony: 24.5 megapixel FF sensor announced | RichA | Digital SLR Cameras | 33 | February 5th 08 11:11 AM |
New Sony 12.4 mpixel 10fps sensor | David Kilpatrick | Digital SLR Cameras | 2 | August 21st 07 11:55 PM |
On the rumored 18.7 Mpix Sony sensor | Alan Browne | Digital SLR Cameras | 9 | July 19th 07 07:23 PM |
Olympus and the Sony sensor recall? | Mike S. | Digital Photography | 3 | March 29th 06 06:30 PM |
D2X (12.4 MP) CMOS sensor is made by Sony, per dpreview / Galbraith | Alan Browne | Digital Photography | 16 | September 20th 04 04:07 PM |