Luminous Landscape Forum
Equipment & Techniques => Medium Format / Film / Digital Backs – and Large Sensor Photography => Topic started by: eronald on October 14, 2009, 04:34:33 am
-
I don't really understand much about technology, especially about electronics, but fortunately there are those on this forum that do.
So please, enlighten me - why can Nikon specify sensors which apparently work well at ISO 12800 and do liveview/video when Dalsa/Kodak sensors streak at 400?
Surely, there is something to be said for Hi-ISO for location, and live-view and on-sensor autofocus for studio work, or is MF dragging itself into the same technological irrelevance as the Leica M system? Are we going to be told that 100 ISO is the "advantage" of MF?
Edmund
-
I think the main reason is economies of scale. The total combined sales of all MFDB is less than 6,000 per year, which is not nearly enough to support the R&D needed to compete with Canikon in low light. I think something will change, though, and some day MFDB will not be just for low ISO.
-
boy,
that D3s sure slipped in under the radar! when was that launched? I'm guessing the D3 is going to be phased out?
but I agree, why can't digi backs ever give a nice, clean file at iso 800? i mean, 33mp would be nice and 800iso to boot would be the cherry on top.
I'm just surprised that Leaf and Phase haven't been selling this current pixelated look as the "film look".
-Dan
-
The MFDB makers are and have been at the mercy of the chip makers for many years. It seems that Dalsa and Kodak don't take this niche market very seriously. If they did release some truly revolutionary sensors, it could only help this niche market to grow. It's rather frustrating...
-
I don't really understand much about technology, especially about electronics, but fortunately there are those on this forum that do.
So please, enlighten me - why can Nikon specify sensors which apparently work well at ISO 12800 and do liveview/video when Dalsa/Kodak sensors streak at 400?
Surely, there is something to be said for Hi-ISO for location, and live-view and on-sensor autofocus for studio work, or is MF dragging itself into the same technological irrelevance as the Leica M system? Are we going to be told that 100 ISO is the "advantage" of MF?
Edmund
Those high ISOs are not real, but are useful for shooting in JPEG. When shooting in raw, one has the choice to increase the ISO either via a camera setting or by increasing exposure in the raw converter. Obviously, increasing the ISO while keeping exposure constant does not affect how many photons are recorded by the sensor but merely changes how those photo-electrons are processed. When light is limited, it makes sense to increase the ISO because read noise is higher at low ISO than higher ISOs. However, as the ISO is raised, there are diminishing returns and at some point, read noise does not change that much but headroom for highlights is diminished. Emil Martinec (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#ETTR) explains the science. With the Canon 1D3 that point is reached at about ISO 1600. Roger Clark reaches a similar conclusion.
With the Nikon D3, amplifier gain is increased up to an ISO of 6400. Higher ISOs are obtained by multiplying the ISO 6400 values post read. Read noise would asymptotically approach a minimum at a lower ISO, but I don't have date on what that ISO is.
Some MFDBs do not vary amplifier gain when ISO is increased, but merely tag the file with the rated ISO so that exposure compensation can be carried out by the raw converter.
-
Simple answer is CCD vs CMOS...
Simple explanation is that in digital imaging, heat=noise and increasing gain on the sensor increases heat, and CCD's generate more heat than CMOS, hence generate more noise sooner...
-
Simple answer is CCD vs CMOS...
Simple explanation is that in digital imaging, heat=noise and increasing gain on the sensor increases heat, and CCD's generate more heat than CMOS, hence generate more noise sooner...
I doubt that explanation. The main components of noise in a digital sensor are photon noise (shot noise), read noise, and thermal noise (dark current). Only the last is significantly affected by heat (see here (http://digitalimagingu.com/articles/ccdsnr.html)). Thermal noise is important in astronomy and scientific imaging, but is negligible for exposures of one second or less (see Roger Clark (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html)).
-
I doubt that explanation.
I am so surprised... . Note that ISO noise was the OP's topic here and is a different discussion than CMOS site-amplifier induced pattern noise which gets subtracted in real-time at the pixel site (at least in most contemporary CMOS cameras).
From http://www.imaging-resource.com/PRODS/D30/D30A4.HTM (http://www.imaging-resource.com/PRODS/D30/D30A4.HTM), "The Active Pixel CMOS image sensors used in digital imaging are very similar to a CCD sensor, but with one major difference — supporting circuitry is actually located alongside each light receptor, allowing noise at each pixel to be canceled out at the site."
And from http://broadcastengineering.com/hdtv/ccd-cmos/ (http://broadcastengineering.com/hdtv/ccd-cmos/), "CCDs typically obtain super high sensitivity by engaging hyper gain. The result is a significant increase in image noise. "
So if you would be so kind as to enlighten me, please direct me to any current non-supercooled consumer CCD-sensored camera that has outstanding ISO performance comparable to Canon or Nikon CMOS.
Cheers,
-
So please, enlighten me - why can Nikon specify sensors which apparently work well at ISO 12800 and do liveview/video when Dalsa/Kodak sensors streak at 400?
come back and ask that question when Nikon specifies a sensor larger than 36x24
Surely, there is something to be said for Hi-ISO for location, and live-view and on-sensor autofocus for studio work, or is MF dragging itself into the same technological irrelevance as the Leica M system? Are we going to be told that 100 ISO is the "advantage" of MF?
there's always something to be said just about "anything".
especially when one is a cheerleader of that anything ...
-
CMOS is the future because full-frame CCDs aren't capable of good live-view which is crucial for coming EVIL-systems. But full-frame CCDs are an excellent choice regarding highest IQ because they don't waste as much space on their surface to circuitry. MFDBs may not have any decent in-camera-processing algorithms but properly processed/converted images are still excellent. Custom CMOS-sensors are already used for some apllications without sufficient "economies of scale".
The current Kodak/Dalsa-offerings aren't inferior, it's a different technology with different strengths and weaknesses.
There are very few people willing to do any simple comparisons on a tripod comparing cameras like the D700 (which is considered to be the best high-ISO-camera - the early samples of the D3s show only minimal improvement) and a CCD-based camera with a similar sensor size (the M9). One guy on the Leica-forum did it - I hope he doesn't mind that I put the two 2500ASA-samples beneath each other - even this comparison is far from being perfect (the light slightly changed between the two shots) - but where's the 2-4stops difference often claimed? Which one is which (to be clear, I wouldn't use any of them for prints)?
[attachment=17196:m9vsd700.JPG]
-
From http://www.imaging-resource.com/PRODS/D30/D30A4.HTM (http://www.imaging-resource.com/PRODS/D30/D30A4.HTM), "The Active Pixel CMOS image sensors used in digital imaging are very similar to a CCD sensor, but with one major difference — supporting circuitry is actually located alongside each light receptor, allowing noise at each pixel to be canceled out at the site."
Somewhat true, but I do see the connection with the on-sensor heat production mentioned in your previous post. And that IR quote is wrong in one major way: the noise cancellation done at the pixel level in CMOS sensors is eliminating a type of noise that CCD's do not have in the first place, so this is a matter of overcoming a weakness of earlier, rather noisy, CMOS sensors, not an advantage over CCDs. The CMOS high ISO speed noise advantage has a different basis.
At normal exposure times (under 1s, say), the main noise source that sensor and amplifier design can do much about is amplifier noise, not heat or any other source of noise within the photosites themselves. CMOS sensors do the initial amplification very early, and massively in parallel, as amplification is done in each column of pixels: thousands of amplifiers in total. With CCD sensors, the amplification happens off the sensor chip itself, using just a few amplifiers which therefore have to operate at a far higher rate. The far lower operating frequency of the on-chip amplifiers of CMOS sensors help to lower amp. noise. (However, the tiny on-chip amplifiers of CMOS sensors might have a lower maximum DR than the far bigger off-chip ones that can be used with good CCD's, and this might relate to why at minimum ISO speed, high end CCD's have higher measured DR.)
On the subject of amplifier operating frequency and noise, Dalsa spec. sheets describe how noise level can be decreased and DR increased by lowering the read-out rate. Also, one often stated reason for increasing the amplifier count with CCD's, is to lower the operating frequency needed by each amplifier, and so reduce its noise.
(Aside: Dalsa builds amplifiers in at each corner of its FF CCD sensor units, and maybe Kodak does too. The dark noise level and DR figures in their spec. sheets are for the signal coming out of the amplifiers, so includes amp. noise.)
-
I am so surprised... . Note that ISO noise was the OP's topic here and is a different discussion than CMOS site-amplifier induced pattern noise which gets subtracted in real-time at the pixel site (at least in most contemporary CMOS cameras).
From http://www.imaging-resource.com/PRODS/D30/D30A4.HTM (http://www.imaging-resource.com/PRODS/D30/D30A4.HTM), "The Active Pixel CMOS image sensors used in digital imaging are very similar to a CCD sensor, but with one major difference — supporting circuitry is actually located alongside each light receptor, allowing noise at each pixel to be canceled out at the site."
And from http://broadcastengineering.com/hdtv/ccd-cmos/ (http://broadcastengineering.com/hdtv/ccd-cmos/), "CCDs typically obtain super high sensitivity by engaging hyper gain. The result is a significant increase in image noise. "
So if you would be so kind as to enlighten me, please direct me to any current non-supercooled consumer CCD-sensored camera that has outstanding ISO performance comparable to Canon or Nikon CMOS.
Cheers,
You seem to be confusing total noise with thermal noise. Pixel response non-uniformity (PRNU) is less a problem with CCD than CMOS, and is not thermal noise. As you may know, noise adds in quadrature and the total variance is the sum of the individual variances (shot noise, read noise, thermal noise). Current CMOS designs have lower read noise than CCDs because the developers have invested in research to lower this type of noise. However, this has nothing to do with thermal noise as per your suggestion. In routine photography, thermal noise is not a problem with either CMOS or CCD. If you want to learn a bit of basic physics, I would suggest you read the paper by Emil Martinec. He addresses thermal noise (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html#thermalnoise).
Your first reference is nearly 10 years old; why do you quote it? It does not address thermal noise, which is not a problem anyway. Your second reference concerns motion picture sensors, although the principles are the same for still cameras, different considerations apply. CMOS can have a much higher readout rate, which is important for movies with a high frame rate. Hyper-gain need not be used, and I image that it would affect read noise, not thermal noise. Shot noise is lilmited by the full well capacity. Please supply a reference to substantiate your theory about heat. I do not expect a reply.
-
The MFDB makers are and have been at the mercy of the chip makers for many years. It seems that Dalsa and Kodak don't take this niche market very seriously. If they did release some truly revolutionary sensors, it could only help this niche market to grow. It's rather frustrating...
They don't because of the small volume being sold. Volume are small because of the pricing.
It has been obvious for years that high performance and small volume in high tech are incompatible requirements in the mid/long run.
The way nuforce is totally trouncing high end hifi equipment costing 5 or 10 times more is yet another example in the supposedely different high end hifi world too.
Cheers,
Bernard
-
There are very few people willing to do any simple comparisons on a tripod comparing cameras like the D700 (which is considered to be the best high-ISO-camera - the early samples of the D3s show only minimal improvement) and a CCD-based camera with a similar sensor size (the M9). One guy on the Leica-forum did it - I hope he doesn't mind that I put the two 2500ASA-samples beneath each other - even this comparison is far from being perfect (the light slightly changed between the two shots) - but where's the 2-4stops difference often claimed? Which one is which (to be clear, I wouldn't use any of them for prints)?
[attachment=17196:m9vsd700.JPG]
Although I agree with you that the M9 might not be as bad a high ISO camera as some people have claimed, let'd compare the latest achievements on both sides.
The sensor of the D700 is 2 years older, first D3s assessment show a one stop improvement in noise although some people claim that ISO12800 on the D3s is closer to ISO 3200 on the D700/D3 than ISO6400, which would mean a 2 stops improvement.
Current rumors claim that the Canon 1DIV to be accounced any day might be even better than the D3s, we'll know for sure in a few days.
Cheers,
Bernard
-
I think the noise-performance of current DSLRs is impressive. The M9 works well up to about 1250ASA, beyond that more serious artifacts occur which are propably more difficult to hide with NR. But I wouldn't use any 20+MP-DSLR beyond 1600ASA for serious printing, either. It even seems that Leica decided to add NR in RAWs just like with CMOS-based cameras, besides the fact that most CCD-sensors are implemented into entirely different systems (no AA-filter, different lenses...) this makes the actual noise comparison even more difficult.
Besides the well-documented differences between CMOS and CCD-cameras we actually know very little about the internal design decisions/tricks of this industry - it's a bit risky to focus that much on CCD vs. CMOS or the noise-characteristics of a few samples in the web (including mine).
It will be interesting to see how the future cameras work (the D3s as well as the MkIV and the S2, the first next-gen CCD with microlenses and ASIC) but I fear that the major breakthroughs have already happened and after the hype with every new model (M9: 2500ASA like 800ASA with M8; D3s: 25600ASA like 6400ASA...) we usually see only incremental improvements When there are not enough electrons hiting the photosite, there's very little we can do...
P.S. the 6400ASA-sample of the D3s looks worse than processed 2500ASA with the M9 and 3200ASA from the D700/D3 as well - at least to my eyes. And I wouldn't print anything beyond 6400ASA (except for stamps...).
-
It even seems that Leica decided to add NR in RAWs just like with CMOS-based cameras,
This is a common misconception. The "NR" everyone talks about in CMOS cameras is just correlated double sampling (CDS), mostly for reduction in kTC noise, which as BJL pointed out above, is stronger in CMOS than CCD (or it would be, if it weren't for CDS). The far majority of CMOS cameras have no NR applied to the raw file, but there are a few exceptions such as the Pentax K20D (30% reduction in mean at ISO 1600 only), and a few Sony models. Sometimes you hear people bring up the long exposure noise reduction or black clipping in Nikon models.
-
The sensor of the D700 is 2 years older, first D3s assessment show a one stop improvement in noise although some people claim that ISO12800 on the D3s is closer to ISO 3200 on the D700/D3 than ISO6400, which would mean a 2 stops improvement.
In fact, in the preview of the D3s by Marcus Bell, linked on 'What's New', Marcus claims that ISO 12,800 seems equal to the old D3 at ISO 3200.
I'm very skeptical. I'm reminded of the hype that surrounded the introduction of the D3. It was claimed to have (up to) 2 stops better high-ISO performance than any other DSLR on the market. In fact, my own tests indicated fairly conclusively the D3 had at most 1/2 a stop lower noise than the old 5D.
What was happening is that D3 jpegs were subject to greater in-camera chroma noise reduction than 5D jpegs and therefore looked much cleaner but were also softer than the 5D images. After applying a suitable amount of chroma noise reduction to the 5D image with Noise Ninja, there was very little image quality advantage to the D3 at ISO 3200 and above, on average about 1/3rd of a stop. D3 RAW images after processing and equal amounts of noise reduction also showed no more than 1/2 a stop improvement over the 5D.
The more significant advantage of the D3 compared with the 5D was not at high ISO but at low ISOs which I didn't have the time and opportunity to test. According to DXOMark, the D3 has more than a whole stop higher DR than the 5D at base ISO.
I would be very surprised if the D3s has as much as one stop image quality improvement over the D3 at ISO 12,800, taking image resolution into consideration also. If it does, then a whole stop improvement is truly remarkable. Two stops improvement is just not credible, but I stand to be corrected.
-
I would be very surprised if the D3s has as much as one stop image quality improvement over the D3 at ISO 12,800, taking image resolution into consideration also. If it does, then a whole stop improvement is truly remarkable. Two stops improvement is just not credible, but I stand to be corrected.
Yes, we'll need to wait for more test reports. I have just sold my D3, so I won't be able to do any test for sure... especially taking into account the fact that I don't intend to buy a D3s.
Cheers,
Bernard
-
@Daniel Browning
You mean the reduction of pattern noise produced by the production tolerances of the amps (within every "pixel")? I'm not sure if this really explains the "look" of high-iso-CMOS-cams and the sometimes bizarre non-linear measurement-results. This look (even with NR turned off in camera) is not too different than CCD-images with noise-reduction, while professional CMOS-(video)-cameras often show noise closer to our known CCD-results - some things go on underneath we don't know about.
-
Are you replacing it Bernard or is the D3X all the camera you need these days?
-
You seem to be confusing total noise with thermal noise.
Nope -- I think you'll find that thermal noise makes up the greater part of ISO noise whether it is amplified after the fact or generated on-chip during capture... We all know you are the most brilliant person sharing photographic knowledge on the internet and so I defer to your supreme intellect in these matters... But I ask you again, can you please point us to ANY consumer CCD camera that outperforms a current consumer CMOS camera on ISO noise?
-
I think you'll find that thermal noise makes up the greater part of ISO noise whether it is amplified after the fact or generated on-chip during capture.
Do you have any source for that? I asked Eric Fossum, inventor of the modern active pixel type of CMOS sensor and still working in the CMOS sensor industry, in a DPReview forum discussion, and he confirmed that to the best of his knowledge (vastly greater than yours or mine) thermal noise from within photosites is not a significant factor, as it is greatly overshadowed by amplifier noise. (Except in long exposures, when dark current noise accumulation can be an issue.)
Also, as soon as the illumination of a pixel is half-way decent, about 100 photons or more detected, photon shot noise overwhelms both read noise (mostly amp. noise) and thermal noise.
can you please point us to ANY consumer CCD camera that outperforms a current consumer CMOS camera on ISO noise?
I do not see the point of that question since no-one is disagreeing that modern CMOS sensors outperform CCDs at high ISO. The only disagreement is about the reasons for that advantage. My answer: mostly a better approach to amplification allowed by the active pixel approach of modern CMOS sensors vs the passive pixels of CCDs. By the way, once upon a time there were "passive pixel" CMOS sensors, but they were only useful as a cheap but noisy low end alternative to CCD's.
-
Nope -- I think you'll find that thermal noise makes up the greater part of ISO noise whether it is amplified after the fact or generated on-chip during capture... We all know you are the most brilliant person sharing photographic knowledge on the internet and so I defer to your supreme intellect in these matters... But I ask you again, can you please point us to ANY consumer CCD camera that outperforms a current consumer CMOS camera on ISO noise?
Jack,
I thought that you had given up on this point . You seem to have edited your response and invented a new type of noise: ISO Noise. Noise at high ISO comes from the same sources as noise at low ISO and the main components (exclusive of PRNU) are shot noise and read noise. Thermal noise is negligible at exposures of less than a second. If you had taken the trouble to read the essays by Roger Clark or Emil Martinec (both of whom are PhDs in astrophysics and physics), you would know that thermal noise is not significant under normal photographic conditons.
In his analysis of the Canon 1D MII (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html), Roger Clark found that average dark currents (thermal noise) were 0.013 to 0.02 electrons/second, but some pixels had dark currents as high as about 0.25 electrons/second. With an exposure of 1 second, 0.25 electrons is insignificant, when compared to total noise, but does become significant with exposures of 623 and 1800 seconds (see Equation 2). The read noise of this camera at an ISO of 3200 is 3.93 electrons and about 1650 electrons are collected at maximum signal (Table 1b). The shot noise for 1650 electrons is sqrt (1650) = 41 electrons. The thermal noise is 0.25 electrons at maximum. The total noise at maximum signal is sqrt (41^2 + 3.93^2 + 0.25^2) = 40.81. With the lens cap on, noise would consist almost entirely of read noise, since there would be no shot noise and this would determine the engineering definition of dynamic range. Figure it out.
Now you are specifying consumer grade cameras and a new kind of noise. What you are asking is irrelevant to the discussion and is pure obfuscation. Your sarcasm only shows what a poor looser you are. Give it up!
-
Now you are specifying consumer grade cameras and a new kind of noise. What you are asking is irrelevant to the discussion and is pure obfuscation. Your sarcasm only shows what a poor looser you are. Give it up!
bjanes,
If I were you, I would give up, with Jack (or Guy), you won't win and he (they) will always find a way to get away, by either editing or changing the subject in a way to present it at his (their) advantage: I have made the experience many many times, here and elsewhere.
I know this post and words will get me again a harsh response, but I don't care, he (they) knows what I am thinking about him (them).
All the best,
Thierry
-
no-one is disagreeing that modern CMOS sensors outperform CCDs at high ISO.
I obviously read certain responses differently than you did... But assuming you're right, then we're all in agreement on the main point which is, "CMOS sensors outperform CCDs at high ISO."
Cheers,
-
In fact, in the preview of the D3s by Marcus Bell, linked on 'What's New', Marcus claims that ISO 12,800 seems equal to the old D3 at ISO 3200.
I'm very skeptical. I'm reminded of the hype that surrounded the introduction of the D3. It was claimed to have (up to) 2 stops better high-ISO performance than any other DSLR on the market. In fact, my own tests indicated fairly conclusively the D3 had at most 1/2 a stop lower noise than the old 5D.
What was happening is that D3 jpegs were subject to greater in-camera chroma noise reduction than 5D jpegs and therefore looked much cleaner but were also softer than the 5D images. After applying a suitable amount of chroma noise reduction to the 5D image with Noise Ninja, there was very little image quality advantage to the D3 at ISO 3200 and above, on average about 1/3rd of a stop. D3 RAW images after processing and equal amounts of noise reduction also showed no more than 1/2 a stop improvement over the 5D.
Ray,
You are right to be skeptical of unsubstantiated claims of people wishing to diss other than their favorite camera brand, but the DXO ratings which you quote gave the Canon 5D, Canon 1DMII and D3 high ISO ratings of 1368, 1003, and 2250 respectively. The D3 was the first Nikon camera to best Canon in that area, but likely Canon will leapfrog Nikon in the near future. In any case my D3 and your Canon will continue to do the jobs for which we purchased them in the first place. Everyone wins from healthy competition.
Bill
-
I obviously read certain responses differently than you did... But assuming you're right, then we're all in agreement on the main point which is, "CMOS sensors outperform CCDs at high ISO."
Cheers,
Jack,
We all know that you are a great photographer with bags of expensive high end medium format digital cameras, but now the question is can you read beyond an 8th grade level?
Cheers,
Bill