Geert Koning wrote:
There is more to it than only SNR
Diffraction will kick in at about f 8 with this pixel density...
The "amount" of diffraction is identical on every 1.6x cropped sensor. The absolute "size" of the diffraction is a function of aperture and focal length, and the also-important "relative size" is a function of sensor dimensions.
There is no more diffraction on a sensor with more photosites.
If you make a print from a camera with X photosites using a particular lens/aperture combination, and then make another print from a second camera with the same sensor dimensions, aperture, and lens but with, say, X times 2 photosites...
I don't know what NR if any was done to it in PP but it looks pretty darn good to me. There's practically no noise in the black areas and very little of the grainy look in said area either.
I am disappointed Canon didn't use 16-bit ADCs instead of using 14-bit ADCs. This would have given the 50D another stop of ISO sensitivity and improved the dynamic range as well.
I thought that the gap between sensors reduced the heat that is produced by the sensor. If they are now touching wouldn't that mean more heat=more noise?
rsg_1 wrote:
I am disappointed Canon didn't use 16-bit ADCs instead of using 14-bit ADCs. This would have given the 50D another stop of ISO sensitivity and improved the dynamic range as well.
... what makes you think so? more bits just mean that more bits will be lost in noise. We only will have bigger files and no benefit at all. Even the 14bits are already too much as some clever people found out.
jbfaulconer wrote:
I thought that the gap between sensors reduced the heat that is produced by the sensor. If they are now touching wouldn't that mean more heat=more noise?
the gap is there because of the transistors who are there. There are 3 or 4 transistors at each photosensor.
rockitman wrote:
I agree, I shoot hummers from f/18 - f/20 on my 1Ds3 with a multi-flash setup for DOF and the pics are sharp. The "useless past f/16" statement wins post of the month.
rsg_1 wrote:
I am disappointed Canon didn't use 16-bit ADCs instead of using 14-bit ADCs. This would have given the 50D another stop of ISO sensitivity and improved the dynamic range as well.
Bit depth, sensitivity and dynamic range are totally independent. I can represent an arbitrarily large dynamic range in one bit. Nobody said it would look pretty, but it would be mathematically and optically correct. More bits just make it "look better".
I'm talking about ADC characteristics. 16-bit monolithic ADC should provide theoretically 96 dB DR and 14-bit 84 dB DR. The SFDR improves considerably. The additional 2-bits should provide at least another stop in sensitivity.
rsg_1 wrote:
I'm talking about ADC characteristics. 16-bit monolithic ADC should provide theoretically 96 dB DR and 14-bit 84 dB DR. The SFDR improves considerably. The additional 2-bits should provide at least another stop in sensitivity.
the A/D is too good for the sensor, the sensor doesnt match the capabilities of 14 or even 16Bit ... imagine someone who thinks he needs a 32bit/96khz audio-ADC to record his old dolby-b tapes from 1975... its nonsense.
ChrisDM wrote:
A couple of misconceptions here. First off, diffraction has nothing to do with pixel count. It is a function of the lens, and all lenses have it. The greater the resolution of the sensor the more visible the flaws of the lens will be. So, stopping down to f16 is not "useless", even if there is diffraction. Also, the sensor being able to "outresolve lenses" as you say, is not a bad thing. It just means the sensor is capable of taking advantage of even better lenses, and ultimately more detail.
Here's an example of diffraction:
Here's the scene I had to shoot, very high depth of field to keep the front corner of the couch all the way to the back wall in focus:
So I had to stop down to f18 to maintain DOF, but suffering from diffraction. But what if I shot at f11 to avoid diffraction? That's right, a portion of the scene becomes out of focus:
So is using f18 on my 1Ds3 "useless" because of diffraction? Of course not, as this example proves. Yes, all lenses (not sensors) suffer from diffraction at narrow apertures. Does this make them "useless" at narrow apertures? No. We shoot at the aperture required to produce the depth of field necessary to get the shot... Does this mean that we should avoid sensors with high resolving power so we don't see the flaws inherent in the designs of the lenses we choose? Of course not.
Chris M
www.imagineimagery.com
Edited by ChrisDM on Aug 26, 2008 at 05:31 AM GMT...Show more →
Isn’t it true that the aperture that diffraction begins to affect image quality has to do with the sensor/film size. So comparing the 50d crop to a 1ds 35mm sensor is not a fair comparison, bigger is more forgiving. That is the reason large format shooters routinely use apertures like f32 and f64 without diffraction issues.
Paul Tessier wrote:
Isn’t it true that the aperture that diffraction begins to affect image quality has to do with the sensor/film size. So comparing the 50d crop to a 1ds 35mm sensor is not a fair comparison, bigger is more forgiving. That is the reason large format shooters routinely use apertures like f32 and f64 without diffraction issues.
Well, it's not so much sensor size as it is pixel size/density, and the 1Ds3 has relatively small/dense photosites. Granted the 50D will probably be even moreso, but the point is that diffraction is more a function of the lens than the pixels/sensor, and is only a consideration when producing very large prints anyways.
The big news, if Canon's claim is to be believed, is that there are no gaps between the micro-lenses. That means there is no room for improvement in that parameter of the noise problem. It also means what I have been banging on about for quite a while: noise PER IMAGE is now completely independent of pixel density. Taking this sensor design to 50 megapixels doesn't introduce any extra noise PROVIDED THE LIGHT WELLS CAN PERFORM AS WELL when scaled down. Of course, this is a gross simplification, because micro-lenses are not 100% efficient, but in practical terms hopefully this means that we no longer need to hear people being misled by the false dogma that high pixel density is intrinsically noisier than low pixel density. Now it's just about the sensor area, the light well silicon, and noise reduction technology. With this sensor design, bigger sensors are better, regardless of how many megapixels you want to capture. In fact, with this design of sensor, the 1Ds4 should be better at iso 12800 than the 1D4, due to the 1.3 crop. The megapixel war should now re-commence with a new ferocity, which I welcome, as 21 is still not enough. I am looking forward to 48, 24 16 and 12 megapixel raw files available from the same camera, depending on your mood, and iso 25600.
brainiac wrote:
The big news, if Canon's claim is to be believed, is that there are no gaps between the micro-lenses. That means there is no room for improvement in that parameter of the noise problem. It also means what I have been banging on about for quite a while: noise PER IMAGE is now completely independent of pixel density. Taking this sensor design to 50 megapixels doesn't introduce any extra noise PROVIDED THE LIGHT WELLS CAN PERFORM AS WELL when scaled down.
Your logic is somewhat flawed. For a given per-sensel fill factor (percentage of each sensel's area that is optically active), increasing sensel count has never impacted noise at the image level. Only sensel-level noise increases and the image-level noise becomes more fine grained.
Unfortunately, keeping fill factor constant isn't easy because the transistors and interconnects associated with each sensel take up space. For a given technology generation, they take up the same area regardless of sensel pitch. So, more sensels means shrinking the optically-active area while keeping the electronics area the same and thus reducing the fill factor. This does increase image-level shot noise, which impacts midtones and highlights. Deep shadow noise is limited by read noise. Here, higher sensel count can be favorable because per sensel read noise can be relatively independent of sensel pitch. More sensels means better noise averaging over any given area of the image.
The gapless microlens design of the 50D sensor doesn't imply any benefit toward future scaling per se. It's more of a one-time improvement.
There is still room for future improvement in light collection, however. For example, backside optical detection places the photodiode below the electronics, illuminated from the wafer backside. This allows the electronics and interconnects to consume effectively zero area.
... what makes you think so? more bits just mean that more bits will be lost in noise. We only will have bigger files and no benefit at all. Even the 14bits are already too much as some clever people found out.
post of the month.
