williamz wrote:
Basically, there is a small amount of random 'signal' reported by each sensor pixel. Any light falling on that pixel will be added to the random signal to give you the signal recorded by the camera.
The more light that falls on to one of the pixels, the smaller the fraction of the overall signal that is noise, therefore the less noticable the random variation is.
Noise exists in several types:
- thermal random noise produced by the electronics in the pixel (depends directly on the ambient temperature),
- read noise produced by the circuitry used to find out the number of electrons in the pixel (resulting in a voltage),
- a variable gain is applied to the voltage to get different ISO settings (the amplifier also introduces some noise of its own),
- quantization noise when the voltage is translated into digits (more bits means less noise),
- photon noise - photons come randomly so if you measure very few of them, you cannot really be sure what the average flow is (increases when pixels - or actually microlenses - decrease in size and there is nothing you can do to it, if all the other parameters remain the same).
The first three noise types have been decreasing due to advances in microelectronics. Quantization noise was decreased by moving to 14 bits. Photon noise increase has been battled with increasing the microlens coverage.
astrolucida wrote:
Actually, it will begin above f7, where the size of the Airy disc equals four pixels (RGBG set). The Airy disc is the smallest circle to which a perfect lens will project an incoming pinpoint light source. In practice, it will be hard to distinguish this effect with non-test targets until maybe f11 and then only through flip-comparing two equivalent images.
With a 30D (8MP), my experience shows that f16 is still very good but f22 no longer looks good. The theoretical limit for the 30D is f9.6. This would mean that f11 would still be OK with the 50D.
So stopping down to f16 is useless like I said for which I got a lot of unpleasent comments.
No, it's still not useless. Yes, image quality will suffer for the in focus portions at f/16 vs. f/11. However, if you need massive depth of field, having a slightly softer, but still in focus f/16 is a lot better than having an important part of your image out of focus at f/11. The softening from diffraction is still less than the softening from being outside the DOF. I've taken shots at f/22 that look fine. Sure, they're a little softer than at f/8, but at f/8, the background would be OOF, and at f/22 it's not.
John Power wrote:
Would someone explain the physics of this to me. Are these pixels in anyway compromised when compared to the 8.2 I have on my 30D?. I thought that when you increased MPs on a small sensor there was a degradation of IQ
Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
But Canon gives you live view, padded CF card doors, bigger LCD's, silent shutters and higher frame rates to help you forget about the fact your 40D photos don't look as good as your 30D photos. And I suspect the 50D files will have more noise & less dynamic range than the 40D files.
Canon should have stopped "improving" the XXD series with the 20D.
Mike Mahoney wrote:
Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
More noise per pixel, but not in the final image-- and at that only if there's no technological improvement, which there clearly has been since the 20D sensor.
But Canon gives you live view, padded CF card doors, bigger LCD's, silent shutters and higher frame rates to help you forget about the fact your 40D photos don't look as good as your 30D photos.
Proof?
Canon should have stopped "improving" the XXD series with the 20D.
Nah. The improvements are welcome. You also left out many useful improvements-- perhaps you failed to consider them?
Mike - perhaps you should look at the massively enhanced technology with the 50D. The gapless microlenses, along with greater area of photosite covered with light means that the 50D actually has about the same amount of light per well as the 40D, but with improved processing and CMOS technology. This is a winner of a sensor.
Mike Mahoney wrote: Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
Repeat after me, without thinking what it really means : Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
You must have read that on the internet somewhere and accepted without question. It is not a law of physics! If the gapless microlens light-collecting surface gets as many photons to the photodiode in the new sensor as the smaller lens did to the older, more widely spaced pixels in the prior sensor(s) then the signal will be the same even if they didn't reduce the "noise" component on the chip or in the processing. This is called incremental progress - the pixels from an older generation sensor cannot be directly compared based on "size" with those of a newer one so the mantra fails
Thanks for the info .. I'm not a theorist nor an electronics engineer, but do use 20,30, and 40D cameras on a daily basis. Probably more than most on this forum.
And can tell you that if I have to shoot a scene with a high contrast such as a meeting room with bright lights and dark floors, or a wedding photo with a bright white dress and deep black suits my 30D's will hold the highlights and keep the shadows better than my 40D's.
So if this is Canon's idea of "incremental progress" they can keep it.
Mike Mahoney wrote:
Thanks for the info .. I'm not a theorist nor an electronics engineer, but do use 20,30, and 40D cameras on a daily basis. Probably more than most on this forum.
And can tell you that if I have to shoot a scene with a high contrast such as a meeting room with bright lights and dark floors, or a wedding photo with a bright white dress and deep black suits my 30D's will hold the highlights and keep the shadows better than my 40D's.
So if this is Canon's idea of "incremental progress" they can keep it.
jvarszegi wrote:
Score one anecdote for Mike Mahoney.
As amusing as this classroom discussion of thermal random noise, electrons, variable gain, photon noise, & microlenses has been, I thought a little reality was in order.
Can someone show me the beef by posting links to high ISO 50D CR2 files?
Talk about it all you want, but there is no way you can cram 50% more pixels onto the same sensor size and retain the same native dynamic range and noise characteristics.
But I will ask how much sensor real estate was saved by the reduction in the microlens gap?
Mike,
Glad you enjoy the banter about test results on a camera that's not been throughly tested yet - I'm looking forward to the test results too! It'll be interesting to see how much the improved photon-gathering of the gapless microlenses actually boosts the DR to offset the increased number of pixels. BTW I also still have, and love, my D30 with its huge pixels
Mike Mahoney wrote:
As amusing as this classroom discussion of thermal random noise, electrons, variable gain, photon noise, & microlenses has been, I thought a little reality was in order.
Can someone show me the beef by posting links to high ISO 50D CR2 files?
Talk about it all you want, but there is no way you can cram 50% more pixels onto the same sensor size and retain the same native dynamic range and noise characteristics.
But I will ask how much sensor real estate was saved by the reduction in the microlens gap?
Wrong, sorry. But I am still waiting for the proof of your claims about 40D files being actually worse than 30D files.
Mike Mahoney wrote:
But Canon gives you live view, padded CF card doors, bigger LCD's, silent shutters and higher frame rates to help you forget about the fact your 40D photos don't look as good as your 30D photos. And I suspect the 50D files will have more noise & less dynamic range than the 40D files.
What are you talking about?! 40D image quality is far superior to the 20D, not to mention the far better AF. If you can't see that, then you should have taken your 20Ds back to the store and get it exchanged
Mike Mahoney wrote:
And can tell you that if I have to shoot a scene with a high contrast such as a meeting room with bright lights and dark floors, or a wedding photo with a bright white dress and deep black suits my 30D's will hold the highlights and keep the shadows better than my 40D's.
Pondria got a few to do a consistent test on the dynamic range of a range of cameras and the 40D was an improvement over the 20 and 30D. And it didn't even include the jump to gapless microlenses
Mike Mahoney wrote:
Smaller pixels (pixel sizes are measured in pitch) mean less dynamic range and more noise.
Ack! When will people stop posting this fallacy?
1) Since gaps between microlenses are very small, increasing the number of pixels has very little effect on the NOISE PER IMAGE, since the noisier pixels are proporionately less significant in the formation of the picture, and random noise averages out. That means that with the ideal gapless sensor we are approaching, sensor size matters more than ever, and pixel density matters less than ever, if at all. In short, smaller pixels DO NOT MAKE significantly MORE NOISE IN THE IMAGE, even though each one is noisier.
2) Dynamic range is mainly set by well capacity. It depends on the sensor technology. With any given technology, while a smaller well has less capacity, and therefore less dynamic range, it also gets fed by a smaller micro-lens, so dynamic range should also be fairly independent of pixel pitch, at least in the highlights. The various noises contributed by the device will imply a lessening of dynamic range in the shadows, but those noise contributions are fairly low already, so loss of dynamic range isn't that dramatic either. The 1Ds3 seems to have slightly less dynamic range than the 1D3 (but also less noise). Cunning techniques, like temporarily reducing the resolution of a sensor by counting fillings of half the wells in the time domain and using the other wells normally for shadows/tonal gradation could mean that a 15 megapixel sensor could double up as a massive dynamic range sensor at 7.5 megapixels. We have yet to see that technology, but potentially it is possible. In other words, future cameras might well be able to have a slider with which you trade resolution for dynamic range. There's no law against it.
At any rate, learn the new gapless micro-lens mantra: higher pixel density does NOT entail significantly more noise in the image, even though each pixel is noisier.
brainiac wrote:
Ack! When will people stop posting this fallacy?
1) Since gaps between microlenses are very small, increasing the number of pixels has very little effect...
and
...higher pixel density does NOT entail significantly more noise in the image, even though each pixel is noisier.
Your equivocal language does little to counter the claims about noise. "Little effects" are still effects and your "NOT entail significantly more noise" concept does acknowledge the possibility of increased noise - though the question of its significance could be open to debate.
The "in the image" noise point would be valid if the noise produced by individual photosites was somehow average with the noise of the surrounding photosites, but unless this is happening there would, indeed, be variations in the signal produced at each photosite that would increase with smaller photosites/pixels. This is not insignificant, as you probably know if you have ever done certain levels/curves adjustments on an area of a photo that is primarily of one color -sky being a great example - and ended up with blotchy effects from noise.
