1 Series replacements?

Is it to early to start speculating about the 1 Series replacements again?

Yup. Too soon.

Andy Biggs said:

Yup. Too soon.

Click to expand...

It looks like Canon has no need to release their 22MP 7 frames/minute flagship camera because there's no camera from 35mm competitors or MF upscale alternates to worry about right now.

Canon does not need to rush.

I would imagine they are using the time to try to increase the BIT rate. Otherwise the cameras are fine. Somewhere between now and March I'd guess. Later if they want to go for higher BIT rate and a larger LCD.

Asher

I don't understand the need to upgrade and the anxiety if there isn't a camera with more megapixels in the horizon.

Somewhere I read that there is a curve of megapixel "output" in the sensor as the Y and actual resolution as the X. and, in case of the Canon the plato is very near or there already.

What I think that it means is that the jump in actual resolution from 12mp to 16mp was very productive in terms of actual res., -- so the myth was born-- but that won't happen necessarily when going from 16 to 22, or even 29.

One of the problems may be, that 16mp is close to the resolution of 35mm optics, and since Canon photographers also prefer to use zooms, then the situation is even worst.

The other problem is that sensel size has to be reduced even further to acomodate more of them in the same real state, so latitud, detail in the shadows and in highlights negates gains in resolution.

This happens even in MF backs. Some photographers, after paying 15k to 20k to quench the upgrade thirst go from 22mp to 39mp just to find out that the sensels of the new back is 6.8 microns and the old 9 microns, so the IQ is not better, only larger in size.

On the other side, do we want a Canon that shoots 200MB files 8fps? so that every second we would have to deal with 1.6Gigas? You would en up with Terrabites and Terrabites of files that have enormous "pixel" resolution, but bad real IQ.

Leonardo,

It all depends on the actual physics limits of noise and ability to get good mathematics of counting small numbers of photons by measuring the elctrical charge accumulated above the noise.

I have yet to see any report of the actual theoretical limits we are approaching. Certainly, lens resolution limits have not been reached especially given the increasing sophistication of lens correction software and advances in lens materials.

The best of Zeiss, Nikon, Olympus and other wide angle lenses need to be considered too.

Asher

Asher, it may be a matter of relative quality. For example, the new batch of compact Panasonic cameras are having an "upgrade" in pixel count, even with new generation of image amplifier the result is not a better image but only one with larger file size.

There may be an increase of image resolution in the new 22-30mp Canon generation, but the return for the "effort" will be diminishing until it is not economically, or practical to continue and photographers will accept the new format for what it is. If they want more resolution they can always get if from a larger size CCD. We have to remember that MF is going to implement all the advances in signal amplification with an advantage in the physical reality of a larger real state from where to collect photons in a given time.

Probably Kodak has that 22-30mp sensor and run tests with Canon and encounter this problems.

I can't wait until we reach this limits and upgrade at a saner pace.

Sprinkle in some new features: Smaller, faster, anti-dust, body antishake, live focus, high-rez chimp screen ,remote EVF, Wifi, Bluetooth, post-shot lens aberration correction, higher ISO, better DR. Yes, yes, a few more MP will help the marketing department too.

Edmund

I am not sure where the practical limit is for pixel densisty (the opposing value of number of pixels versus sensor size), but I do know the physics limits for detector size.
And note with micro lenses that would have to be bigger then this for the lens to work.
And also note there is a signal to noise component at work meanning they have to be even larger.

But the physics limit is the wavelength of light or you can't get smaller then 750nm for detector size and still see all the red and 450nm ish* and still see all the blue. Some of the point and shoots are down to 2500nm in detector size, while some of the DSLRs are running in the 5-6000nm range.

Some image scientists from Kodak (Government sensor division - now ITT) I once talked to said that 8000nm was the optimum size for small pixels versus SNR, below that noise starting becoming a problem, but that they figured someday we would be able to get to somewhere around 2800nm and possibly as small as 1400nm before we would have so much noise as to completely drown out the signal. They said that below 2800 they expected noise to always show up in any picture as captured. This was told to me about 6 years ago now and still seems to be valid.

Or so I have been informed,

* I didn't feel like looking up the blue to green line.

Edmund Ronald said:

Sprinkle in some new features: Smaller, faster, anti-dust, body antishake, live focus, high-rez chimp screen ,remote EVF, Wifi, Bluetooth, post-shot lens aberration correction, higher ISO, better DR. Yes, yes, a few more MP will help the marketing department too.

Edmund

Click to expand...

Edmund,

Great idea on the remote EVF! Lens aberration correction is a no brainer. Could even special series of lenses with their specifications for exact corrections.

DR is the most important!

Asher

Tim

Interesting information, if we where to apply this to actual cameras, how would the 1DsMk2 be? I think that digital backs have sensels of 9x9 microns, is that 9000nm ?

Do we know the maximum size that a Canon 1DsMk3 would have to have in microns in case they want to go 22MP of 39MP? I know that the P 45 is 6.8 Microns @ 39MP. The Canon chip is half the size, so the microns would have to be halved = 3.4 Microns. That seams to be conformably over the "no way" line, but the image purity, if compared to a 9 Microns sensel size sensor array -- like a P 25 -- would make them beasts of two different species.

It would be like comparing a Panasonic 10MP against a D200 10MP. In other word, pixel count is just like cylinder count in an automobile analogy, or comparing wines by alcohol %...

Tim Dolan (Longwatcher) said:

I am not sure where the practical limit is for pixel densisty (the opposing value of number of pixels versus sensor size), but I do know the physics limits for detector size.
And note with micro lenses that would have to be bigger then this for the lens to work.
And also note there is a signal to noise component at work meanning they have to be even larger.

But the physics limit is the wavelength of light or you can't get smaller then 750nm for detector size and still see all the red and 450nm ish* and still see all the blue. Some of the point and shoots are down to 2500nm in detector size, while some of the DSLRs are running in the 5-6000nm range.

Some image scientists from Kodak (Government sensor division - now ITT) I once talked to said that 8000nm was the optimum size for small pixels versus SNR, below that noise starting becoming a problem, but that they figured someday we would be able to get to somewhere around 2800nm and possibly as small as 1400nm before we would have so much noise as to completely drown out the signal. They said that below 2800 they expected noise to always show up in any picture as captured. This was told to me about 6 years ago now and still seems to be valid.

Or so I have been informed,

* I didn't feel like looking up the blue to green line.

Click to expand...

I'm afraid my eyes cross when I think about microns, and such. However, I can think of one good reason why I am cheering for a quick introduction of a new, higher end Canon 1 body. It will allow me to shop for a used 1dsMkII. My 1ds is getting a little beat up, though it's still plugging along. I love it dearly, but I'd like some of those new-fangled color histo-thingies, and some of that faster write speed.

Victor

leonardobarreto.com said:

Interesting information, if we where to apply this to actual cameras, how would the 1DsMk2 be? I think that digital backs have sensels of 9x9 microns, is that 9000nm ?

Click to expand...

The 1DsMk2 is approx. 7.2 micron sensel pitch. The actual sensel dimensions are slightly smaller than the pitch, because there is some separation between sensel structures. Yes, millimetres/1000 = microns (micrometres officially), microns/1000 = nanometres.

Do we know the maximum size that a Canon 1DsMk3 would have to have in microns in case they want to go 22MP of 39MP?

Click to expand...

They would approx. have a 6.2 or 4.7 micron pitch respectively, too small for my taste if dynamic range is important. That would be better served by a 9 micron sensel pitch, with a 16-bit ADC.

I know that the P 45 is 6.8 Microns @ 39MP. The Canon chip is half the size, so the microns would have to be halved = 3.4 Microns. That seams to be conformably over the "no way" line, but the image purity, if compared to a 9 Microns sensel size sensor array -- like a P 25 -- would make them beasts of two different species.

Click to expand...

Indeed, there is a trade-off between 'on sensor' resolution with its resulting output magnification, and Dynamic Range. The latter, Dynamic Range, benefits from larger sensels. Therefore, in order to satisfy both, large sensor arrays with many large (>8 micron) sensels are needed.

You may like this information. It attempts to explain the issues at hand, and the fact that current solutions are often photon noise limited in the signal/noise ratio that can be achieved, and that that photon noise limited S/N ratio also benefits from large area sensels.

Bart

"Because good digital cameras are photon noise limited, the larger pixels will always have higher signal-to-noise ratios unless someone finds a way around the laws of physics, which is highly unlikely.

Image detail can be blurred by diffraction. Diffraction is more of an issue with smaller pixels, so again cameras with larger pixels will perform better, giving sharper images."

That information is difficult to ignore. It may mean that people holding their breaths for a 1DsMk3 better than the Mk2 in the IQ level are waiting for the laws of physics to change.

I think that the camera can improve, and it should, but not by "upgrading" the Megapixel count, -- and splitting each sensel in two with half the size -- but by living that as is and concentrating in purifying the image, speed, etc.

Bart_van_der_Wolf said:

You may like this information. It attempts to explain the issues at hand, and the fact that current solutions are often photon noise limited in the signal/noise ratio that can be achieved, and that that photon noise limited S/N ratio also benefits from large area sensels.

Click to expand...

I don't agree with this "current (Canon) DSLRs are photon-noise-limited for dynamic range" stuff at all. Readout noise reduces practical DR by 2 to 4 stops, easily, depending on your standards of DR. Here's the interplay of noises in the Canon 5D at ISO 100:

5D noises ISO 100

The whole chart shifts one stop to the left for the red channel in daylight, two stops to the left for the blue channel in incandescent light - only the highlights and brighter midtones have mainly shot noise. As you move to higher ISOs, the chances of a conservative standard of dynamic range being limited by shot noise increases, but readout noise is still the main limiter for more liberal standards.

There are other improvements than resolution that one can hope for. Buffer depth, for one. I would love to have at least 30 frames in raw+S. It's about the only thing that I'd call limiting with the current 1Ds2 - 9 frames is not enough, even at 4fps, forget about 7fps (7fpm would be a good joke )

Bart_van_der_Wolf said:

Indeed, there is a trade-off between 'on sensor' resolution with its resulting output magnification, and Dynamic Range. The latter, Dynamic Range, benefits from larger sensels. Therefore, in order to satisfy both, large sensor arrays with many large (>8 micron) sensels are needed.

You may like this information. It attempts to explain the issues at hand, and the fact that current solutions are often photon noise limited in the signal/noise ratio that can be achieved, and that that photon noise limited S/N ratio also benefits from large area sensels.

Bart

Click to expand...

Bart,

A very interesting reference for sure. You found a treasure! In that article, the author states that for a sensel that can contain say 40,000 electrons (from receiving 40,000 photons), the error rate would according to Poisson distribution be the square rot of 40,000 if the well was full.

Well that is very small!

However, the same well measuring light from a shadow area might get only 100 photons and the error would be the sqare root of 100, ir 10%.

Now the % error is the square root of the number divided by that number x100.

So for 100 electrons counted, the error is 10%, obviously way to noisy!

As a result of this fact, the author gives a thumbs down to small sensors.

Well there's one way out of this. The brighter of the areas of the image can be accurately measured in the first microseconds, A smart pixel will then switch off.

The rest of the pixels will keep counting electrons until a at least say 3000 electrons have been counted. Or else 4 adjacent pixels in a shadow area will be binned to produce a single large pixel with a consequently higher signal to noise ration.

So with dynamic binning and early pixel shut off, and 16 BIT A to D convertors, I see a lot of capability of 5-6 micron pitch pixels.

Dynamic range will be expanded, IMHO, as soon as Canon delivers its new CMOS sensor with individually addressed pixels.

Just that Canon has not enough reason to give out advances when no one is breathing down their necks!

Asher

leonardobarreto.com said:

On the other side, do we want a Canon that shoots 200MB files 8fps? so that every second we would have to deal with 1.6Gigas? You would en up with Terrabites and Terrabites of files that have enormous "pixel" resolution, but bad real IQ.

Click to expand...

This is an excellent point. I hate even the files from 11 megapixel and 16 megapixel cameras. They're too large and slow to manipulate. They consume huge amounts of hard drive space.

And when I look back at studio shots from the old 5 megapixel D1X, I see little to no real improvement in IQ.

Asher Kelman said:

In that article, the author states that for a sensel that can contain say 40,000 electrons (from receiving 40,000 photons), the error rate would according to Poisson distribution be the square rot of 40,000 if the well was full.

Well that is very small!

However, the same well measuring light from a shadow area might get only 100 photons and the error would be the sqare root of 100, ir 10%.

Now the % error is the square root of the number divided by that number x100.

So for 100 electrons counted, the error is 10%, obviously way to noisy!

Click to expand...

Not very noisy for a shadow, but noisy for a highlight. Regardless, any capture of ~100 photons is going to have far more noise from the readout process at low ISOs. Readout noise is typically about 15 to 25 electrons/photons in intensity at ISO 100, and about 3.5 at ISO 1600 with current Canons. This renders the 10 photons of shot noise at ISO 100 very small in the overall noise ((10^2+20^2)^0.5 = 64) at that signal level. For ISO 1600, then the shot noise is more significant (10^2+3.5^2)^0.5 = 10.6.

John Sheehy said:

Not very noisy for a shadow, but noisy for a highlight. Regardless, any capture of ~100 photons is going to have far more noise from the readout process at low ISOs. Readout noise is typically about 15 to 25 electrons/photons in intensity at ISO 100, and about 3.5 at ISO 1600 with current Canons. This renders the 10 photons of shot noise at ISO 100 very small in the overall noise ((10^2+20^2)^0.5 = 64) at that signal level. For ISO 1600, then the shot noise is more significant (10^2+3.5^2)^0.5 = 10.6.

Click to expand...

Did you really mean " Readout noise is typically about 15 to 25 electrons/photons in intensity at ISO 100, and about 3.5 at ISO 1600 with current Canons." If so why?

The readout process is open to further optimization.

Asher

John Sheehy said:

I don't agree with this "current (Canon) DSLRs are photon-noise-limited for dynamic range" stuff at all.

Click to expand...

For general understanding, I'd like to stress the distinction between Dynamic Range (saturation signal level, divided by Read-noise), and Signal to Noise ratio. Those are different (though somewhat related) entities. So it would become "current DSLRs are photon-noise-limited for photon signals". Dynamic Range is only affected by the sensel's saturation level and Read-noise, while S/N ratio is affected by local signal level (thus photon noise) and Read-noise.

If we plot the actual recorded noise (e.g. standard deviation of a centre crop from a flat-field image) as a function of average signal level (mean Data Number for the same crop), it correllates quite nicely with the Read noise+Photon noise we can derive from a black frame+Square root of the average input signal. Again, here we are looking at Signal/Noise ratio.

Readout noise reduces practical DR by 2 to 4 stops, easily, depending on your standards of DR.

Click to expand...

Indeed, since Read noise is constant (independent of signal level) it will have more impact on low exposure levels, and as such limit the Dynamic Range by raising the noise floor.

Here's the interplay of noises in the Canon 5D at ISO 100:

5D noises ISO 100

Click to expand...

Yes, but that doesn't conflict with what was suggested. High exposure levels are mainly photon shot-noise limited, and low exposure levels are mainly Read-noise limited. Both play a role at the same time in images that test the full dynamic range.

Bart

Asher Kelman said:

Did you really mean " Readout noise is typically about 15 to 25 electrons/photons in intensity at ISO 100, and about 3.5 at ISO 1600 with current Canons." If so why?

Click to expand...

Yep; I really meant that. Readout noise is about 1.3 RAW levels (as a standard deviation) in the 1DmkII, and as high as 2.1 in the 20D at ISO 100, and there are about a dozen photons for each RAW level at ISO 100. They generally have around 4.7 for ISO 1600 (except for the Rebel, which is a bit higher) where there are about 3/4 of an electron for each RAW level.

The readout process is open to further optimization.

Click to expand...

That's an understatement. Readout noise is abysmal at ISO 100, considering how much less the signal is amplified. The noise loses a few stops of DR, easily. High ISOs aren't just to get a brighter review image for weak exposures; they are necessary to cleanly digitize the weakest signals.

Imagine a device like a sensor that could give an accurate count of photons collected, in units of the photons themselves, as a 16-bit number capable of recording 65,535 photons. Exposure compromises would be very simple (only absolute exposure, determines by scene illumination and Tvs and Avs, would affect noise). It would have 3 to 4 stops (maybe more) practical DR than what 50,000-photon ISO 100s are giving now.

Asher Kelman said:

Did you really mean " Readout noise is typically about 15 to 25 electrons/photons in intensity at ISO 100, and about 3.5 at ISO 1600 with current Canons." If so why?

Click to expand...

I'm in the process of doing a more detailed analysis of my 1Ds Mark II, and for that unit I record, e.g. for the Green filtered sensels, a Standard Deviation of 1.329 (in 12-bit scale) Data Numbers.

The ADC apparently uses an amplification (gain) of 1 DN per 14.154 electrons so a recorded StdDev of 1.329 equals 18.8 electrons of Read-noise at ISO 100 without any input signal.

At ISO 1600, I have to test that in more detail, it looks like the amplification is boosted in such a way that it generates 1 DN per 0.88 electrons (slightly more than unity gain), so a recorded Read-noise StdDev of 5.160 (in 12-bit scale) was caused by 4.56 electrons, and a lot of amplification.

The readout process is open to further optimization.

Click to expand...

Actually it is already very low in Canon cameras, but I assume you were tricked a bit by the electron count which gets reduced to a standard deviation of 1.329 to 5.160 in 12-bit Data Numbers, for ISO 100 and 1600 respectively.

What's really needed is larger sensels (or deeper potential wells) for a higher saturation level to boost the Dynamic range, but that would require larger sensels and better ADCs (14 or 16-bit). It apparently isn't easy to get deeper wells, so for a fixed sensor array size like 24x36mm there are currently only trade-offs between sensel count and Dynamic Range possible.

Bart

Probably the best alternative for Canon upgrade seekers is a new 1DsMk2+ and a 1DsMk3, the second with 22 or 39MP, and the first an improved 16MP model. The photographers that insist in having a "muscle" 35mm dslr could opt for the new one and upgrade --at a premium cost-- and for the purists that want a noiseless cleaner image Canon will get the one with "only" 16MP.

Thank heavens Leonardo came in and saved me from having to bring this up mid-photon/electron/particle physics discussion!

But there's a "new" rumor on Northlight (is it okay to mention other websites?) about an upgrage of the Mk2N, which would be the 1DMk3, according to the, ah, information culled and babel-fished from a Spanish website. If this were to turn out to be true (or in the ball-park of true), I would happily eschew my plans for a 1Ds (in whatever iteration) in favor of this:

17mp, 1.25x crop, 8.5 fps, RAW buffer fill to 25 images, anti-dust technology a la 400D (but improved), WiFi, better signal-to-noise ratio, and a lighter/higher cap. battery (slightly reducing weight) are the main points.

Yup, another internet rumor. This one, however, would just about nail down my "dream" 35mm SLR. I'd be able to retain the blistering speed of the Mk2N (can be VERY useful in photographing wildlife), would have my mind eased of the wide-angle problems of the FF Canons (for landscapes), and I'd have more mp's for more enlargability. With the (alleged) improved noise levels, maybe even more useful mp's. Dare to dream.

I have to believe that Canon has been all over the limiting factors of the more/smaller pixel situation, otherwise, why even bother with a 20 or 22 (or more) mp sensor?? Nobody wants the performance of their big expensive wonderful pro camera to drop, least of all Canon who would have to try and pitch that $8k flop to professionals. From what I've seen here (some amazing grasp of the minutia of the physical properties of sensors and processors!), there IS room for improvement. If mp's cannot be increased without losing DR or noise-limiting, why bother? They wouldn't, in my opinion. It would have to be improved proportionally, or they'd have an unsalable dog on their hands.

Anyway, this is one rumor that I hope could be true. It also shows up on a Korean website in a 12mp version, with everything else being approximately the same as above. This would also be a satisfactory version. I apologize if internet rumors are held in the deepest contempt, as on some other photog forums, but this was a rumor that actually got me excited.

Oh, and they also mention a 10-24 f/2.8 L, a BLACK 400 f/4 L IS, as well as a black 100-400 F/4 L IS (III), a 200 f/2.8 L IS, and a 24-70 f/2.8 L IS. And a weather-proofed 5D with 16 mp's.

Whoa, a 100-400 f/4L IS would be very cool indeed.

Nill
~~
www.toulme.net

And a weather-proofed 16 mp's 5D with at least 5fps would be awesome too.

Wow, you're right. What a nice wildlife kit those two would make!

Nill
~~
www.toulme.net

I wonder if any IQ improvement would be had if:

5D and 1DSmkII had no AA filter ala M8

and

They ship with moire' removal plugin for various raw converters or include it in DPP.

Harvey Moore said:

I wonder if any IQ improvement would be had if:

5D and 1DSmkII had no AA filter ala M8

and

They ship with moire' removal plugin for various raw converters or include it in DPP.

Click to expand...

Harvey,

You are quite right on the removal of the Moire filter. That would be a remarkable choice to give us.

Asher