Has Canon fallen hopelessly behind? (cross-posted)

[ErikKaffehr]

Hi,

I'm not exactly of the opinion that Canon DR at base ISO is sufficient. It's more like that I essentially feel that most cameras I had did have enough DR for most situations.

On the other hand, Fred Miranda's recent comparison of the Canon 5DIII and the Nikon D800 clearly indicated that the D800 has real advantages in that area.

All my cameras have been Sonys and Minoltas.

Best regards
Erik

There seem to be many photographers who think the Canon DR at base ISO is sufficient. Erik Kaffehr of this site, for example, and many studio workers who can control their lighting.

...

I personally am more interested in improved dynamic range, improved image stabilisation, improved 'merge to HDR' software, and improved stitching software that can overcome the lack of a pano head.

[BJL]

The patent for this type of technology was issued a long time ago, 12 years ago to be precise. I believe it's held by a Chinese person, Peter Hong Xiao. Refer attached link. http://www.google.com/patents/US6137432

it is not just the patents that matters, and certainly not a single patent. By the way, that patent is not launching the idea of a column parallel ADC itself, but is on one particular approach to the idea, with the key words being "low power". Note also the prior art cited there from Stanford and Kodak. That 1997 Kodak patent might in fact be the real starting point! http://www.google.com/patents/US5613156

Multiple companies have been making sensors with column-parallell ADCs for some years, including Samsung, though primarily for video cameras, where the advantages are perhaps even greater. My guess is that Panasonic also uses column-parallel ADC in its more video-oriented sensor for the GH2, but Panasonic only says that that sensor produces a digital output signal rathe than the analog signal form other Panasonic 4/3" sensors. A search will reveal numerous papers on improvements in column-parallel ADC technology form a wide array of authors (mostly Korean and Japanese lately, it seems) so it is far form being controlled by one company or one patent holder.

No: it is the quality of implementation of the new technology that matters more. For example, the first Sony efforts only gave 12-bit output, with Nikon also offering 14-bit output but only at a very reduced frame rate. This parallels what happened a few years ago with active pixel CMOS sensors: Canon did not invent the active pixel CMOS sensor (Eric Fossum at the JPL did), but even when Sony and Panasonic followed Canon in adopting that active pixel CMOS technology for DSLRs, Canon for some years had a lead in the quality of its implementations.

[Justinr]

I must admit that I am a little remiss in not noting the figures you mention, however, when I was asked to shoot a candle lit wedding in the evening three days before Christmas I was fully confident in the cameras ability to handle the situation along with judicious use of bounced flash. I warned the bride that there would be a degree of noise and she was comfortable with that. The Pentax did not disappoint.

[Ray]

it is not just the patents that matters, and certainly not a single patent. By the way, that patent is not launching the idea of a column parallel ADC itself, but is on one particular approach to the idea, with the key words being "low power". Note also the prior art cited there from Stanford and Kodak. That 1997 Kodak patent might in fact be the real starting point! http://www.google.com/patents/US5613156

Multiple companies have been making sensors with column-parallell ADCs for some years, including Samsung, though primarily for video cameras, where the advantages are perhaps even greater. My guess is that Panasonic also uses column-parallel ADC in its more video-oriented sensor for the GH2, but Panasonic only says that that sensor produces a digital output signal rathe than the analog signal form other Panasonic 4/3" sensors. A search will reveal numerous papers on improvements in column-parallel ADC technology form a wide array of authors (mostly Korean and Japanese lately, it seems) so it is far form being controlled by one company or one patent holder.

No: it is the quality of implementation of the new technology that matters more. For example, the first Sony efforts only gave 12-bit output, with Nikon also offering 14-bit output but only at a very reduced frame rate. This parallels what happened a few years ago with active pixel CMOS sensors: Canon did not invent the active pixel CMOS sensor (Eric Fossum at the JPL did), but even when Sony and Panasonic followed Canon in adopting that active pixel CMOS technology for DSLRs, Canon for some years had a lead in the quality of its implementations.

All that may be true. I can't dispute it. There are many ways of skinning a cat. Sony may well have a clear advantage in a particular type of miniaturisation of a process which Canon has decided not to emulate for various reasons which might include unattractive trade-offs in cost and frame rate, for example.

I will get back to my analogy of the frame rate of the P&S camera. Is there any fundamental technological or patent impediment why a P&S camera with 12 fps could not be offered, and a bracketing capacity of 5 or 9 frames with a 1EV interval? Such a facility would be tremendously useful for merging to HDR and improving the inherently poor noise characteristics of the P&S..

[hjulenissen]

The main principle in reducing noise, as you probably already know, is to boost the analog signal from the sensor prior to A/D conversion and all other signal processing, not so that over all noise is reduced in absolute terms, (in fact it must be increased to some degree at the time of analog amplification), but so that the noise is less as a proportion of the signal. That is, SNR is improved.

Boosting signal _before_ a noise source contributes to SNR. Boosting signal _after_ a noise source does not improve SNR (wrgt that noise source at least). If the signal is noisy early in the signal chain, no amount of amplification later on will improve things.

If you want a good SNR, you want to have a large signal and little noise. I think that what Sony/Nikon did is the opposite of what you are suggesting: they reduced the noise contributed by electronics.

-h

[hjulenissen]

All that may be true. I can't dispute it. There are many ways of skinning a cat. Sony may well have a clear advantage in a particular type of miniaturisation of a process which Canon has decided not to emulate for various reasons which might include unattractive trade-offs in cost and frame rate, for example.

The most interesting internet hearsay that I have read was that Canon have invested heavily in manufacturing capacity that limits their development. Surely, this investement must have secured them large amounts (or larger dimensions) of sensors at sensible prices.

I will get back to my analogy of the frame rate of the P&S camera. Is there any fundamental technological or patent impediment why a P&S camera with 12 fps could not be offered, and a bracketing capacity of 5 or 9 frames with a 1EV interval? Such a facility would be tremendously useful for merging to HDR and improving the inherently poor noise characteristics of the P&S..

suggestions:
Processing bandwidth, bus bandwidth, storage bandwidth. Sensor thermals. Contrast-based AF.

Lack of interest among typical customers.

-h

[Ray]

Boosting signal _before_ a noise source contributes to SNR. Boosting signal _after_ a noise source does not improve SNR (wrgt that noise source at least). If the signal is noisy early in the signal chain, no amount of amplification later on will improve things.

If you want a good SNR, you want to have a large signal and little noise. I think that what Sony/Nikon did is the opposite of what you are suggesting: they reduced the noise contributed by electronics.

-h

I understand what you are saying, and I did mention in my post that boosting an analog signal must also boost any inherent noise in that signal. That seems clear.

Nevertheless, the high-ISO noise in the 5D3 is on a par with that of the D800, so that boosting of the analog signal is clearly working for Canon, despite the fact that the initial boosting also bossts the inherent noise at the instant of capture.

[Ray]

Processing bandwidth, bus bandwidth, storage bandwidth. Sensor thermals. Contrast-based AF.

Lack of interest among typical customers.

-h

You want your cake and eat it too? 

Your first objections are not insurmountable. The lack of interest amongst typical customers is more likely the real reason.

[Ray]

Hi,

I'm not exactly of the opinion that Canon DR at base ISO is sufficient.

Sorry! I must have got the wrong impression. You have mentioned several times that you haven't felt the need for greater DR than your A900 provides.

[BJL]

Decisions as to trade-offs between frame-rate and  pixel count are marketing decisions. 5D3 owners are pleased that there 22mp sensor has a faster frame rate than Nikon's 36.3mp.

True, and that is a quite legitimate factor: indeed, I suspect that this is the only real reason that a camera maker ever offers a new sensor with less resolution than another it already has in the same format, or another that it could have offered instead, and that
"less noise at high ISO (but only if you measure per pixel or view at 100% on screen, so that our images are viewed smaller than the competition's)"
is mostly a marketing smoke-screen, blown away by DX0's "print" normalized measurements.

But did you miss my example about the far less expensive Sony A77 having both a far higher frame rate (12 vs 6) and somewhat higher pixel count (24 vs 22) than the 5D3? Do you seriously propose that "lack of interest" dissuaded Canon from offering either a higher frame rate than 6fps at the same 22MP, or a higher pixel count at the same 6fps?

It is getting very hard to pretend that the 5D3's disadvantages are entirely about Canon's choices and not at all due to a (temporary) technological disadvantage.


By the way: Eric Kaffehr's A900 is a Sony with column-parallel ADC; the sensor that moved Sony and Nikon ahead of Canon on DR. SO I do not see why you take his satisfaction with its DR as evidence of satisfaction with the DR of Canon's sensors.

[Ray]

By the way: Eric Kaffehr's A900 is a Sony with column-parallel ADC; the sensor that moved Sony and Nikon ahead of Canon on DR. SO I do not see why you take his satisfaction with its DR as evidence of satisfaction with the DR of Canon's sensors.

The DR of the A900 is only 1/2 a stop better than the 5D3 at its base ISO. That's the minimum improvement that DXO considers significant. Above base ISO the 5D3 excels. At ISO 3200 it's 1 & 2/3rds stops better than the A900. That's a huge difference.

[MatthewCromer]

The DR of the A900 is only 1/2 a stop better than the 5D3 at its base ISO. That's the minimum improvement that DXO considers significant. Above base ISO the 5D3 excels. At ISO 3200 it's 1 & 2/3rds stops better than the A900. That's a huge difference.

The Alpha 900 is a four year old camera, not a four day old camera.  It has extremely dense color filtration for better color purity, the 5D2 and 5D3 have some of the least-dense color filters of any dSLRs.

It also doesn't have banding in the deep shadows like Canon -- that difference isn't captured by the DxO Mark scores.

[BJL]

Above base ISO ...

Ray, you have an amazing ability to change the subject in an tempt to win an argument. Eric, and I, are clearly talking about maximum possible dynamic range, with proper exposure, meaning at base ISO speed (or whatever ISO speed setting maximizes it.) That is, the dynamic range that the sensor is capable of, between full well capacity and the noise floor.

Indeed, "dynamic range at elevated ISO speed", where the maximum well capacity is not even being made use of, is rather an abuse of the established meaning of "dynamic range". Once one is significantly above base ISO speed, so-called DR is essentially just another measure of the dark noise floor.  But that is another topic that I should probity not get into here.

[Ray]

Ray, you have an amazing ability to change the subject in an tempt to win an argument. Eric, and I, are clearly talking about maximum possible dynamic range, with proper exposure, meaning at base ISO speed (or whatever ISO speed setting maximizes it.) That is, the dynamic range that the sensor is capable of, between full well capacity and the noise floor.

Indeed, "dynamic range at elevated ISO speed", where the maximum well capacity is not even being made use of, is rather an abuse of the established meaning of "dynamic range". Once one is significantly above base ISO speed, so-called DR is essentially just another measure of the dark noise floor.  But that is another topic that I should probity not get into here.

I'm not changing the subject at all. I'm rational and focussed. My point all along has been that Canon could deliver better DR at base ISO if it were to boost the analog signal in a manner similar to its boosting of the underexposed signal, at high ISO.

In order to do that they would need larger and more robust processors and D/A converters etc which would no doubt incur some disadvantages which the marketing department decided against.

[BJL]

My point all along has been that Canon could deliver better DR at base ISO if it were to boost the analog signal in a manner similar to its boosting of the underexposed signal, at high ISO.

So the 5D3 could have been a better camera if it had been done differently? That is a vacuous truism in the lack of evidence that Canon could have done so. Instead, the new highest resolution model in the Canon line-up is inferior in (maximum) dynamic range, and in resolution, to a discontinued four year old model. (Which to add insult to injury, cost significantly less.)

I will repeat: there is a very likely technological explanation for Canon not improving DR to the level already offered by the competition years ago: with its continued use of off-board ADC technology, extra noise is introduced in the additional fast transmission of the analog signal along the sensor's edge and then to the ADC, and if the extra amplification used at high ISO speeds to overcome that noise were applied at lower ISO speeds, it would produce signals too strong to be handled by the transmission between ISO amplifier to ADC. If not, why on earth would Canon not use that extra amplification and improve the DR? I cannot see any major cost factor. The fundamental difference in almost certainly sensor design and different ADC approaches, not just marketing decisions like "DR does not matter so much any more in out highest resolution camera."

Aside: weren't you the one who, a few years ago, went around outside your house with a light meter, proving the occasional need for about 15 stops or more or DR?


But I should give up: it is futile fighting "my imaginary camera is better than your real camera" hypotheticals: you are clearly in full-on brand-loyalty obfuscatory defense mode!

[MatthewCromer]

. . .which Canon has decided not to emulate for various reasons which might include unattractive trade-offs in cost and frame rate, for example.

Which is why the $900 24MP Alpha 65 shoots at 10 FPS with DR equal to the 5D3, right?

[BJL]

Which is why the $900 24MP Alpha 65 shoots at 10 FPS with DR equal to the 5D3, right?

Oh yes, I was behind the times withe the 12fps 24MP $1400 A77 now that there is the $900 10fps 24MP A65.
Which pretty much brings the arguments down to cost: maybe Canon could do it, but at a higher cost than the competition is currently offering --- which is a manifestation of being behind technologically.

In the case of the A65 and A77, there is another obvious technological factor: by eliminating the flipping mirror with its SLT approach, Sony can bring the cost of high frame rates (with AF t every frame) way down. The expected 36x24mm format Sony SLT will be interesting in that regard: if 12fps @ 24MP is possible, so should be 8fps @ 36MP, and maybe a 36MP sensor could offer a 16MP to 24MP crop mode at 12fps or more. That could shake up both Nikon and Canon a bit, and make them think more seriously about ditching the flippin' mirror!

I can see that Canon has good reasons for choosing 22MP rather than higher (better video, insufficient IQ benefit to most customers of higher resolution, etc.) but I do to see any good reason for not then getting the expected "per pixel" quality advantage over its competitors from those fatter pixels.

[MatthewCromer]

But I should give up: it is futile fighting "my imaginary camera is better than your real camera" hypotheticals: you are clearly in full-on brand-loyalty obfuscatory defense mode!

Doesn't Ray own at least one Nikon dSLR?!

[MatthewCromer]

The expected 36x24mm format Sony SLT will be interesting in that regard: if 12fps @ 24MP is possible, so should be 8fps @ 36MP, and maybe a 36MP sensor could offer a 16MP to 24MP crop mode at 12fps or more. That could shake up both Nikon and Canon a bit, and make them think more seriously about ditching the flippin' mirror!

I'm completely sold on EVF/SLT for legacy SLR lens mounts and wouldn't consider another dSLR myself, but it seems that most photographers with lots of disposable income for purchasing aspirational photo gear are in love with optical viewfinders.

Sony could release a FF 30FPS 36 MP camera for $1000 and the majority wouldn't consider purchasing it.

Meanwhile Nikon releases a D800 that basically offers the technology of the F5 with a really nice (Sony!) digital sensor and a nice (albeit fixed) LCD and the majority of photographers think it's the second coming.

I think this is, in large part, because serious photographers are typically over 40 or 50 and pretty well set in their ways.

[ihv]

When looking at the processing speed then the D800 is doing little bit faster than the 5D3: 4 * 36 = 144 MP/s > 6 * 22 = 132 MP/s
While maybe not a big deal, the flash sync is also better on the D800 side: 250 vs 200.

Which kind of gives impression Canon held back in (too) many departments, yet the price suggests a premium product over the competition.

Nevertheless knowing about the D800 Canon gave green light to the 5D3 and generally speaking it sort of looks doing fine.

Now with the 1Dx the story looks quite opposite, I imagine it is the competition which:
1) forced for a very early in advance announcement (very unusual for Canon announcing cameras)
2) delayed the final product even more