October 22, 2014, 11:21:24 PM

Author Topic: Can Canon deliver a FF sensor that is class leading?  (Read 8954 times)

StudentOfLight

  • Canon 70D
  • ****
  • Posts: 319
  • I'm on a life-long journey of self-discovery
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #15 on: June 03, 2014, 05:04:41 PM »
Guys, didn't you read DxOmark? Just downsample your images down to 8x12" and print it. Then you gain at least 2 stops of dynamic range. And if 14-stops isn't enough then just downsample some more.  :P
Fantasy Gear:
TS-E: 45mm f/2.8 L-II,  EF: 40mm f/0.8,  100mm f/1.4,  35-85mm f/1.8, 
EF with 1.4xInt: 100-300mm f/4 ,  500mm f/5.6 L

canon rumors FORUM

Re: Can Canon deliver a FF sensor that is class leading?
« Reply #15 on: June 03, 2014, 05:04:41 PM »

ouuzi

  • SX60 HS
  • **
  • Posts: 2
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #16 on: June 03, 2014, 05:11:11 PM »
Marsu where do you download ML MINI ISO?

Marsu42

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4816
  • ML-66d / 100L / 70-300L / 17-40L / 600rts
    • View Profile
    • 6D positive spec list
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #17 on: June 03, 2014, 05:35:36 PM »
Marsu where do you download ML MINI ISO?

Atm unfortunately it isn't available for the broad public as it's undergoing major refactoring to be compatible with all camera models :-\ ... I'm happy enough to have a usable dev version. But as the information about this upcoming feature is available on the ML forum I feel free to spread the good news, even if it currently takes longer than expected...

... but speaking of upcoming goodies from ML: It seems the module will be accompanied with a postprocessing path to remove fixed pattern noise which is a problem for 5d3 when using the max. dynamic range you can squeeze out of the Canon in-camera image path (6d is fine out of the box).

Don Haines

  • Canon EF 300mm f/2.8L IS II
  • *******
  • Posts: 3355
  • Posting cat pictures on the internet since 1986
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #18 on: June 03, 2014, 05:38:14 PM »
Guys, didn't you read DxOmark? Just downsample your images down to 8x12" and print it. Then you gain at least 2 stops of dynamic range. And if 14-stops isn't enough then just downsample some more.  :P
OK.
I downsampled this picture of my cat to 16x16 pixels, how much dynamic range should I have?
The best camera is the one in your hands

ajfotofilmagem

  • 1D Mark IV
  • ******
  • Posts: 981
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #19 on: June 03, 2014, 05:53:07 PM »
Guys, didn't you read DxOmark? Just downsample your images down to 8x12" and print it. Then you gain at least 2 stops of dynamic range. And if 14-stops isn't enough then just downsample some more.  :P
Holy God! We found the solution to all problems of DR. ::) From now on, I'll shoot everything with 36 megapixel, and then downsample to 2 megapixel. Then print with 18 stops DR. :P
« Last Edit: June 03, 2014, 05:55:04 PM by ajfotofilmagem »

MichaelHodges

  • Canon 6D
  • *****
  • Posts: 368
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #20 on: June 03, 2014, 06:10:17 PM »
And if Canon can do that, can it finally deliver a FF sensor that is also class leading? By class leading, I'm referring to noise control and DR

Canon already has a class-leading sensor, it's in the 6d: nearly zero (esp. vertical) banding (better than 1dx @iso100) means almost the full theoretical dr is actually usable. It has good dr @base iso (boosted +1/3ev by ML and =15ev with dual_iso) and superior dynamic range at high iso: http://sensorgen.info/CanonEOS_6D.html


I can vouch for the 6D sensor.  Fantastic looking images.

jrista

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4467
  • EOL
    • View Profile
    • Nature Photography
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #21 on: June 03, 2014, 06:16:27 PM »
almost the full theoretical dr is actually usable
This is not true. The 6D does NOT have the "full theoretical DR", not even close.

Talking of correct quoting :-) ... but thanks for the explanation anyway, always great to read your in-depth posts. You did mis-understand me in this case though: the "usable dynamic range" was related to pattern noise, not to the theoretical limits of a Canon sensor if you glue it inside a Nikon camera body.

I think you may be misunderstanding what dynamic range is. Dynamic range is defining the "usable" range of signal. I think it becomes a bit redundant to say "usable dynamic range". The entire dynamic range is usable, because it's the range from the noise floor to the maximum signal...dynamic range implicitly excludes the "unusable" part of the signal. In an "ideal" system, one without noise, dynamic range would (theoretically) be infinite (The formula: 20 log(FWC/NoiseRMS) fails if NoiseRMS == 0...that's effectively a vertical and instantaneous asymptotic explosion to infinity. :P). That would mean that you have an infinite range of USABLE signal.

So, while I understand what your trying to say...I think it's a misnomer. Dynamic range is itself defining what's usable...so saying that you can only use part of what's usable... Redundant and inaccurate. And, incorrect. ;P

however the ADC bit depth limits it to 14, and downstream electronics (namely the ADC) introduce so much noise that it flattens the curve, rather than leaving it in it's linear state.

If you're interested in how the Magic Lantern people to improve the adc chain, see their forum (I posted the link above)


I've read some of their stuff. I haven't read it all. From what I have read, a lot of what they do benefits from Canon's downstream secondary amplifier.

It doesn't matter if the 6D ISO 100 noise doesn't have banding...the problem is that it still has a ton of noise.

Noise doesn't worry me that much as the 5d3/1dx/6d manage to make it look like film nose, so that elevates it way above my 60d (and as far as I see 5d2) concerning "destructive noise". My eye is not a technical instrument to measure snr.

I totally agree that the NATURE of noise can make it acceptable or not. Horizontal and vertical bands are horrid and completely unnnatural. The 6D is indeed a massive improvement there. The noise is quite aesthetically pleasing and random. Canon has certainly made strides here.

canon rumors FORUM

Re: Can Canon deliver a FF sensor that is class leading?
« Reply #21 on: June 03, 2014, 06:16:27 PM »

Radiating

  • Canon 6D
  • *****
  • Posts: 336
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #22 on: June 03, 2014, 06:58:44 PM »
Whilst we're going to have to wait a month to find out what the distortion is like on the 16-35/f4L, the MTF graphs and samples tend to suggest it is going to be a class leading lens. And how long have we had to wait for it?!

And if Canon can do that, can it finally deliver a FF sensor that is also class leading? By class leading, I'm referring to noise control and DR that is close to linear right through the ISO range and that matches or exceeds that of Sony sensors.

What's important about it being linear? At the moment Canon sensors have a DR function that is flat in the low ISO range as opposed to Sony's which is linear: below ISO 200 (or is it 400?) there is very little gain in DR from Canon sensors whereas Sony's deliver substantial gains the lower the ISO goes. I suspect that prior to Sony's sensors coming out, people thought that the "flat area" at the start of Canon's graphs looked good because there was no degradation. What people didn't realise is that there shouldn't be any flat spots, rather the DR (colour, noise, etc) response of the sensor should be close to having a linear relationship with ISO.


You do realize that the 5D3 1DX and 70D sensors are better than anything Sonikon makes in several categories right? Just because those aren't your favorite categories doesn't make them any worse.

David Hull

  • EOS M2
  • ****
  • Posts: 222
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #23 on: June 03, 2014, 07:39:40 PM »
Whilst we're going to have to wait a month to find out what the distortion is like on the 16-35/f4L, the MTF graphs and samples tend to suggest it is going to be a class leading lens. And how long have we had to wait for it?!

And if Canon can do that, can it finally deliver a FF sensor that is also class leading? By class leading, I'm referring to noise control and DR that is close to linear right through the ISO range and that matches or exceeds that of Sony sensors.

What's important about it being linear? At the moment Canon sensors have a DR function that is flat in the low ISO range as opposed to Sony's which is linear: below ISO 200 (or is it 400?) there is very little gain in DR from Canon sensors whereas Sony's deliver substantial gains the lower the ISO goes. I suspect that prior to Sony's sensors coming out, people thought that the "flat area" at the start of Canon's graphs looked good because there was no degradation. What people didn't realise is that there shouldn't be any flat spots, rather the DR (colour, noise, etc) response of the sensor should be close to having a linear relationship with ISO.

Who says?  FWIW I don't believe that the flat spot has anything to do with the sensor at all -- it is due to the noise being dominated by that of the ADC which follows the sensor in the Canon designs.  If you really dig into things you will find that the Canon sensors are actually the class leaders in terms of dynamic range, they just foul it up in terms of their system architecture.  The answer to the question "can Canon deliver a class leading sensor" is yes... they already do.  Take a look at the data from Sensorgen for the 5DIII for example:  Min read noise 2.4e-, FWC=67531 for a DR of 14.7 stops.

If they were to put an ADC on-chip similar to what Sony does, they would have equivalent performance maybe even better.

What Canon needs is a "class leading" system architecture and ADC.  This is a nit, I know but it is important because Canon DOES have class leading sensors and has had them for quite some time.
« Last Edit: June 03, 2014, 07:43:12 PM by David Hull »

David Hull

  • EOS M2
  • ****
  • Posts: 222
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #24 on: June 03, 2014, 07:49:01 PM »
And if Canon can do that, can it finally deliver a FF sensor that is also class leading? By class leading, I'm referring to noise control and DR

Canon already has a class-leading sensor, it's in the 6d: nearly zero (esp. vertical) banding (better than 1dx @iso100) means almost the full theoretical dr is actually usable. It has good dr @base iso (boosted +1/3ev by ML and =15ev with dual_iso) and superior dynamic range at high iso: http://sensorgen.info/CanonEOS_6D.html

This is not true. The 6D does NOT have the "full theoretical DR", not even close. The 6D has the same problem as every other Canon camera: A flattened DR curve at low ISO. Canon's DR tops out at around 12 stops of DR. Their sensors are more than capable of more than that, according to Roger Clark the sensor itself is actually probably capable of a little over 15 stops of DR natively (in analog space, before digitization), however the ADC bit depth limits it to 14, and downstream electronics (namely the ADC) introduce so much noise that it flattens the curve, rather than leaving it in it's linear state. It doesn't matter if the 6D ISO 100 noise doesn't have banding...the problem is that it still has a ton of noise. It has 26.8e- worth of read noise, which while less than the 1DX's 38.2e-, it also has a lower FWC, so it's dynamic range is roughly the same.

The problem isn't the sensor. Canon's sensors are very good. Canon's problem is their high frequency off-die ADC units housed in the DIGIC chips. They are just plain noisy.

Roger Clark has evaluated a lot of Canon sensors. His work finds the lowest noise level in the sensor itself, which would be intrinsic sensor noise, devoid of actual read noise. That's dark current noise in the sensor, along with whatever noise the per-pixel amplifiers might introduce. In Canon sensors, that noise level is around 2e-. In the case of the 6D, the sensor's analog dynamic range would be 20log(76606/2) dB, or 91.65 dB, which in terms of stops is 15.27. In the case of the 1D X, which also has ~2e- intrinsic sensor noise, the maximum possible dynamic range would be 93.1 dB or 15.52 stops. These levels aren't realizable due to the amount of read noise at ISO 100. If Canon can get their read noise under control, and get their ISO 100 noise levels down to 3-4e-, their dynamic range would be ~14-14.5 stops. Throw in a little bit of quantization error and PRNU and a 14-bit ADC, and Canon's DR jumps up to the level of Sony Exmors. Throw in a 16-bit ADC, and Canon should be able to achieve 14.5 stops of DR pretty easily. If they can lower their read noise levels even more, they could achieve well more than 15 stops of DR.

Their SENSORS are capable. The rest of their electronics are not. Canon's biggest problem is their approach of offloading the ADC into the DIGIC chip, and running them at very high frequency. Plus, their use of a downstream secondary amplifier doesn't help, but that only kicks in at higher ISO as far as I am aware.

Yep.... someone actually "gets it".  It just blows me away that so many don't really understand what that curve really says with regard to the system implementation. I didn't get to your post before I stuck mine on further down.  I bet Sony looses a bit of performance by running all of that digital crapola in the same die with the sensor stuff -- I suspect their sensor is probably as good as Canon's.
« Last Edit: June 03, 2014, 07:52:02 PM by David Hull »

tomri

  • SX60 HS
  • **
  • Posts: 2
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #25 on: June 04, 2014, 03:26:26 AM »
Isn't the problem that in order to integrate the ADC on the Sensor, Canon needs to switch to a more modern silicon process (using finer structures, from hundreds of nanometers today to tens of nanometers)?

Doing so while producing the sensor in own factories would mean to build an entirely new fab, which cost in the billions of US-$. So from a commercial perspective almost *any* other type of R&D effort is cheaper than coming up with that new sensor tech.

What makes me wonder is the dual pixel AF of the 70D. It is essentially a 40mpix sensor. In order to collect similar amounts of light as 20mpix sensors, the gaps between the photo diodes must be very small. Wouldn't this require a small structure silicon process already?

Has there been any insight in whether Canon might already be using smaller structure silicon?

« Last Edit: June 04, 2014, 03:28:18 AM by tomri »

sanj

  • 1D X
  • *******
  • Posts: 1557
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #26 on: June 04, 2014, 03:30:01 AM »
I am sure it can. And I hope it does soon.

Marsu42

  • Canon EF 400mm f/2.8L IS II
  • *******
  • Posts: 4816
  • ML-66d / 100L / 70-300L / 17-40L / 600rts
    • View Profile
    • 6D positive spec list
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #27 on: June 04, 2014, 03:37:14 AM »
So, while I understand what your trying to say...I think it's a misnomer. Dynamic range is itself defining what's usable...so saying that you can only use part of what's usable... Redundant and inaccurate. And, incorrect. ;P

Ok, thanks for explaining, I'm really not much of a tech geek even after all this time on CR :-p ... so for my education and to avoid further confusion: With which word(s) would you label the dynamic range starting from the shadows when no banding is visible anymore? Because that's what I called "usable" as fpn is what makes your shot "unusable".

canon rumors FORUM

Re: Can Canon deliver a FF sensor that is class leading?
« Reply #27 on: June 04, 2014, 03:37:14 AM »

TLN

  • Power Shot G7X
  • **
  • Posts: 25
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #28 on: June 04, 2014, 03:47:18 AM »
There are actually three questions

1.  Can Canon deliver this sensor
2.  Should Canon deliver this sensor
3.  Will Canon deliver this sensor

Only the first question has anything to do with the technology.  Unfortunately, the second question is the limiting factor.  Canon may be technologically capable of delivering this sensor, but it the marketing/production/sales numbers don't add up, all the technology won't matter.

Assuming that the first two questions are yes, the third question may dictate that they will do this in 2-3/3-5 years from now. 

I think gone are the days when cameras were limited by technical capability.  It is all about the business case.  And that's a good thing.

That's it.

I'm sure canon can do it. And in my opinion it will do it quite soon.

You can notice, that canon and nikon lineup doesn't have equal cameras. Yep, they're quite close, but I see it goes like:
D3100-600D-D5200-60D-D7100-7D for crop cameras one by one. Same for Fullframe cameras, but it's easier to show with bigger range of cropped.

Second thing you can notice, that canon and nikon got some cycles. Say 2-4 years. Reason? Same - marketing.
When 5d2 came out it was a deal-breaker. Great performance, 21mp. Hard to compare with D700. Yep, megapixels doesn't matter that much, as people believe, but we all know that 5d2 sell wery well.
4 years later Nikon makes D800 and canon stays with 5d3(which is cool, but difference between 5d-5d2 and 5d2-5d3 is not that big).
Canon is upgrading their lenses, from tele(easiest to design) to wide to II versions(70-200 II, 24-70II) or adds IS (24-28-35 IS, 16-35 IS) and they all have great sharpness and performance.  I think canon will announce some high-MP camera if 1-2 years, as well as upgraded primes.
This change of leadership easily falls to 4 year cycle. Later we'll see some nikon on top.

Companies makes a bigger profits, if the leader changes with time, this makes users go to one side or another.
Fortunately now is canon turn. Unfortunately, previous announces makes me think, that price will be quite high.

eml58

  • 1D X
  • *******
  • Posts: 1504
  • 1Dx
    • View Profile
Re: Can Canon deliver a FF sensor that is class leading?
« Reply #29 on: June 04, 2014, 03:57:39 AM »
Yep
The only thing necessary for the triumph of evil is for good men to do nothing

canon rumors FORUM

Re: Can Canon deliver a FF sensor that is class leading?
« Reply #29 on: June 04, 2014, 03:57:39 AM »