Is it really the off-sensor ADC?

Started Jul 18, 2013 | Questions
TTMartin
TTMartin Veteran Member • Posts: 7,304
Is it really the off-sensor ADC?

In the thread titled dynamic range confusion David Hull wrote:

texinwien wrote:

Sensor Dynamic Range vs. Camera Dynamic Range

This windowing of DR is occurring because the camera doesn't deliver the full DR that the sensor is capable of at base ISO; rather it is limited by the rest of the electronics -- the ISO amplifier and ADC. The sensor has a DR of about 14 stops, while the amp/ADC have a bit less than 12 stops DR. There are about two stops of extra dynamic range to be had in DSLR's that is currently being thrown away by limitations of components other than the sensor.

This appears to be exactly the case, a point I have been making for quite a while. From a sensor perspective what Canon is putting in there appears to be every bit as good as what Sony is making:

Nikon D600: QE=53%, Read Noise = 2.8e-, FWC=76231, DR~14.7 stops

Canon 6D: QE=50%, Read Noise 1.6 e-, FWC=76606, DR~15.5

Source: Bob's Sensorgen site (which is ultimately DxO, I guess)

. . .

People need to stop slamming the Canon sensor technology because it is actually pretty darn good. This also says the front end noise figure of ISO amp is pretty good as well. The demons are hiding in the rear end of it as Mr. Martinec points out. The sensor is not where bodies are buried in these things.

Thanks for putting this up, I had not seen it before – good stuff.

With the above showing that the 6D sensor is actually capable of producing 15.5 EV of DR it's time to again ask if it is really the off-sensor ADC?

It has been repeated ad nauseum that the off-sensor ADC is dumping tons of read noise in Canon RAW files.

If it were ADC inefficiencies, it should manifest itself as clipping not as read noise. Or in the off chance it was introducing read noise, that read noise should not decrease as ISO increases. Nor should the amount of read noise vary as much as it does from one Canon camera to another.

If some other component were introducing read noise prior to the ADC, then the read noise should not decrease as ISO increases.

None of the scenarios involving the off sensor ADC should manifest itself as increasing read noise, as you lower the ISO.

But, then where is that read noise coming from?

I've been ridiculed for my suggestion of where that tons of read noise comes from.

Is there anyone now willing to have a mature discussion on the subject?

 TTMartin's gear list:TTMartin's gear list
Canon PowerShot S100 (2000) Canon PowerShot SD1000 Canon PowerShot G1 X Canon EOS 40D Canon EOS 400D +26 more
ANSWER:
This question has not been answered yet.
Canon EOS 6D
If you believe there are incorrect tags, please send us this post using our feedback form.
Press Correspondent
Press Correspondent Veteran Member • Posts: 3,362
Re: Is it really the off-sensor ADC?

TTMartin wrote:

Is there anyone now willing to have a mature discussion on the subject?

No.

 Press Correspondent's gear list:Press Correspondent's gear list
SeaLife DC2000 Panasonic LX100 II Canon EOS 5D Mark III Canon EF 50mm F1.4 USM Canon EF 70-200mm F4L USM +12 more
David Hull
David Hull Veteran Member • Posts: 6,820
Re: Is it really the off-sensor ADC?
3

TTMartin wrote:

In the thread titled dynamic range confusion David Hull wrote:

texinwien wrote:

Sensor Dynamic Range vs. Camera Dynamic Range

This windowing of DR is occurring because the camera doesn't deliver the full DR that the sensor is capable of at base ISO; rather it is limited by the rest of the electronics -- the ISO amplifier and ADC. The sensor has a DR of about 14 stops, while the amp/ADC have a bit less than 12 stops DR. There are about two stops of extra dynamic range to be had in DSLR's that is currently being thrown away by limitations of components other than the sensor.

This appears to be exactly the case, a point I have been making for quite a while. From a sensor perspective what Canon is putting in there appears to be every bit as good as what Sony is making:

Nikon D600: QE=53%, Read Noise = 2.8e-, FWC=76231, DR~14.7 stops

Canon 6D: QE=50%, Read Noise 1.6 e-, FWC=76606, DR~15.5

Source: Bob's Sensorgen site (which is ultimately DxO, I guess)

. . .

People need to stop slamming the Canon sensor technology because it is actually pretty darn good. This also says the front end noise figure of ISO amp is pretty good as well. The demons are hiding in the rear end of it as Mr. Martinec points out. The sensor is not where bodies are buried in these things.

Thanks for putting this up, I had not seen it before – good stuff.

With the above showing that the 6D sensor is actually capable of producing 15.5 EV of DR it's time to again ask if it is really the off-sensor ADC?

It has been repeated ad nauseum that the off-sensor ADC is dumping tons of read noise in Canon RAW files.

If it were ADC inefficiencies, it should manifest itself as clipping not as read noise. Or in the off chance it was introducing read noise, that read noise should not decrease as ISO increases. Nor should the amount of read noise vary as much as it does from one Canon camera to another.

Why would this be the case?

If some other component were introducing read noise prior to the ADC, then the read noise should not decrease as ISO increases.

A camera is just like a satellite receiver. Let's say you have a LNA (low Noise Amplifier) in the front end that has a noise figure of 1 dB (very low noise) and its gain is 40 dB. This amplifier is followed by a subsequent stage that has a noise figure of 25 dB (lots of noise). In this case the effective noise figure of the cascaded stages will be 1.107 dB.

If you now reduce the gain of that first stage to 5 dB (think of that first stage as the ISO amplifier and you are turning its gain down) the effective noise figure of the chain is now 20 dB.

This is known as Friis Formula or more generally as the "cascade Noise Figure Formula" it is fundamental receiver lineups -- this is radio engineering 101.

http://en.wikipedia.org/wiki/Friis_formulas_for_noise

Looking at the 5DIII, you have 22 MPix sampled at 6 FPS, so the converter needs to run at 132 MSPS. In this camera there are 8 channels so each of the converters will be running at 16.5 MSPS. They are probably using 20 MSPS or 40 MSPS converters. You can look these up yourself if you want (ADI-9257 might be a good one). You can look at the data sheet for these things and you will see that the SINAD rating is in the range of 73-75 dB (depending on the speed and quality of the part). This equates to an ENOB [ENOB=(SINAD-1.76)/6.02] of 11.8 to 12.2 dB.

Basically what you are doing is amplifying the front end noise to the point where it dominates over the back end noise when you increase the gain. Let’s say that you have a very weak signal but your front end is very quiet (very low noise) you pass that directly (unamplified) to a second stage that has 10 times the noise. That second stage noise will be dominant and will swamp out your signal and your noise. Now let’s say you increase the gain of that very quiet amplifier to 1000. Now the noise (and the signal) out of that first stage is 100 times the noise of the second stage. The SNR of the first stage dominates and the noise of the second one is rendered irrelevant.

Essentially you have a noise floor of 74 dB in the back end, it will never get any better than that.

None of the scenarios involving the off sensor ADC should manifest itself as increasing read noise, as you lower the ISO.

But, then where is that read noise coming from?

I've been ridiculed for my suggestion of where that tons of read noise comes from.

Is there anyone now willing to have a mature discussion on the subject?

-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon EF 100mm F2.8L Macro IS USM Canon RF 35mm F1.8 IS STM Macro Canon RF 24-105mm F4L IS USM Canon RF 15-35mm F2.8L IS USM +1 more
TTMartin
OP TTMartin Veteran Member • Posts: 7,304
Re: Is it really the off-sensor ADC?

David Hull wrote:

If some other component were introducing read noise prior to the ADC, then the read noise should not decrease as ISO increases.

A camera is just like a satellite receiver. Let's say you have a LNA (low Noise Amplifier) in the front end that has a noise figure of 1 dB (very low noise) and its gain is 40 dB. This amplifier is followed by a subsequent stage that has a noise figure of 25 dB (lots of noise). In this case the effective noise figure of the cascaded stages will be 1.107 dB.

If you now reduce the gain of that first stage to 5 dB (think of that first stage as the ISO amplifier and you are turning its gain down) the effective noise figure of the chain is now 20 dB.

This is known as Friis Formula or more generally as the "cascade Noise Figure Formula" it is fundamental receiver lineups -- this is radio engineering 101.

http://en.wikipedia.org/wiki/Friis_formulas_for_noise

Looking at the 5DIII, you have 22 MPix sampled at 6 FPS, so the converter needs to run at 132 MSPS. In this camera there are 8 channels so each of the converters will be running at 16.5 MSPS. They are probably using 20 MSPS or 40 MSPS converters. You can look these up yourself if you want (ADI-9257 might be a good one). You can look at the data sheet for these things and you will see that the SINAD rating is in the range of 73-75 dB (depending on the speed and quality of the part). This equates to an ENOB [ENOB=(SINAD-1.76)/6.02] of 11.8 to 12.2 dB.

Basically what you are doing is amplifying the front end noise to the point where it dominates over the back end noise when you increase the gain. Let’s say that you have a very weak signal but your front end is very quiet (very low noise) you pass that directly (unamplified) to a second stage that has 10 times the noise. That second stage noise will be dominant and will swamp out your signal and your noise. Now let’s say you increase the gain of that very quiet amplifier to 1000. Now the noise (and the signal) out of that first stage is 100 times the noise of the second stage. The SNR of the first stage dominates and the noise of the second one is rendered irrelevant.

Essentially you have a noise floor of 74 dB in the back end, it will never get any better than that.

Thanks Dave,

I'll have to look into most of this in the morning.

My initial thought reading it, is that would seem to hold true if the signal being received remained constant, then increasing the amplifier would do exactly what you are saying.

But, with the camera, the ISO is really only increased to compensate for less signal (less light), so the front end non-noise signal keeps going down, relative to the amplification, so the amount of non-noise signal reaching the later stage noise source remains relatively constant.

Again, I'll look into it all, more in the morning.

I appreciate your taking the time to talk this through with me.

 TTMartin's gear list:TTMartin's gear list
Canon PowerShot S100 (2000) Canon PowerShot SD1000 Canon PowerShot G1 X Canon EOS 40D Canon EOS 400D +26 more
David Hull
David Hull Veteran Member • Posts: 6,820
Re: Is it really the off-sensor ADC?
1

TTMartin wrote:

David Hull wrote:

If some other component were introducing read noise prior to the ADC, then the read noise should not decrease as ISO increases.

A camera is just like a satellite receiver. Let's say you have a LNA (low Noise Amplifier) in the front end that has a noise figure of 1 dB (very low noise) and its gain is 40 dB. This amplifier is followed by a subsequent stage that has a noise figure of 25 dB (lots of noise). In this case the effective noise figure of the cascaded stages will be 1.107 dB.

If you now reduce the gain of that first stage to 5 dB (think of that first stage as the ISO amplifier and you are turning its gain down) the effective noise figure of the chain is now 20 dB.

This is known as Friis Formula or more generally as the "cascade Noise Figure Formula" it is fundamental receiver lineups -- this is radio engineering 101.

http://en.wikipedia.org/wiki/Friis_formulas_for_noise

Looking at the 5DIII, you have 22 MPix sampled at 6 FPS, so the converter needs to run at 132 MSPS. In this camera there are 8 channels so each of the converters will be running at 16.5 MSPS. They are probably using 20 MSPS or 40 MSPS converters. You can look these up yourself if you want (ADI-9257 might be a good one). You can look at the data sheet for these things and you will see that the SINAD rating is in the range of 73-75 dB (depending on the speed and quality of the part). This equates to an ENOB [ENOB=(SINAD-1.76)/6.02] of 11.8 to 12.2 dB.

Basically what you are doing is amplifying the front end noise to the point where it dominates over the back end noise when you increase the gain. Let’s say that you have a very weak signal but your front end is very quiet (very low noise) you pass that directly (unamplified) to a second stage that has 10 times the noise. That second stage noise will be dominant and will swamp out your signal and your noise. Now let’s say you increase the gain of that very quiet amplifier to 1000. Now the noise (and the signal) out of that first stage is 100 times the noise of the second stage. The SNR of the first stage dominates and the noise of the second one is rendered irrelevant.

Essentially you have a noise floor of 74 dB in the back end, it will never get any better than that.

Thanks Dave,

I'll have to look into most of this in the morning.

My initial thought reading it, is that would seem to hold true if the signal being received remained constant, then increasing the amplifier would do exactly what you are saying.

But, with the camera, the ISO is really only increased to compensate for less signal (less light), so the front end non-noise signal keeps going down, relative to the amplification, so the amount of non-noise signal reaching the later stage noise source remains relatively constant.

Again, I'll look into it all, more in the morning.

I appreciate your taking the time to talk this through with me.

Keep in mind that the DR is defined as the difference between the saturation of the converter (all 14 bits are flipped) and the noise floor. Once the front end noise gets amplified to the point where the bottom three or four bits are wiggling, then the input noise is starting to dominate. This is the point where the DxO curve starts to "knee over" and leave the flat part at low ISO and start that linear downhill ramp. The “rear end” noise is being overshadowed now by the “amped-up” front end noise.

I am not sure what you meant by your clipping comment but let’s say that all of this is well under way by the time the time the 7bit starts wiggling. If those 7 bits are wiggling due to noise then the remaining 7 are giving you 7 stops of DR. You crank up the front end gain a couple stops more, now 9 bits will be wiggling (from noise) then 14-9=5 bits (or 5 stops) are left for signal and your DR is 5 stops. We don’t care about the signal (it can be anywhere in there) but the DR is defined as the difference between full scale SAT of the ADC and the “amped-up” noise floor.

Now if you were to get a different kind of converter (say an SAR converter with 14 effective bits) and stick it right at the sensor diode (integrated onto the sensor die itself), and… that converter was quiet enough so as not to add much noise, then you could do your “amping-up” in the digital domain (with bit shifting or something like that) after the ADC. In that case that linear (so-called ISO-less) region would start right at ISO=100, like the Sony’s do.

That is the real difference, IMO, Nikon/Sony have an architecture that is optimized for maximum DR across the entire range of gain/ISO settings Canon does not.

-- hide signature --
 David Hull's gear list:David Hull's gear list
Canon EOS R Canon EF 100mm F2.8L Macro IS USM Canon RF 35mm F1.8 IS STM Macro Canon RF 24-105mm F4L IS USM Canon RF 15-35mm F2.8L IS USM +1 more
Keyboard shortcuts:
FForum MMy threads