A simple way to put Canon's sensors on top.

[Great Bustard]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

 

[Press Correspondent]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

At low ISO the DR is limited mostly by the sensor read noise. At high ISO the DR is limited by the well capacity and therefore the photon noise. In other words, at high ISO the DR limitation is in the light itself even before it hits the sensor. Therefore increasing the number of bits in the electronics would not increase the DR at high ISO.

 

[Wayne Larmon]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

Wouldn't increasing the bit depth to 20 bits drive the price up to NASA project level? $50,000+ medium format cameras only claim 16 bits (don't know how much they actually deliver. Maybe you could achieve 20 bits by starting with a digital medium format camera and adding a cooling unit the size of a refrigerator?

Most of the technical people that hang out on the science and technology forum consider Canon's claim of 14 bits to be pure fiction.

An ISOless interface would be cool but I think that Canon needs to address the already well known issues with sensor noise first. If you were to post this on the Nikon or Sony forums, you'd might get a more reasonable response. They have cameras that are close to being capable of ISOless operation. (Well, I do also. I have a Canon G9 and John Sheehy has said that the G9 can run ISOless. But I think that Canon squelched that with later models.)

Wayne

 

[brightcolours]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

Huh? The low noise at "base ISO" for Sony sensors is due to low read noise. The Canon sensor will still have a higher read noise at ISO 3200 too. Just that besides the read noise the Canon sensors are better, does not mean that your solution will make the read noise go away, does it?

Also, why would increasing the RAW bit depth (nothing to do with what the sensor collects) drown out the read noise and the noise from at ISO 3200 needed high signal amplification and resulting noise (because of a weak signal)?

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

 

[rrccad]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

no canon simply has to get or decide to use better conversions from analog signals to digital.

they also have to come out with a CR3 format as well.

people say "sony sensors" or "nikon sensors" but it's the entire systematic aspect - including how one actually measures against a sony, nikon and canon RAW file.

 

[Horshack]

http://image-sensors-world.blogspot.com.es/2013/08/canon-files-for-dual-range-column.html

 

[Great Bustard]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

At low ISO the DR is limited mostly by the sensor read noise. At high ISO the DR is limited by the well capacity and therefore the photon noise. In other words, at high ISO the DR limitation is in the light itself even before it hits the sensor. Therefore increasing the number of bits in the electronics would not increase the DR at high ISO.

Increasing the number of bits will increase the DR since the read noise is lower (much lower) at higher ISOs on Canon sensors (up to ISO 3200, or thereabouts, after which the sensor has flat read noise). That's the point of my OP.

 

[Great Bustard]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

Huh? The low noise at "base ISO" for Sony sensors is due to low read noise.

Yes.

The Canon sensor will still have a higher read noise at ISO 3200 too.

They don't -- it's actually a bit lower. Not a lot, but a little.

Just that besides the read noise the Canon sensors are better, does not mean that your solution will make the read noise go away, does it?

If the bit depth were raised so that you could always shoot at ISO 3200, then the read noise would meet or beat current Sony sensors.

Also, why would increasing the RAW bit depth (nothing to do with what the sensor collects) drown out the read noise and the noise from at ISO 3200 needed high signal amplification and resulting noise (because of a weak signal)?

It's due to the noisy ADC in Canon sensors. That is, Canon makes very good pixels, but they have to pass through a lousy ADC. At higher amplifications, the ADC becomes more efficient.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

 

[Great Bustard]

http://image-sensors-world.blogspot.com.es/2013/08/canon-files-for-dual-range-column.html

This idea was suggested long ago by Emil Martinec in these forums. Would be awesome to see it become a reality!

 

[brightcolours]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

Huh? The low noise at "base ISO" for Sony sensors is due to low read noise.

Yes.

The Canon sensor will still have a higher read noise at ISO 3200 too.

They don't -- it's actually a bit lower. Not a lot, but a little.

Just that besides the read noise the Canon sensors are better, does not mean that your solution will make the read noise go away, does it?

If the bit depth were raised so that you could always shoot at ISO 3200, then the read noise would meet or beat current Sony sensors.

Also, why would increasing the RAW bit depth (nothing to do with what the sensor collects) drown out the read noise and the noise from at ISO 3200 needed high signal amplification and resulting noise (because of a weak signal)?

It's due to the noisy ADC in Canon sensors. That is, Canon makes very good pixels, but they have to pass through a lousy ADC. At higher amplifications, the ADC becomes more efficient.

So that takes "care" of readnoise levels. But, when you collect less light (shorter exposure time due to ISO 3200) you will end up with more noise, which limit DR again. Do you mean that with ISO 3200 and the higher bit depth, you "over expose" compared to what the situation is now (which then takes care of collecting enough light), and put the mid level "higher"?

I bet it would mean a redesign of the Digic processors.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

 

[Great Bustard]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

Huh? The low noise at "base ISO" for Sony sensors is due to low read noise.

Yes.

The Canon sensor will still have a higher read noise at ISO 3200 too.

They don't -- it's actually a bit lower. Not a lot, but a little.

Just that besides the read noise the Canon sensors are better, does not mean that your solution will make the read noise go away, does it?

If the bit depth were raised so that you could always shoot at ISO 3200, then the read noise would meet or beat current Sony sensors.

Also, why would increasing the RAW bit depth (nothing to do with what the sensor collects) drown out the read noise and the noise from at ISO 3200 needed high signal amplification and resulting noise (because of a weak signal)?

It's due to the noisy ADC in Canon sensors. That is, Canon makes very good pixels, but they have to pass through a lousy ADC. At higher amplifications, the ADC becomes more efficient.

So that takes "care" of readnoise levels.

Yes.

But, when you collect less light (shorter exposure time due to ISO 3200)...

But you don't with the greater bit depth -- that's my point. If the image file had 19 bits, then it could record the same amount of light at ISO 3200 as it could record at ISO 100 for the same shutter speed.

...you will end up with more noise, which limit DR again. Do you mean that with ISO 3200 and the higher bit depth, you "over expose" compared to what the situation is now (which then takes care of collecting enough light), and put the mid level "higher"?

The point is that the ISO setting has nothing, whatsoever, to do with the exposure, except inasmuch as the ISO setting indirectly affects those settings that do have something to do with exposure (f-ratio, shutter speed, flash power), depending on the shooting mode you're using.

That is, f/2.8 1/200 ISO 100 and f/2.8 1/200 ISO 3200 result in the *exact same exposure* for a given scene luminance, but the signal from the ISO 3200 photo has been amplified by a factor of 32 (five stops). So, if you have five more bits in the image file, you won't clip any more at ISO 3200 than you would at ISO 100, for a given exposure.

I bet it would mean a redesign of the Digic processors.

For a fact. But, as I said, if the firmware allowed an *optional* ISOless UI, which set the amplification to a constant ISO 3200, then not only would Canon match the DR of Sony sensors, but also be the first to enter into the realm of ISOless shooting.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

 

[Great Bustard]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

Wouldn't increasing the bit depth to 20 bits drive the price up to NASA project level?

Dunno. Is 20 bit architecture that much more expensive than 14 bit? I suspect not. The main issue, methinks, is if 20 bit architecture would slow down the frame rate (of course, it would require greater storage and processing times, but...).

$50,000+ medium format cameras only claim 16 bits (don't know how much they actually deliver. Maybe you could achieve 20 bits by starting with a digital medium format camera and adding a cooling unit the size of a refrigerator?

Well, if there are technical issues that make 20 bits infeasible, then you certainly have a point.

Most of the technical people that hang out on the science and technology forum consider Canon's claim of 14 bits to be pure fiction.

This is simply because the DR is less than 14 bits, so the two lowest bits are merely noise. Of course, 20 bits would be very inefficient, but it would allow for significantly less read noise with the same photon noise, and thus less noise overall.

An ISOless interface would be cool but I think that Canon needs to address the already well known issues with sensor noise first.

I'm merely suggesting one of many possible ways around the problem.

If you were to post this on the Nikon or Sony forums, you'd might get a more reasonable response.

;-)

They have cameras that are close to being capable of ISOless operation. (Well, I do also. I have a Canon G9 and John Sheehy has said that the G9 can run ISOless. But I think that Canon squelched that with later models.)

In at least one of the Sigma cameras (I forget which), it is full-on ISOless, with the ISO simply as a tag in the metafile. Unfortunately, the camera does not have an ISOless UI, even if just an option, which is a real pity.

 

[Press Correspondent]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

At low ISO the DR is limited mostly by the sensor read noise. At high ISO the DR is limited by the well capacity and therefore the photon noise. In other words, at high ISO the DR limitation is in the light itself even before it hits the sensor. Therefore increasing the number of bits in the electronics would not increase the DR at high ISO.

Increasing the number of bits will increase the DR since the read noise is lower (much lower) at higher ISOs on Canon sensors (up to ISO 3200, or thereabouts, after which the sensor has flat read noise). That's the point of my OP.

I see now what you meant, but did not say thus making your proposal unclear. You apparently assume that the read noise would remain at the same count number while the numbers are lower with more bits and therefore the read noise would reduce relatively to the signal. Did I get it right?

I believe this is a wrong assumption. If you simply increase the resolution of the ADC by 2 and leave everything else the same, the ADC would simply measure 1 photon as 2 and 1 electron as 2 and therefore the SNR would stay the same. But this is not what you want. You want to increase the number of bits while keeping the ADC count 1 to 1 with photons and electrons. This effectively means increasing the well capacity that would indeed increase the DR. Therefore you are saying that the well capacity is limited only by the ADC resolution. Is this a fact? I am not a sensor engineer, but I do specialize in solid state physics and I suspect that the sensor saturation level is an actual physical limit on the number of electrons the CMOS can produce that happens before ADC and has nothing to do with it. The ADC may indeed be tuned to have the highest bit just below the physical saturation level of the sensor, but it does not mean at all that adding an extra bit at the top of the ADC would allow the CMOS producing more electrons than it can.

If your assumption were correct, it would indeed be an easy fix. High resolution 24-bit codecs (ADCs and DACs) have been used in the audio industry for at least a decade and by now cost little, if any, more than 16-bit ones to produce.

 

[brightcolours]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

At low ISO the DR is limited mostly by the sensor read noise. At high ISO the DR is limited by the well capacity and therefore the photon noise. In other words, at high ISO the DR limitation is in the light itself even before it hits the sensor. Therefore increasing the number of bits in the electronics would not increase the DR at high ISO.

Increasing the number of bits will increase the DR since the read noise is lower (much lower) at higher ISOs on Canon sensors (up to ISO 3200, or thereabouts, after which the sensor has flat read noise). That's the point of my OP.

I see now what you meant, but did not say thus making your proposal unclear. You apparently assume that the read noise would remain at the same count number while the numbers are lower with more bits and therefore the read noise would reduce relatively to the signal. Did I get it right?

I believe this is a wrong assumption. If you simply increase the resolution of the ADC by 2 and leave everything else the same, the ADC would simply measure 1 photon as 2 and 1 electron as 2 and therefore the SNR would stay the same. But this is not what you want. You want to increase the number of bits while keeping the ADC count 1 to 1 with photons and electrons. This effectively means increasing the well capacity that would indeed increase the DR. Therefore you are saying that the well capacity is limited only by the ADC resolution. Is this a fact? I am not a sensor engineer, but I do specialize in solid state physics and I suspect that the sensor saturation level is an actual physical limit on the number of electrons the CMOS can produce that happens before ADC and has nothing to do with it. The ADC may indeed be tuned to have the highest bit just below the physical saturation level of the sensor, but it does not mean at all that adding an extra bit at the top of the ADC would allow the CMOS producing more electrons than it can.

If your assumption were correct, it would indeed be an easy fix. High resolution 24-bit codecs (ADCs and DACs) have been used in the audio industry for at least a decade and by now cost little, if any, more than 16-bit ones to produce.

I thinks he means:

At ISO 3200 settings, for "normal exposure" for the tonal curve used and the 14 bits format it is stored in, you get low read noise but higher than ISO 100 total noise due to few photon's collected. At ISO 3200 with the short exposure times, you are not going to use the well capacity (but you are going to blow out highlights due to the way it is stored).

So I think he suggests: with 20 bits you have a crazy amount of room to store a crazy amount of DR. You then easily "over expose" compared to above scenario, with exposure times which would be normal with ISO 100. You then will use the well capacity to full extent again, and just have to define where you place mid tones within the 20 bits. By shifting the mid tone "point" within the 20 bits, you have a "ISO-less" camera and you regulate the exposure time not by setting a different ISO amplification level, but by defining where within the 20 bits the DR range lays you want to use during RAW development.

It would need some clever thinking about the UI, to make it easy and intuitive to use, though.

 

[rhlpetrus]

Wouldn't increasing bit depth also increase file size (raw) by a large amount?

 

[Great Bustard]

Wouldn't increasing bit depth also increase file size (raw) by a large amount?

Yep. But storage is cheap, so I don't consider that an issue. What I do consider as potential issues is if it affects the frame rate and how it affects processing time.

 

[Great Bustard]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

At low ISO the DR is limited mostly by the sensor read noise. At high ISO the DR is limited by the well capacity and therefore the photon noise. In other words, at high ISO the DR limitation is in the light itself even before it hits the sensor. Therefore increasing the number of bits in the electronics would not increase the DR at high ISO.

Increasing the number of bits will increase the DR since the read noise is lower (much lower) at higher ISOs on Canon sensors (up to ISO 3200, or thereabouts, after which the sensor has flat read noise). That's the point of my OP.

I see now what you meant, but did not say thus making your proposal unclear.

My apologies.

You apparently assume that the read noise would remain at the same count number while the numbers are lower with more bits and therefore the read noise would reduce relatively to the signal. Did I get it right?

Yes, but it's not an assumption -- it's how things work. For example, observe the read noise for the 6D:

http://www.sensorgen.info/CanonEOS_6D.html

or any camera, for that matter. The only question is when the read noise levels out (sensors that are level from the start are called "ISOless", but all sensors become ISOless after some point).

I believe this is a wrong assumption. If you simply increase the resolution of the ADC by 2 and leave everything else the same, the ADC would simply measure 1 photon as 2 and 1 electron as 2 and therefore the SNR would stay the same. But this is not what you want. You want to increase the number of bits while keeping the ADC count 1 to 1 with photons and electrons. This effectively means increasing the well capacity that would indeed increase the DR. Therefore you are saying that the well capacity is limited only by the ADC resolution. Is this a fact?

The well capacity of the pixel is not a function of the ADC or the ISO setting.

I am not a sensor engineer, but I do specialize in solid state physics and I suspect that the sensor saturation level is an actual physical limit on the number of electrons the CMOS can produce that happens before ADC and has nothing to do with it.

Correct!

The ADC may indeed be tuned to have the highest bit just below the physical saturation level of the sensor, but it does not mean at all that adding an extra bit at the top of the ADC would allow the CMOS producing more electrons than it can.

My understanding is that while the pixels themselves are rather clean with respect to noise, the noise from the ADC, at least in Canon sensors, is, to some extent, a function of the signal, and thus lower signals result in less noise from the ADC, and thus noise levels much closer to the pixel noise.

If your assumption were correct, it would indeed be an easy fix. High resolution 24-bit codecs (ADCs and DACs) have been used in the audio industry for at least a decade and by now cost little, if any, more than 16-bit ones to produce.

Well, the read noise charts from sensorgen, as linked above, seem to support my idea, and it's good to know that 24 bit ADCs are not an issue in terms of cost.

 

[Great Bustard]

It would need some clever thinking about the [ISOless] UI, to make it easy and intuitive to use, though.

Here we go:

http://www.dpreview.com/forums/post/38224181

Q: How does the camera know what f-ratio and shutter speed to use if there is no ISO setting?

A: It doesn't. When using an ISOless camera, the photographer would choose the f-ratio for the desired DOF/sharpness and choose the shutter speed for the desired motion blur. The camera's meter would then tell the photographer how far "under" or "over" exposed the metering software thinks the photo is, and the photographer would compromise on f-ratio and/or shutter speed as they see fit.

Q: I don't get what you're saying. Could you give me an example?

A: Sure. Let's say I'm taking a pic of people at an indoor party, and f/2.8 would give me the DOF/sharpness I want, but people are moving a lot, so I choose 1/200 to avoid motion blur. The camera meter shows I'm two stops under what it considers "optimum". I trust the meter, and think, OK, I'll go with 1/50 and try to take pics only of people who are relatively still. Alternatively, I might stick to my guns, and just deal with the greater noise of photos that are two stops "underexposed", since the noise is preferable to the risk of motion blur.

http://www.dpreview.com/forums/post/42625308

Essentially you use it in manual mode, the meter has two scales, each measuring in a unit which is essentially milli-lumen-seconds (as I remember) so is sensor size independent. It is a matrix meter, the top line meters highlights, the bottom average illumination. The carets are user settable to their own quality threshold. All you do is adjust f-number and shutter as you want, if either bar is the wrong side of your quality threshold, choose which compromise to make. One can imagine a set of auto modes that would do that for you, a bit more sensibly than the current ones - all based on the idea that you adjust exposure to adjust image brightness - do. Of course, with an EVF like the E-M5, the highlight bar could probably be better done with blinkies.

 

[brightcolours]

It would need some clever thinking about the [ISOless] UI, to make it easy and intuitive to use, though.

Here we go:

http://www.dpreview.com/forums/post/38224181

Q: How does the camera know what f-ratio and shutter speed to use if there is no ISO setting?

A: It doesn't. When using an ISOless camera, the photographer would choose the f-ratio for the desired DOF/sharpness and choose the shutter speed for the desired motion blur. The camera's meter would then tell the photographer how far "under" or "over" exposed the metering software thinks the photo is, and the photographer would compromise on f-ratio and/or shutter speed as they see fit.

Q: I don't get what you're saying. Could you give me an example?

A: Sure. Let's say I'm taking a pic of people at an indoor party, and f/2.8 would give me the DOF/sharpness I want, but people are moving a lot, so I choose 1/200 to avoid motion blur. The camera meter shows I'm two stops under what it considers "optimum". I trust the meter, and think, OK, I'll go with 1/50 and try to take pics only of people who are relatively still. Alternatively, I might stick to my guns, and just deal with the greater noise of photos that are two stops "underexposed", since the noise is preferable to the risk of motion blur.

http://www.dpreview.com/forums/post/42625308

Essentially you use it in manual mode, the meter has two scales, each measuring in a unit which is essentially milli-lumen-seconds (as I remember) so is sensor size independent. It is a matrix meter, the top line meters highlights, the bottom average illumination. The carets are user settable to their own quality threshold. All you do is adjust f-number and shutter as you want, if either bar is the wrong side of your quality threshold, choose which compromise to make. One can imagine a set of auto modes that would do that for you, a bit more sensibly than the current ones - all based on the idea that you adjust exposure to adjust image brightness - do. Of course, with an EVF like the E-M5, the highlight bar could probably be better done with blinkies.

This is not a very good solution, is it?

In the situation now, we see if the photo is exposed "correctly" with the aperture and exposure time we select. If it is too dark, we up the ISO setting.

The ISOless situation should make it easier, not more complex to understand. More intuitive. We have to be able to do two things in the same situation: set a long exposure time (to maximise the number of photons collected) or to set a short exposure time (to for instance prevent motion blur). In both situations one should be able to just do a simple RAW conversion, where each photo looks correctly exposed to us. This means that the camera has to set a mid grey and/or black/white point per shot somewhere in that 20bit range. In RAW conversion we of course can shift those points, when wanted.

We also have to be able to meter for for instance high key and low key images, without having to think in "quality threshold" and mapping that to what we need.

So, somehow we have to be able to meter midlevels like we are used to, but without having to set ISO. The camera has to set the midlevel position in the RAW. It would basically come down to being able to set any aperture and exposure time combination and the camera has to assume we have metered it "correctly", and set a marker in the 20 bits RAW file. But we also have to be able to have that marker set in a way which results in an "under exposed" or "over exposed" default RAW conversion, and be able to manage flash too, in an understandable, intuitive manner.

 

[Press Correspondent]

At higher ISO settings, Canon sensors are as good as the best of them. So, if Canon were to offer an optional ISOless interface, increase the bit depth of the capture files to 20 bits, and have the cameras shoot permanently at ISO 3200 in the ISOless shooting mode, they'd match or beat Sony's sensors in terms of noise and DR.

By making the ISOless UI optional, it would still allow those who find the current noise and DR levels to be "good enough" and prefer setting the ISO themselves to continue as before without any bother.

At low ISO the DR is limited mostly by the sensor read noise. At high ISO the DR is limited by the well capacity and therefore the photon noise. In other words, at high ISO the DR limitation is in the light itself even before it hits the sensor. Therefore increasing the number of bits in the electronics would not increase the DR at high ISO.

Increasing the number of bits will increase the DR since the read noise is lower (much lower) at higher ISOs on Canon sensors (up to ISO 3200, or thereabouts, after which the sensor has flat read noise). That's the point of my OP.

I see now what you meant, but did not say thus making your proposal unclear.

My apologies.

You apparently assume that the read noise would remain at the same count number while the numbers are lower with more bits and therefore the read noise would reduce relatively to the signal. Did I get it right?

Yes, but it's not an assumption -- it's how things work. For example, observe the read noise for the 6D:

http://www.sensorgen.info/CanonEOS_6D.html

or any camera, for that matter. The only question is when the read noise levels out (sensors that are level from the start are called "ISOless", but all sensors become ISOless after some point).

I see. So you believe that the read noise is not an analog noise in the sensor and the connecting circuitry external to the ADC and only measured by the ADC on its input, but something that happens inside the ADC to occupy a few lowest bits. If this is true, your idea would work. Say, we increase the ADC resolution from 14 to 16 bits with everything else the same. Then we simply divide the output value by 4 to keep the 1 to 1 photon count. According to you, the read noise would remain the same in the ADC and would divide by 4 at the output thus adding 2 extra bits to the DR.

I am not familiar with the sensor technology enough to tell if your belief is true. My feeling is that it is not and that the read noise, at least in part, does reflect the value of the analog circuitry noise at the input of the ADC. In fact, if I recall correctly, the main sources of the read noise are the noise in the connection line, because the Canon ADC is not on the sensor (unlike Sony) plus the fact that a large number of pixels are read sequentially by the same ADC and this process does not account for different sensitivity of different pixels. Most likely I am not describing this correctly, but hopefully enough to illustrate that at least some of the read noise originates from outside of the ADC And would not be affected by the ADC resolution.

I have a different idea on the same issue. Why not put a separate ADC on every pixel right there on the sensor? This might allow for a virtually unlimited DR, because the pixel could be read and reset multiple times during the exposure thus eliminating the well capacity limitation. I understand that aside from the cost that is irrelevant in the long term (remember $100 per 1MB of RAM?), the issue is space, that pixels are too small to host multiple transistors in addition to photocells. I believe this limitation comes from the old 2D lithography, but there is the third dimension, the depth of the sensor where you could put as much circuitry as you need. I believe this technology is coming and hopefully not too far away.