Why do we still use analog gain with ISO invariant sensors?

[Chris Noble]

I just shot a dark corner in my room and checked the results with RawDigger ...

100 ISO: SNR 12.78

400 ISO: SNR 7.38

LOL no improvement there ... again, quite the opposite ... minus 0.8EV ...

... and no playing with words either. Just cold, hard numbers. :-D

Same exposure (f-stop and shutter) in both cases?

SNR of the darkest part of the image or of the entire frame?

And is it your observation that measured Raw SNR is well-correlated with visible noise when you have adjusted the brightness in Post?

 

[bobn2]

The A/D comes with the sensor.

If "comes with" means that they are on the same chip, the statement is not generally correct.

The sensor can be integrated with the A/D on the same chip, (Panasonic has done it and I'm sure others), but you take a performance hit as the optimal semiconductor process is different for these very different types of circuit.

Especially with video performance requirements, you need fast, low-noise, pretty-high resolution (12-16 bit) A/Ds, and also sensors with pretty deep wells. Very different semiconductor processing.

Not really true at all. A specialist process is required for image sensor, but it works just fine for ADCs. It is essentially a 'mixed mode' process, ideal for ADCs, but with rather deeper implants to give the photodiodes sufficient depth to capture the longer wavelengths.

Also, stacked sensors are becoming a thing. Canon's stacked architecture puts the ADCs on the helper chip, but I suspect that's because Canon's own sensor process has too coarse a geometry to do column ADCs well.

Most of the best performing sensors available (Sony) have column ADCs integrated onto the sensor chip. Even their stacked architecture does this.

 

[xpatUSA]

I just shot a dark corner in my room and checked the results with RawDigger ...

100 ISO: SNR 12.78

400 ISO: SNR 7.38

LOL no improvement there ... again, quite the opposite ... minus 0.8EV ...

... and no playing with words either. Just cold, hard numbers. :-D

Same exposure (f-stop and shutter) in both cases?

No. It was a real quick test. I left it in aperture priority, ergo the ISO setting changed the shutter period.

SNR of the darkest part of the image or of the entire frame?

A dark-ish selection in RawDigger, same for each image. SNR calculated as mean/std.dev.

And is it your observation that measured Raw SNR is well-correlated with visible noise when you have adjusted the brightness in Post?

Yes. Well enough for such a rough test.

Was trying to show my correspondent that upping the ISO doesn't improve SNR for every camera on the planet ... ;-)

 

[bobn2]

I'm old-school. Variable analog gain has much lower noise than any sensor or A/D and the delay is zero. Not bad.

I agree. It's in most cases a better engineering solution than a wider ADC. I'm not a fan of the term 'analog gain'. You're not gaining analogs, and 'gain' itself is an analog concept, applied to whatever quantity is being 'gained' (in the case of a camera VGA it's voltage). In the digital domain, if you make a number larger it's called multiplication.

Analog gain and digital gain were very commonly used terms when I did electronic design, but that was decades ago. However I just did a Google search and they are still commonly used by circuit designers.

I've never heard the terms "gaining analogs" or "gaining a quantity" though.

As I said, you're not gaining analogs, which was the point. On the quantity, electronic engineers will talk about voltage gain or current gain and occasionally charge gain. That's 'gaining a quantity', the quantity being voltage, current or charge.

And gain can be greater or less than unity in an electronic circuit.

Yes, they are. Within a discipline people tend to have their own framework of jargon. Electronic engineers are well accustomed to talking about 'gain', but the will say in general which kind of gain it is (unless it's obvious by contexts) so they'll talk about 'current gain' or 'voltage gain'. When digital circuits began to be incorporated into analog systems, they they found that digital multiplication operators could provide the same function as could 'gain', so they called it 'digital gain'.

In the world of computer science things were different. Circuits that performed multiplication were called 'multipliers'. For analog computing the variable gain amplifiers that did this function were still called 'multipliers' (and still are).

Photography is neither electronic engineering or computer science, so if we adopt those communities' terminology without understanding then we confuse ourselves. I say this with a fair amount of confidence, because I am an electronic engineer, a computer scientist and a photographer.

The reason I'm concerned about terminology is that poor use can lead thought patterns down a garden path. That's the case here. If we consider a camera as a black box, it takes light in at one end and puts out perceptual specifications at the other. It doesn't emit light. Inside the black box a translation is made from the input to the output, and gain is no part of that translation. There is no reason why any arbitrary amount of light might not be translated to any arbitrary lightness (lightness being the component of that perceptual specification which says how light or dark something should look). Internal 'gains' are as irrelevant to the essence that conversion as are the details of the computer code used to do it. So, you do not need to invoke 'gain' to explain how a smaller amount of light translates to a lighter image. And doing so sometimes leads to erroneous thought patterns. People logically assume that 'gain' means that something is being 'gained', and then the thing that is gained is either light or some unspecified analog to light they they often call 'signal'.

In most digital cameras, ISO control is effected over most of its range by a variable analog gain stage between the sensor and the A/D. It's a pretty simple concept, and variable gain between a sensor and an A/D is a common circuit in many applications. I speak as a photographer and also someone who used to design data-conversion circuits.

Is the gain continuously variable, or is it switched from low to high at a certain ISO number ? Bill Claff's graphs suggest the latter, and so do the occasional nuggets of information released by camera companies.

Don

There are two 'gains' at work here, which is one reason why you need to be specific about which 'gain' your talking about. The first 'gain' is a variable voltage gain stage (or, more often, two of them) at the input to the ADC. The purpose of this gain stage is to allow the full bit depth of the ADC to be used a low exposures (assumed to go along with high ISO settings). The second 'gain' is 'conversion gain', which isn't really 'gain' at all. It determines the relationship between charge in the pixel and voltage output of the pixel (which is why it sin't 'gain', different quantities in and out). This is more recent. A switchable conversion gain system was developed by Aptina (now On Semiconductor) and Sony gained access to it via a patent swap. Using this system the conversion gain is switched from low to high at one iSO setting in the range, and causes a characteristic 'step' in Bill's graphs. The reason for switching conversion gain is that it is responsible for how the downstream electronic noise looks in 'input referred' terms. That is, the higher the conversion gain the less noisy a given amount of voltage noise looks. However, a high conversion gain limits the saturation capacity of a pixel (so-called 'full well capacity' is another highly misleading term), so if you build sensors with a very high conversion gain, they will require a high base ISO. Switching the conversion gain squares this particular circle.

 

[xpatUSA]

The A/D comes with the sensor.

If "comes with" means that they are on the same chip, the statement is not generally correct.

The sensor can be integrated with the A/D on the same chip, (Panasonic has done it and I'm sure others), but you take a performance hit as the optimal semiconductor process is different for these very different types of circuit.

Especially with video performance requirements, you need fast, low-noise, pretty-high resolution (12-16 bit) A/Ds, and also sensors with pretty deep wells. Very different semiconductor processing.

Not really true at all. A specialist process is required for image sensor, but it works just fine for ADCs. It is essentially a 'mixed mode' process, ideal for ADCs, but with rather deeper implants to give the photodiodes sufficient depth to capture the longer wavelengths.

Also, stacked sensors are becoming a thing. Canon's stacked architecture puts the ADCs on the helper chip, but I suspect that's because Canon's own sensor process has too coarse a geometry to do column ADCs well.

Most of the best performing sensors available (Sony) have column ADCs integrated onto the sensor chip. Even their stacked architecture does this.

Well it turned out that @chrisfisheye didn't mean on-the-chip, so this side-discussion is a bit moot, gents.

 

[bobn2]

In most digital cameras, ISO control is effected over most of its range by a variable analog gain stage between the sensor and the A/D. It's a pretty simple concept, and variable gain between a sensor and an A/D is a common circuit in many applications. I speak as a photographer and also someone who used to design data-conversion circuits.

I would quibble with the exact choice of words. ISO control is not 'effected' by a variable 'analog' (actually voltage) stage.

Not just voltage; a lot of current gain as well due to the impedance mismatch.

Not in any example I've ever seen. It would also be remarkably poor electronic engineering that gave the ADC such a complex input impedance that it required both variable current and voltage gain to drive it. In any case, CMOS circuits (which these invariably are nowadays) are very high input impedance, so certainly don't need variable current gain to 'match' them

Rather, the control of ISO includes a change of gain over some of the ISO range. But that's not the point really. The control of ISO includes different operations, very often camera dependent. To focus on one, which is not universal, and say that ISO 'is' that one, and use it as an explanation, when it is largely irrelevant to what the ISO control actually does is just misleading. People who are not electronic engineers don't know what 'gain' is in any detail, so it's just swapping one unknown concept for another. Along with it will come misconceptions, such as that so-called 'ISO noise' is cause by 'amplification', which, if you have designed data conversion circuits, you will know is false.

I fully agree that these concepts are challenging and largely irrelevant to practicing photographers, but the concept and execution of variable signal gain between a sensor and an A/D are basic processes for experienced electronic engineers. That does not disparage photographers; electronics is a profession, like surgery and accounting, and those last two are just as mysterious to me as electronics is to those who have not done electronics design.

I'm not suggesting anything else. It makes an even stronger case for not explaining photographic concepts to lay people in electronic engineering terms, especially if one gets it wrong.

 

[xpatUSA]

I'm old-school. Variable analog gain has much lower noise than any sensor or A/D and the delay is zero. Not bad.

I agree. It's in most cases a better engineering solution than a wider ADC. I'm not a fan of the term 'analog gain'. You're not gaining analogs, and 'gain' itself is an analog concept, applied to whatever quantity is being 'gained' (in the case of a camera VGA it's voltage). In the digital domain, if you make a number larger it's called multiplication.

Analog gain and digital gain were very commonly used terms when I did electronic design, but that was decades ago. However I just did a Google search and they are still commonly used by circuit designers.

I've never heard the terms "gaining analogs" or "gaining a quantity" though.

As I said, you're not gaining analogs, which was the point. On the quantity, electronic engineers will talk about voltage gain or current gain and occasionally charge gain. That's 'gaining a quantity', the quantity being voltage, current or charge.

And gain can be greater or less than unity in an electronic circuit.

Yes, they are. Within a discipline people tend to have their own framework of jargon. Electronic engineers are well accustomed to talking about 'gain', but the will say in general which kind of gain it is (unless it's obvious by contexts) so they'll talk about 'current gain' or 'voltage gain'. When digital circuits began to be incorporated into analog systems, they they found that digital multiplication operators could provide the same function as could 'gain', so they called it 'digital gain'.

In the world of computer science things were different. Circuits that performed multiplication were called 'multipliers'. For analog computing the variable gain amplifiers that did this function were still called 'multipliers' (and still are).

Photography is neither electronic engineering or computer science, so if we adopt those communities' terminology without understanding then we confuse ourselves. I say this with a fair amount of confidence, because I am an electronic engineer, a computer scientist and a photographer.

The reason I'm concerned about terminology is that poor use can lead thought patterns down a garden path. That's the case here. If we consider a camera as a black box, it takes light in at one end and puts out perceptual specifications at the other. It doesn't emit light. Inside the black box a translation is made from the input to the output, and gain is no part of that translation. There is no reason why any arbitrary amount of light might not be translated to any arbitrary lightness (lightness being the component of that perceptual specification which says how light or dark something should look). Internal 'gains' are as irrelevant to the essence that conversion as are the details of the computer code used to do it. So, you do not need to invoke 'gain' to explain how a smaller amount of light translates to a lighter image. And doing so sometimes leads to erroneous thought patterns. People logically assume that 'gain' means that something is being 'gained', and then the thing that is gained is either light or some unspecified analog to light they they often call 'signal'.

In most digital cameras, ISO control is effected over most of its range by a variable analog gain stage between the sensor and the A/D. It's a pretty simple concept, and variable gain between a sensor and an A/D is a common circuit in many applications. I speak as a photographer and also someone who used to design data-conversion circuits.

Is the gain continuously variable, or is it switched from low to high at a certain ISO number ? Bill Claff's graphs suggest the latter, and so do the occasional nuggets of information released by camera companies.

Don

There are two 'gains' at work here, which is one reason why you need to be specific about which 'gain' your talking about. The first 'gain' is a variable voltage gain stage (or, more often, two of them) at the input to the ADC. The purpose of this gain stage is to allow the full bit depth of the ADC to be used a low exposures (assumed to go along with high ISO settings). The second 'gain' is 'conversion gain', which isn't really 'gain' at all. It determines the relationship between charge in the pixel and voltage output of the pixel (which is why it sin't 'gain', different quantities in and out). This is more recent. A switchable conversion gain system was developed by Aptina (now On Semiconductor) and Sony gained access to it via a patent swap. Using this system the conversion gain is switched from low to high at one iSO setting in the range, and causes a characteristic 'step' in Bill's graphs. The reason for switching conversion gain is that it is responsible for how the downstream electronic noise looks in 'input referred' terms. That is, the higher the conversion gain the less noisy a given amount of voltage noise looks. However, a high conversion gain limits the saturation capacity of a pixel (so-called 'full well capacity' is another highly misleading term), so if you build sensors with a very high conversion gain, they will require a high base ISO. Switching the conversion gain squares this particular circle.

An excellent clarification!

For example, my Sigma's F7 sensor has a conversion factor of 7.14 uV/e- and a FWC of 77,000 e- for what that's worth ...

... both figures for the so-called 'green' layer.

 

[bobn2]

The A/D comes with the sensor.

If "comes with" means that they are on the same chip, the statement is not generally correct.

The sensor can be integrated with the A/D on the same chip, (Panasonic has done it and I'm sure others), but you take a performance hit as the optimal semiconductor process is different for these very different types of circuit.

Especially with video performance requirements, you need fast, low-noise, pretty-high resolution (12-16 bit) A/Ds, and also sensors with pretty deep wells. Very different semiconductor processing.

Not really true at all. A specialist process is required for image sensor, but it works just fine for ADCs. It is essentially a 'mixed mode' process, ideal for ADCs, but with rather deeper implants to give the photodiodes sufficient depth to capture the longer wavelengths.

Also, stacked sensors are becoming a thing. Canon's stacked architecture puts the ADCs on the helper chip, but I suspect that's because Canon's own sensor process has too coarse a geometry to do column ADCs well.

Most of the best performing sensors available (Sony) have column ADCs integrated onto the sensor chip. Even their stacked architecture does this.

Well it turned out that @chrisfisheye didn't mean on-the-chip, so this side-discussion is a bit moot, gents.

Well, if he didn't mean 'on the chip' then he's wrong. There's any number of analog front-end chips that one could choose to use with a any given analog output image sensor. Nowadays just about every sensor (including Canon) has integrated ADCs, so it's still moot.

 

[bobn2]

With a fixed shutter speed and aperture, there are two ways we can brighten an image:

- by using a higher ISO (analog gain)

- by increasing the exposure slider in post (digital gain)

My understanding is that most modern sensors are approximately ISO invariant (beyond a second base ISO, if they have one), meaning that these two approaches will produce very similar noise levels in the final image.

But the digital gain approach retains the maximum amount of highlight room, while every additional stop of ISO decreases highlight room by a stop. As a result, by using ISO instead of digital gain we could lose multiple stops of dynamic range without seeing any real noise improvement.

Wouldn't we be better off if, beyond their second base ISO, cameras just baked an exposure adjustment into the exif info instead of applying more analog gain? So if "proper" image brightness required ISO 1600, the camera would instead use its highest base ISO (say, ISO 400) and then tell Lightroom to start at +2 on the exposure slider. Compared to using ISO 1600 we'd get more highlight protection and about the same noise.

Why do cameras not do this? Is there a problem I'm not seeing?

The sensor has no way of "knowing" what comes after it and how ISO control is effected. The sensor is not affected by the ISO control.

The stage that is affected by the ISO control is the A/D. If the A/D has sufficient ENOB (effective number of bits) to capture all of the data from the sensor, it is ISO invariant.

So we should really be talking about cameras with ISO-invariant digital conversion.

The A/D comes with the sensor.

If "comes with" means that they are on the same chip, the statement is not generally correct.

Actually, it is nowadays. Any camera with a Sony or Panasonic (or whoever is selling them now, I lose track) image sensor has its ADCs integrated on the sensor chip, as do most cameras with a Canon sensor. That makes it 'generally correct' in my eyes.

 

[xpatUSA]

The A/D comes with the sensor.

If "comes with" means that they are on the same chip, the statement is not generally correct.

Actually, it is nowadays. Any camera with a Sony or Panasonic (or whoever is selling them now, I lose track) image sensor has its ADCs integrated on the sensor chip, as do most cameras with a Canon sensor. That makes it 'generally correct' in my eyes.

Oh boy ...

... neither my correspondent nor myself mentioned a sensor era.

However, to keep the peace, I take your point.

--
what you got is not what you saw ...

 

[Iliah Borg]

In most digital cameras, ISO control is effected over most of its range by a variable analog gain stage between the sensor and the A/D. It's a pretty simple concept, and variable gain between a sensor and an A/D is a common circuit in many applications. I speak as a photographer and also someone who used to design data-conversion circuits.

I would quibble with the exact choice of words. ISO control is not 'effected' by a variable 'analog' (actually voltage) stage.

Not just voltage; a lot of current gain as well due to the impedance mismatch.

Not in any example I've ever seen. It would also be remarkably poor electronic engineering that gave the ADC such a complex input impedance that it required both variable current and voltage gain to drive it. In any case, CMOS circuits (which these invariably are nowadays) are very high input impedance, so certainly don't need variable current gain to 'match' them

Rather, the control of ISO includes a change of gain over some of the ISO range. But that's not the point really. The control of ISO includes different operations, very often camera dependent. To focus on one, which is not universal, and say that ISO 'is' that one, and use it as an explanation, when it is largely irrelevant to what the ISO control actually does is just misleading. People who are not electronic engineers don't know what 'gain' is in any detail, so it's just swapping one unknown concept for another. Along with it will come misconceptions, such as that so-called 'ISO noise' is cause by 'amplification', which, if you have designed data conversion circuits, you will know is false.

I fully agree that these concepts are challenging and largely irrelevant to practicing photographers, but the concept and execution of variable signal gain between a sensor and an A/D are basic processes for experienced electronic engineers. That does not disparage photographers; electronics is a profession, like surgery and accounting, and those last two are just as mysterious to me as electronics is to those who have not done electronics design.

I'm not suggesting anything else. It makes an even stronger case for not explaining photographic concepts to lay people in electronic engineering terms, especially if one gets it wrong.

Photographic terms would be push, pull, intensifier, reducer, sensitization, hypersensibilization ?

 

[xpatUSA]

I don't agree with the your assumption that a camera's ISO setting is an analog adjustment.

Agreed.

It's not like volume control by means of a potentiometer.

With many if not most systems, the ISO value is set by a digital input to a Programmable Gain Amplifier.

 

[bobn2]

With many if not most systems, the ISO value is set by a digital input to a Programmable Gain Amplifier.

Not really. The ISO value is set by doing a lot of things to the image. It might (and usually does) involve a variable gain amplifier, but quite rarely is the gain of that (or those, there is often two of them)amplifier set directly according to the iSO.

In the very early days, when the processing chips in cameras were slow, 16 bit affairs without multipliers, dedicated imaging pipelines, SIMD or vector instructions, a bit of analog computing, via the use of a VGA to scale the input to the ADC was a useful thing to do. More recently, variable gain has been more about optimising the use of the VGA.

 

[bobn2]

In most digital cameras, ISO control is effected over most of its range by a variable analog gain stage between the sensor and the A/D. It's a pretty simple concept, and variable gain between a sensor and an A/D is a common circuit in many applications. I speak as a photographer and also someone who used to design data-conversion circuits.

I would quibble with the exact choice of words. ISO control is not 'effected' by a variable 'analog' (actually voltage) stage.

Not just voltage; a lot of current gain as well due to the impedance mismatch.

Not in any example I've ever seen. It would also be remarkably poor electronic engineering that gave the ADC such a complex input impedance that it required both variable current and voltage gain to drive it. In any case, CMOS circuits (which these invariably are nowadays) are very high input impedance, so certainly don't need variable current gain to 'match' them

Rather, the control of ISO includes a change of gain over some of the ISO range. But that's not the point really. The control of ISO includes different operations, very often camera dependent. To focus on one, which is not universal, and say that ISO 'is' that one, and use it as an explanation, when it is largely irrelevant to what the ISO control actually does is just misleading. People who are not electronic engineers don't know what 'gain' is in any detail, so it's just swapping one unknown concept for another. Along with it will come misconceptions, such as that so-called 'ISO noise' is cause by 'amplification', which, if you have designed data conversion circuits, you will know is false.

I fully agree that these concepts are challenging and largely irrelevant to practicing photographers, but the concept and execution of variable signal gain between a sensor and an A/D are basic processes for experienced electronic engineers. That does not disparage photographers; electronics is a profession, like surgery and accounting, and those last two are just as mysterious to me as electronics is to those who have not done electronics design.

I'm not suggesting anything else. It makes an even stronger case for not explaining photographic concepts to lay people in electronic engineering terms, especially if one gets it wrong.

Photographic terms would be push, pull, intensifier, reducer, sensitization, hypersensibilization ?

Maybe some of those.

 

[Chris Noble]

I just shot a dark corner in my room and checked the results with RawDigger ...

100 ISO: SNR 12.78

400 ISO: SNR 7.38

LOL no improvement there ... again, quite the opposite ... minus 0.8EV ...

... and no playing with words either. Just cold, hard numbers. :-D

Same exposure (f-stop and shutter) in both cases?

No. It was a real quick test. I left it in aperture priority, ergo the ISO setting changed the shutter period.

Sorry, but I think you missed the essential point that the discussion is about increasing ISO at constant exposure. In your test, what reduced your SNR was the decrease in exposure. The change in ISO had no impact.

 

[xpatUSA]

With many if not most systems, the ISO value is set by a digital input to a Programmable Gain Amplifier.

Not really. The ISO value is set by doing a lot of things to the image. It might (and usually does) involve a variable gain amplifier, but quite rarely is the gain of that (or those, there is often two of them)amplifier set directly according to the iSO.

In the very early days, when the processing chips in cameras were slow, 16 bit affairs without multipliers, dedicated imaging pipelines, SIMD or vector instructions, a bit of analog computing, via the use of a VGA to scale the input to the ADC was a useful thing to do. More recently, variable gain has been more about optimising the use of the VGA.

I fold.

 

[tbcass]

In that case, the change in shutter speed changed the exposure and therefore the S/N ratio.

 

[xpatUSA]

In that case, the change in shutter speed changed the exposure and therefore the S/N ratio.

All of a sudden, I'm the bad guy in this thread!

The lower exposure gave a lower mean signal value but also and consequently a lower standard deviation. So can someone please explain to why the SNRs in that quick rough test can not be compared ??

Also can someone please define the exact exposures needed for a perfect, satisfactory test that would actually pass muster here?

 

[tbcass]

In that case, the change in shutter speed changed the exposure and therefore the S/N ratio.

All of a sudden, I'm the bad guy in this thread!

The lower exposure gave a lower mean signal value but also and consequently a lower standard deviation. So can someone please explain to why the SNRs in that quick rough test can not be compared ??

Also can someone please define the exact exposures needed for a perfect, satisfactory test that would actually pass muster here?

Try full manual control, set the aperture and shutter speed, and change only the ISO. By doing that the exposure remains constant with the only variable being any gain supplied by ISO. I'm curious to find out your results. I'm no expert and maybe I'm wrong but that seems like the best approach to me.

--
Tom

 

[bodeswell]

With many if not most systems, the ISO value is set by a digital input to a Programmable Gain Amplifier.

A clear assertion, I can't judge whether it is true or false.

Not really.

It is followed by an ambiguous negation.

The ISO value is set by doing a lot of things to the image.

Perhaps I should read "is implemented" for "is set"? In the vernacular the ISO value "is set" (I will not hedge with "often") by turning a dial.

It might (and usually does) involve a variable gain amplifier, but quite rarely is the gain of that (or those, there is often two of them)amplifier set directly according to the iSO.

"Directly" powerfully adds to the overall ambiguity. Was xpatUSA right but insufficiently precise in his formulation? It's hard to say.

In the very early days, when the processing chips in cameras were slow, 16 bit affairs without multipliers, dedicated imaging pipelines, SIMD or vector instructions, a bit of analog computing, via the use of a VGA to scale the input to the ADC was a useful thing to do.

We meander down Memory Lane. Many technical terms are mentioned. They don't clear anything up.

More recently, variable gain has been more about optimising the use of the VGA.

The coup de grace, though no particular light is shed.

I fold.

That's what happens when you enter the twilight zone.

This exchange is what I mean by clear as milk.

Please note that I follow these threads in Open Talk with interest.