Sensor Sensitivity: an ISO question

[dje]

Hi Dave,

So, if the constant 10 refers to 18% saturation and the constant 78 refers to 100% saturation less 1/2 a stop, should the constant used for a saturated sensor (full well) be 78 lx.s plus 1/2 a stop, i.e. 110 lx.s?

Ted it's really all a bit beyond me but wouldn't it be nice to be twenty years younger and working as a designer for a camera manufacturer !!

Dave

[Mike Buckley]

you've got to allow us...to "excercise our minds"

Being able to understand this thread is unimportant. (I'm proof of that.) Reviewing it on a regular basis will surely exercise the brain cells enough to ward off Alzheimer's disease.

[blakemcguire]

Math is fun, photography is fun, why not enjoy both? :P

[xpatUSA]

No it doesn't Colin but you've got to allow us old codgers, with too much time on our hands [:-)] and our grasp of technical issues slipping gradually away, a little bit of latitude to "excercise our minds" !!

Dave

At 72, I need all the exercise I can get . .

While seated today, the final penny dropped re: the mysterious constants 10 & 78.

The formula S = 10/H applies, as you said, to the 18% saturation case.
The formula S = 78/H applies to the saturated case but with 1/2 stop headroom.

H, the exposure at the sensor face, is therefore different for each of the two cases.

So, it may well be that S doesn't change between cases for a given sensor, in theory anyway.

[Simon Garrett]

No it doesn't Colin but you've got to allow us old codgers with too much time on our hands and our grasp of technical issues sliiping gradually away a little bit of latitude to "excercise our minds" !!
Dave

Quite! And you could say that any technical information beyond "point the camera and press the button" is a matter of diminishing returns. The more detailed the information, the smaller the incremental benefit to a photographer. But for those of us with a technical background it's fun!

However, a couple of things come out of this that reinforce what most of know anyway:

• For digital cameras, ISO ratings are somewhat arbitrary, especially if the REI method is uses
• For that matter, metering is somewhat arbitrary, especially for matrix (multi-zone) metering

In other words, as most of us have found out, you have to learn the metering of each camera you use. However deeply technical the matter is, there's still a large subjective element - judgement to be exercised by the photographer.

So I suppose Colin's right, really

[sarangiman]

Mike; while I'm by no means an expert, I do have a basic understanding of how sensors work. The simple answer to your question is no. You cannot separate the photo detector part of the sensor from the amplifier part; they are part and parcel of the same piece of hardware. The actual signal from the photodiodes is so low as to be impossible for other circuit elements to use, so amplification is necessary so that other external circuits can actually use it. Circuits don't care about ISO settings, rather they need a specific voltage range to operate. As photographers we do care and this base voltage requirement defines the base ISO of the sensor.

Manfred, thank you for that response. Do you, furthermore, know if in modern CMOS designs the amplification at the pixel level is, in fact, separate from an ISO amplifier downstream? Or are they one & the same?

Might base ISO (e.g. 100 for most cameras) still apply some level of amplification simply to match the input voltage requirements of the ADC?

Or am I thinking about this incorrectly? I have a general curiosity about this, but also specifically wish to understand the validity of building models to fit constants like conversion gain, read noise, etc., for any given camera.

[jcuknz]

I'm an 'old codger' but with not enough time and countless projects waiting to be started ... and finished ... so I'll just type this and move on ... have a great day

[Manfred M]

Manfred, thank you for that response. Do you, furthermore, know if in modern CMOS designs the amplification at the pixel level is, in fact, separate from an ISO amplifier downstream? Or are they one & the same?

Might base ISO (e.g. 100 for most cameras) still apply some level of amplification simply to match the input voltage requirements of the ADC?

Or am I thinking about this incorrectly? I have a general curiosity about this, but also specifically wish to understand the validity of building models to fit constants like conversion gain, read noise, etc., for any given camera.

Sorry, I don't know and would only be guessing. From what I can glean from the literature, there are at least two levels of amplifier on the sensor. The first is incorporated into the pixel assembly itself and is very compact to minimize interfering with the light measurement functionality. The second seems to be there and boosts the signal some more as the sensor data is read and transferred off chip. Unfortunately, the manufacturers are not all that forthcoming as to how they implement the various camera functions.

[dje]

I can't claim any expertise either but from what I've read I would agree with Manfred. Each pixel has an amplifier built in and I believe these would be fixed gain. A second stage of amplfication lies between the pixel array and the A/D converter. This would be gain controlled by the camera's ISO setting.

Dave

[sarangiman]

I'm an 'old codger' but with not enough time and countless projects waiting to be started ... and finished ... so I'll just type this and move on ... have a great day

I'm utterly perplexed as to the purpose of this response.

I can't claim any expertise either but from what I've read I would agree with Manfred. Each pixel has an amplifier built in and I believe these would be fixed gain. A second stage of amplfication lies between the pixel array and the A/D converter. This would be gain controlled by the camera's ISO setting.

Dave

Yes, this'd be my guess from everything I've read as well. I'm attempting to measure the read noise upstream vs. downstream of this second 'ISO' amplifier for a number of cameras, with the idea that the upstream noise's main contributors being dark current & any other noise associated with the circuitry measuring the charge stored in the pixel's capacitors. Was hoping that latter bit was valid and, yes, it's very hard to find this information for specific sensors (not that I blame them; this is pretty esoteric, but fascinating to me, stuff).

Thanks very much Manfred & Dave.

[dje]

In that case you may be interested in this article on Noise and DR (if you haven't already seen it).

Also the measurement results in the DXO sensor comparison Database might be useful.

Dave

[sarangiman]

In that case you may be interested in this article on Noise and DR (if you haven't already seen it).

Also the measurement results in the DXO sensor comparison Database might be useful.

Dave

Thanks Dave -- yes I'm aware of both of those resources. Emil's work is fantastic, but I was hoping to dive a little further into the architecture to address some of the trends that become evident when doing this sort of modeling to determine performance aspects of different sensors. For example, read noise across sensors of any one generation often seem to independent of pixel size... I'm curious about what architectural design is the reason behind this; e.g., whether it has to do with the capacitors, transistors, or pixel amplifiers, or some combination thereof.

[Inkanyezi]

I look at ISO for digital cameras as follows :

• ISO values are a carry over from the days of film and were a measure of the speed of the film
• The film speed was determined by the chemical make-up of the film. The higher the speed, the more grainy-ness (noise)
• With digital, an attempt to retain the same sensitivity ratings as for film was made by introducing ways of "standardiing" ISO ratings for digital cameras. This allowed the same exposure settings "triangle" to be retained (shutter speed, aperture and ISO).
• With digital, the same compromise applies. With the base ISO, the noise is as low as it gets. To get a higher ISO setting and hence more sensitivity to light, amplication must be added and this adds noise.

That's my simple view of the world !!

Dave

Although it deviates from the thread topic, I will challenge these conceptions, as I believe they can do some harm in understanding the basic properties of both film and digital sensors.

Grain and noise are two completely different entities, and even though they may seem visually similar in a finished image, they are so widely different, that we should not simplify it to equalling noise with grain.

The noise threshold of electronic media is more akin to the underlying veiling of film. Veil is the density a film has when developed without being exposed to light. Grain is different altogether. Grain is what comprises the image, it is entirely built up of grains, although those grains may be substituted with dyes that take their place. Also the veil will display graininess, but grain in itself is not related in any way to the noise of electronic media.

Note that I use the term "electronic media" and not digital. The noise we are discussing here is analogue, before conversion of the signal to digital. We have similar noise also in analogue television tubes, as those used in the fifties and sixties, before the CCD chip came into our trade.

Higher amplification pushes more of the (analogue) data over the maximum threshold, up through the roof, where it cannot be recorded. There is an absolute maximum output, defined by the max voltage.

BUT - amplifying also raises the noise level, and the same noise is present at all voltage levels of the sensor.

Here's where noise is definitely different from grain. The film had its grain at all levels, and it was a necessary property, without which there would be no image. The noise of electronic media is always unwanted, so we suppress it if we can. It will always be less important when the signal level is high, and it can be regarded invisible in the image, above certain levels. High amplification will shrink dynamic range, and in those less exposed areas that are then pushed up to visible levels, noise becomes more apparent. So raising ISO does not increase noise any more than it increases the signal level. It's only that in the final image, we use signal levels for the brighter areas of the image, where noise is more apparent.

We can make an analogy with film of this, because "pressing" film to a higher exposure index is very much like raising the magnification of the electric output of a sensor. When we under-expose and over-develop a film, we raise contrast, thus shrink the dynamic range that will be reproduced in the final image. "More graininess" is not a result of under-exposure itself, but graininess may change appearance because of processing differences. Longer processing time and higher temperature will make grain coarser in the film, thus creating "more graininess", regardless of exposure level. It should be noted that using a higher exposure index with harder development does not change the speed of the film, much similarly with raising amplification when using electronic media. The sensitivity remains the same, only amplification is changed. We use a higher exposure index, thus under-exposing the medium, and then we compensate for this by either development or amplification. In both cases we lose dynamic range, side-stepping the ISO specification, hopefully with some creative intention.

[Manfred M]

Although it deviates from the thread topic, I will challenge these conceptions, as I believe they can do some harm in understanding the basic properties of both film and digital sensors.

Grain and noise are two completely different entities, and even though they may seem visually similar in a finished image, they are so widely different, that we should not simplify it to equalling noise with grain.

The noise threshold of electronic media is more akin to the underlying veiling of film. Veil is the density a film has when developed without being exposed to light. Grain is different altogether. Grain is what comprises the image, it is entirely built up of grains, although those grains may be substituted with dyes that take their place. Also the veil will display graininess, but grain in itself is not related in any way to the noise of electronic media.

Note that I use the term "electronic media" and not digital. The noise we are discussing here is analogue, before conversion of the signal to digital. We have similar noise also in analogue television tubes, as those used in the fifties and sixties, before the CCD chip came into our trade.

Higher amplification pushes more of the (analogue) data over the maximum threshold, up through the roof, where it cannot be recorded. There is an absolute maximum output, defined by the max voltage.

BUT - amplifying also raises the noise level, and the same noise is present at all voltage levels of the sensor.

Here's where noise is definitely different from grain. The film had its grain at all levels, and it was a necessary property, without which there would be no image. The noise of electronic media is always unwanted, so we suppress it if we can. It will always be less important when the signal level is high, and it can be regarded invisible in the image, above certain levels. High amplification will shrink dynamic range, and in those less exposed areas that are then pushed up to visible levels, noise becomes more apparent. So raising ISO does not increase noise any more than it increases the signal level. It's only that in the final image, we use signal levels for the brighter areas of the image, where noise is more apparent.

We can make an analogy with film of this, because "pressing" film to a higher exposure index is very much like raising the magnification of the electric output of a sensor. When we under-expose and over-develop a film, we raise contrast, thus shrink the dynamic range that will be reproduced in the final image. "More graininess" is not a result of under-exposure itself, but graininess may change appearance because of processing differences. Longer processing time and higher temperature will make grain coarser in the film, thus creating "more graininess", regardless of exposure level. It should be noted that using a higher exposure index with harder development does not change the speed of the film, much similarly with raising amplification when using electronic media. The sensitivity remains the same, only amplification is changed. We use a higher exposure index, thus under-exposing the medium, and then we compensate for this by either development or amplification. In both cases we lose dynamic range, side-stepping the ISO specification, hopefully with some creative intention.

A very complete overview, Urban.

The only other analogy that seems to be consistent, whether we apply additional gain using an electronic sensor or conventional film; we are trading off grain size / noise levels AND image quality. Pushed film has lower contrast and detail and we lose dynamic range as we increase ISO values in digital photography.

As with anything else in photography (and life) we are managing tradeoffs....

[dje]

Grain and noise are two completely different entities, and even though they may seem visually similar in a finished image, they are so widely different, that we should not simplify it to equalling noise with grain.

Urban it was not my intention to imply that film "grainy-ness" and electrical noise in digital cameras were the same. Rather I was drawing a very simple analogy between the image quality degradation in film and digital associated with high ISO. I used the term noise in a very broad sense- ie something that visibly degrades the image.

I don't think anything I said about electrical noise and ISO setting is incorrect. I simply avoided discussing the impications of higher noise levels associated with more amplifier gain.

Dave

[Inkanyezi]

Urban it was not my intention to imply that film "grainy-ness" and electrical noise in digital cameras were the same. Rather I was drawing a very simple analogy between the image quality degradation in film and digital associated with high ISO. I used the term noise in a very broad sense- ie something that visibly degrades the image.

I don't think anything I said about electrical noise and ISO setting is incorrect. I simply avoided discussing the impications of higher noise levels associated with more amplifier gain.

Dave

Somehow, my point probably did not come through, the crucial one being in the closing sentence: hopefully with some creative intention. The rest of it was most techno-babble, which might not be too easily interpreted and the intention of which could be misunderstood. Maybe because I stated that noise is always unwanted. Maybe it is, but it is unavoidable, it is an inherent property of our photographic medium.

Noise is a technical trade-off, but grain is not. Grain is the essence of silver-based photography, and if a comparison should be made, grain is in a way similar to the pixels of the digital image, or the sweep and line resolution of an analogue television image. Grain is what builds the image, without grain, it would be blank, absolutely blank.

A coarser grain is somewhat like fewer pixels, but the fact that grain became somewhat coarser when pressing film was not much of an issue. However, fast film had coarser grain than slower films, and the grain of fast films was loved by many photographers, and it was used as a means of expression. When high speed film grain was made finer with more refined film making techniques in the latter decades, many photographers would still prefer the older films with coarser grain and a different kind of sharpness. We looked at sharpness in a different way, and grain sharpness in the print was a sign of darkroom work well done.

Photography is an expressive art, depicting the play of light over our scenes. In photography with small negative sizes, grain is very apparent, particularly when using high speed film before chromogenic and T-grain technologies. Grain is part of the very expression, and maybe we can learn to use noise expressively too. In the usual technobabble noise is mostly overrated as a detractor of image quality.

Mostly noise is not an issue, and in that respect, in our vision, they appear similar. It is an unavoidable technical aspect of our images made with a particular technique. When we paint with oil on canvas, the brush strokes, and the canvas itself, are in a similar way inherent properties of the technique we are using. Aquarel is different, so is charcoal or pastel crayons. Different techniques all have their quirks.

So what I missed to put forth is that noise is not necessarily degrading any image, but those where noise is most unwanted. As a creative artist, I must understand when I provoke noise, and learn to use it creatively, instead of just regarding it as a negative component in the image, which always will degrade and sometimes destroy the image that I create. Photography, as I see it, is a creative art, and as all media used for art, it has certain technical properties, inherent in the media, that are used for expression.

[jcuknz]

I'm utterly perplexed as to the purpose of this response. .

Sorry to confuss a newcomer to the forum but I was expressing my complete disinterest in the subject as really it has nothing to do with photography as I practice it ... a gentle rib at the technicians from an operator