Question for Barry re ISO settings (or anyone else who has a definitive answer)

[Rod Pike]

What is the effect of reducing ISO setting to say 200? Does it improve anything or just reduce sensitivity? are there any detrimental effects? I recall from my still days that slower films offered greater latitude, but not sure if this applies to digital imagery on video. (edit: I am referring to the standard iso400/0db setting)
Cheers!
Rod

 

[Ken Hull]

Rod,
The more you increase gain (ISO setting), the more likely you will see noise in the dark areas of the shot. Regarding latitude, I've read a few posts from experienced videographers, saying they found that using absolute lowest ISO setting available on the camera (I don't recall which brands/models) seemed to slightly REDUCE dynamic range! Sounds counter-intuitive, but that's what they say happened. So, if this is true for the AF100, then maybe you should avoid 200 ISO, and use 400 or 500 for you "slow" ISO speed. BTW, I've found I can get very good images with the ISO as high as 800, but I keep the detail setting low. A high detail setting can emphasize noise; and it doesn't REALLY increase sharpness, it's just an illusion. Barry's book and DVD talk about this and other tips for getting quality images.

 

[Shooter]

I think this is a scenario that one does controlled tests for.

Personally...I shoot lowest ISO as possible...and I have never tested them.

 

[Rod Pike]

Rod,
The more you increase gain (ISO setting), the more likely you will see noise in the dark areas of the shot. Regarding latitude, I've read a few posts from experienced videographers, saying they found that using absolute lowest ISO setting available on the camera (I don't recall which brands/models) seemed to slightly REDUCE dynamic range! Sounds counter-intuitive, but that's what they say happened. So, if this is true for the AF100, then maybe you should avoid 200 ISO, and use 400 or 500 for you "slow" ISO speed. BTW, I've found I can get very good images with the ISO as high as 800, but I keep the detail setting low. A high detail setting can emphasize noise; and it doesn't REALLY increase sharpness, it's just an illusion. Barry's book and DVD talk about this and other tips for getting quality images.

Hi Ken, yes thanks for that. I have carried out some trials and must say I can see no visual difference between 200 and 400. I also read the comments re decreased D/R- I believe it was a canon camera- hence the question-it seems counter intuitive.. Barry's book is my bible for this camera (and I have the 160A as well) and I have always kept my general purpose scene file as follows for excellent results: detail-5 vdetail-5 detail coring -3 Gamma b.press knee low and matrix norm1 (occasionally norm2).
Cheers!
Rod

 

[Rod Pike]

Just to clarify: I have assumed that 0db/iso400 is no electronic enhancement for sensitivity, so what electronically is happening when these figures go negative? ( I wish I had said that the first time!)

Cheers! Rod

 

[harddrive]

Regarding latitude, I've read a few posts from experienced videographers, saying they found that using absolute lowest ISO setting available on the camera (I don't recall which brands/models) seemed to slightly REDUCE dynamic range! Sounds counter-intuitive, but that's what they say happened.

In very general terms, it's true that there may be considered to be a "sweet" setting for ISO (or gain) - raising it has the detrimental effect of increasing noise (which is fairly easy to understand), lowering it (negative gain) has the detrimental effect of adversely affecting highlight performance. Yes - maybe counterintuitive, but I'll try to explain why it happens in principle.

Firstly, think only of sensor output, without any processing applied. From a certain threshold, increase the light falling on it, and the output will increase proportionally - up to a point when you reach saturation. Put on more light, and the output won't get any higher.

In most modern video cameras, "white" in the picture corresponds NOT to 100% of sensor output, but to a lower level - let's assume it's 25% of the maximum.

And this is where the "knee" comes in. It acts to compress brightness changes in the highlights. Acts so that the output doesn't hard clip when the sensor output exceeds 25% of max, but means that the change from 25 to 100% is reflected in the output by a far smaller change. The image still sees detail in the highlights - but the difference between "bright" and "very bright" is lessened. The camera will still see detail if the level is 4x nominal "white" - but such won't be 4x as bright any more.

Now, apply negative gain, and it must mean that the sensor output has to be higher to give the same output level as at 0dB, yes? Reduce it by 6dB, and to make nominal output "white", the sensor must now produce not 25% of max, but 50%. And that reduces the highlight headroom. Now you can only overexpose by one stop before getting to sensor clipping, not two.

It's strictly not quite true to say negative gain reduces dynamic range - such is set by chip characteristics - more accurate to say it redistributes the dynamic range between high and lowlights. Introduce negative gain and the shadows should benefit at the expense of the highlights. In practice, it may be that the noise floor is set by other factors, so then the potential shadow improvement may get wasted - so then, yes, the net effect will indeed be a reduction in dynamic range, though to say "highlight handling" is probably a more accurate way to describe it.

Does all that make sense?

 

[jdv]

Awesome post Harddrive - thanks!

 

[Rod Pike]

In very general terms, it's true that there may be considered to be a "sweet" setting for ISO (or gain) - raising it has the detrimental effect of increasing noise (which is fairly easy to understand), lowering it (negative gain) has the detrimental effect of adversely affecting highlight performance. Yes - maybe counterintuitive, but I'll try to explain why it happens in principle.

Firstly, think only of sensor output, without any processing applied. From a certain threshold, increase the light falling on it, and the output will increase proportionally - up to a point when you reach saturation. Put on more light, and the output won't get any higher.

In most modern video cameras, "white" in the picture corresponds NOT to 100% of sensor output, but to a lower level - let's assume it's 25% of the maximum.

And this is where the "knee" comes in. It acts to compress brightness changes in the highlights. Acts so that the output doesn't hard clip when the sensor output exceeds 25% of max, but means that the change from 25 to 100% is reflected in the output by a far smaller change. The image still sees detail in the highlights - but the difference between "bright" and "very bright" is lessened. The camera will still see detail if the level is 4x nominal "white" - but such won't be 4x as bright any more.

Now, apply negative gain, and it must mean that the sensor output has to be higher to give the same output level as at 0dB, yes? Reduce it by 6dB, and to make nominal output "white", the sensor must now produce not 25% of max, but 50%. And that reduces the highlight headroom. Now you can only overexpose by one stop before getting to sensor clipping, not two.

It's strictly not quite true to say negative gain reduces dynamic range - such is set by chip characteristics - more accurate to say it redistributes the dynamic range between high and lowlights. Introduce negative gain and the shadows should benefit at the expense of the highlights. In practice, it may be that the noise floor is set by other factors, so then the potential shadow improvement may get wasted - so then, yes, the net effect will indeed be a reduction in dynamic range, though to say "highlight handling" is probably a more accurate way to describe it.

Does all that make sense?

Thanks Harddrive, I have a better understanding now, that explanation would not have occurred to me- I assumed that a minimum/negative ISO/gain would have improved everything across the range of tones at the expense of sensitivity- facts always clarify assumptions!
Cheers!
Rod

 

[harddrive]

Thanks Harddrive, I have a better understanding now, that explanation would not have occurred to me- .......

I'm old enough to remember tube cameras (as an engineer) and in those days the battle was to achieve an acceptable noise figure at the low end. Hence white tended to correspond closely with max tube output if I remember right, so no head room even available to think about a knee - signal got to white and clipped! It was one of the things that then distinguished "video" from "film" look, and to the detriment of video it has to be said!

Even earlier cameras were even worse if the tube got overloaded, and gave "comet tails" on highlights - if you're interested, there's a lot about it in the description of the EMI2001 at http://en.wikipedia.org/wiki/EMI_2001 (Probably the most common colour camera in the UK during the 1970's.) As the article says, ACT circuits greatly lessened that problem, but it was still necessary for white to mean near peak tube signal - so forget about highlight subtleties!

If negative gain had been possible on the cameras of the time, it would have meant that the output could never even have reached a full volt - no matter how bright the scene or how much the iris was opened.

The cameras did have a positive gain option, but from memory even small amounts increased the gain noticeably (and undesirably!). It wasn't until tubes got a better dynamic range that it was even possible to think about a knee, highlight rolloff, and nominal white being given by anything much less than full sensor output.

It's important to distinguish between input dynamic range and output dynamic range. Output dynamic range normally corresponds to something like 8 bit video - which is why it's so widely used - and it roughly corresponds to what the human eye can distinguish in a single scene. So if you ask "why do we even think about 12 stops of DR then?", it's because at some point the original signal will get it's luminance levels condensed down to about 8 stops output - but not in a uniform manner. If the compression was uniform, the output would look very flat. Hence compress the highlights far more than the midtones.

If that gets done in camera, it's fine if it's correct and factors like 10 bit, RAW etc won't then add anything. But the value of 10 bit is that it allows for more flexibility in post. It gives flexibility for having more ability to redistribute luminance values post recording without detriment. It's also worth saying that for acquisition, 10 bit is only really worthwhile if the original signal justifies it in terms of S/N, and it really needs something like s-log to make much difference.