ASA settings

[Mike Williamson]

I worked on a short test yesterday for the Red camera shooting some available light at night, which lead to some questions about how the Red camera does different ASA settings and how it compares to Canon's digital still cameras.

 

On one hand, I've read that in the Red camera, the ASA is a metadata setting that can be changed anywhere in the workflow. To me, this would imply that the sensor has a native sensitivity that is not changed prior to outputting a digital signal/RAW file.

 

On the other hand, I've read that the Red camera works "the same as a digital stills camera". Using my Canon camera, there is a definite difference in a shot taken at 400 ASA and boosted up to 1600 ASA afterwards, and a shot taken at 1600 ASA in camera.

 

Looking at the basic construction of a CMOS chip, the diagrams I've looked at show a video processing amplifier as part of the chip itself, amplifying the signal before it goes through the A/D conversion. This would suggest that there is the potential to alter the sensitivity of the chip itself, though I don't know whether this is done.

 

After reading some Canon white papers, my unconfirmed guess is that Canon DSLR's do change the sensitivity of the chip, while the Red camera does not. Hopefully I'll know more after I look at the Red footage in color correction later this week. Anyone who can shed some light on this would be appreciated, I don't have an agenda besides understanding how the two cameras compare with one another. Thanks for the help!

[Brian Drysdale]

You can't adjust the sensitivity of the RED chip itself, the ASA is in the metadata and everything takes place in post. Interestingly, the alternative method is available in the SI 2K, so that gain can be applied on the camera chip.

[Peter Moretti]

Yes, from what I understand the Red always shoots at 320 ASA. Any ASA changes are to metadata only.

[Mike Williamson]

Thank you guys for the info. If anyone is interested, here's a similarly themed chat thread from photo.net, apparently confirming my suspicion that Red and Canon cameras create ASA/ISO options in different ways.

http://photo.net/bboard/q-and-a-fetch-msg?msg_id=00PTCF

[Keith Walters]

You can't adjust the sensitivity of the RED chip itself, the ASA is in the metadata and everything takes place in post. Interestingly, the alternative method is available in the SI 2K, so that gain can be applied on the camera chip.

I believe that with CMOS sensors like the Mysterium, the Analog-to-digital converter is essentially part of the photosite multiplexer, and so there is no practical way of varying the analog signal level fed into it.

[Guest Glen Alexander]

I believe that with CMOS sensors like the Mysterium, the Analog-to-digital converter is essentially part of the photosite multiplexer, and so there is no practical way of varying the analog signal level fed into it.

 

well if you really wanted to go balls to the wall, reduce the dark current on the chip thermally. bolt on a peltier cooler, liquid nitro works well, ha ha. electrically isolate the chip, no ground loops, no high frequency transients, etc..

[Keith Walters]

well if you really wanted to go balls to the wall, reduce the dark current on the chip thermally. bolt on a peltier cooler, liquid nitro works well, ha ha. electrically isolate the chip, no ground loops, no high frequency transients, etc..

I have been down that path with designing video tap cameras.

 

One overlooked area is phase jitter caused by digital clock squaring circuits, which are in essence, simply high-gain analog amplifiers. By replacing the original CCD drive waveform with phase-locked signals from an LC oscillator working in a strictly linear mode, you can get over a stop improvement in signal to noise, which is vital when you are scratching for light with modern film stocks.

 

However, personally I do not believe that silicon photosensors will ever match the dynamic range performance of modern film stocks. (Or even not so modern ones).

Maybe some other technology will come along to do the jog, but I really doubt silicon is up to the task.

[Brian Drysdale]

I believe that with CMOS sensors like the Mysterium, the Analog-to-digital converter is essentially part of the photosite multiplexer, and so there is no practical way of varying the analog signal level fed into it.

 

On CML, Jason Rodriguez says that the SI-2K does use analog gain as against digital gain and because it's analog rather than digital gain it does actually change the sensors "ASA".

[Guest Glen Alexander]

I have been down that path with designing video tap cameras.

 

One overlooked area is phase jitter caused by digital clock squaring circuits, which are in essence, simply high-gain analog amplifiers. By replacing the original CCD drive waveform with phase-locked signals from an LC oscillator working in a strictly linear mode, you can get over a stop improvement in signal to noise, which is vital when you are scratching for light with modern film stocks.

 

However, personally I do not believe that silicon photosensors will ever match the dynamic range performance of modern film stocks. (Or even not so modern ones).

Maybe some other technology will come along to do the jog, but I really doubt silicon is up to the task.

 

 

eventually someone will come up with some little MEMs device

[Mike Williamson]

I believe that with CMOS sensors like the Mysterium, the Analog-to-digital converter is essentially part of the photosite multiplexer, and so there is no practical way of varying the analog signal level fed into it.

 

Looking at the following diagram of a generic CMOS sensor here (the blue diagram two thirds of the way down the page) you can see that there's a video processing amplifier in line before the A/D conversion. This is apparently where the signal is initially amplified in Canon's still cameras and I would expect the basic chip structure to be the same for the Red camera, as it seems you would have to have an amplifier between the photosites and digital output. But perhaps there is a different design for CMOS chips that I'm not aware, I'm on the front end of this learning curve so I'll defer to those with greater experience.

[Keith Walters]

Looking at the following diagram of a generic CMOS sensor here (the blue diagram two thirds of the way down the page) you can see that there's a video processing amplifier in line before the A/D conversion. This is apparently where the signal is initially amplified in Canon's still cameras and I would expect the basic chip structure to be the same for the Red camera, as it seems you would have to have an amplifier between the photosites and digital output. But perhaps there is a different design for CMOS chips that I'm not aware, I'm on the front end of this learning curve so I'll defer to those with greater experience.

 

You can certainly have a preamplifier between the photosite and the ADC. The hard part is varying its gain in any useful manner. The other problem is that most CMOS imagers have large numbers of separate on-chip ADCs (thousands in some cases), and apart from the problems of implementing gain variation at all, it is even more difficult to make it consistent for all the amplifiers. If there is any variation in the control voltage/gain response curves for individual amplifiers, colour shading errors would result.

 

The big problem here is that it's hard enough to get 12 million photodiodes on a chip all working properly, any additional processing transistor count has to be multiplied by 12 million! The yield of workable chips rapidly drops to zero!

[Mike Simpson]

Actually adjusting the ASA on the red seems to affect the image, but in a pretty counterintuitive way. Adjusting the ASA shifts the dynamic range of the camera and can help preserve highlight detail or reduce noise in blacks.

 

You can check out the red forums for a technical explaination, but the jist of it is you essentially increase the ASA in order to increase highlight detail, but at the cost of added noise in the shadows, and vice versa.

 

So on bright exteriors where you are worried about highlights clipping, you can increase the ASA to, say 500 or 800, and essentially the dynamic range shifts. Lowering the ASA provides helps reduce noise in the shadows, so in low light situations, lower ASAs might actually be more useful.

 

The whole thing seems counter intuitive to me but that seems the be the results of testing over on the red forums.

[John Sprung]

Actually adjusting the ASA on the red seems to affect the image, but in a pretty counterintuitive way. .... you essentially increase the ASA in order to increase highlight detail, but at the cost of added noise in the shadows, and vice versa.....

Yup, you can't get much more counterintuitive than that.

 

 

 

-- J.S.

[Michel Hafner]

The whole thing seems counter intuitive to me but that seems the be the results of testing over on the red forums.

Changing ASA on RED merely changes metadata or how the RAW data is interpreted in post or the monitoring output. It changes nothing about what is recorded in the RAW data unless you fiddle with your exposure as well to do 'the right thing' with the ASA you chose (whatever your idea of 'the right thing' is).

Edited May 22, 2008 by Michel Hafner

[Keith Walters]

Yup, you can't get much more counterintuitive than that.

 

-- J.S.

Well, no. Think about it.

 

Suppose you have the "ASA" on the RED set to 320.

You would also set your light meter to 320 ASA and follow its directions, would you not?

 

Well, if you then set the RED to say, 500 ASA, and do likewise to your light meter, the light meter is obviously going to tell you to reduce the exposure (close down the iris, add NDs, turn down the wicks on your kerosene lanterns etc.)

 

Which is what you would do if you had a highlight clipping problem. QED

 

So yes, if you have highlight clippping, you set the the "ASA" to a higher figure, and (I presume) Redcode will reduce the gain of its digital amplifier. But that will only work if you do the same to your light meter and follow its instructions.

 

Which will be of absolutely no @$#!?!! help whatever if the clipping is happening in the CMOS chip, but I'll leave the Fanboys to work that out.

 

The reality is, assigning an ASA number to a video camera is likely to be a counter productive exercise and will only end in tears. I had started writing a new thread on what an film ASA rating REALLY means (with pictures to prove it) but I just couldn't bothered any more.

 

One of the essential features MY hypothetical version of the RED would have, is a separate video output that shows you exactly what is coming off the chip, BEFORE you do any De-Bayer R3D2 jiggery-pokery. One of the major advantages of shooting on video is that theoretically you don't need a light meter, because your viewfinder and waveform monitor do the same job more accurately and in real time. However, doing this accurately requires a considerable level of skill and technical knowledge. Trying to turn the RED into a Handycam is simply not going to work, no matter how desperately people want it to.

 

But anyway, RED features are coming out to cinemas, so at last the rubber meets the road, and we'll now see how much poop hits the fan :P

[David Mullen ASC]

Exposing RED above 320 ASA, at 500 ASA or 1000 ASA, etc. just means you are underexposing to some degree; the ASA settings are just metadata which you can use when opening up and converting the RAW files.

 

The CMOS sensor itself or the A-D converters, etc. none of these are affected by the inputted ASA rating. There is no analog or digital gain feature in the camera.

 

The ASA rating does affect the signal levels of the live debayering to 720P for monitoring purposes though. But it has no affect on the RED RAW files other than, of course, the rating affects how you expose the image. REDCODE is just the wavelet compression scheme.

 

The RED does have histograms and waveform displays, but they are reading the 720P monitor signal, not the RAW files (not sure if you can read luminance information for RGB for a RAW file anyway without converting it first to RGB.) So if you have selected an ASA rating like 1000 ASA, for example, your onboard exposure displays are reading that boosted signal, so it is not a representation of the RAW recording. Setting the camera to 320 ASA comes the closest to matching the 720P monitoring signal to the RAW recording in terms of exposure values being measured by the camera's histogram and waveform.

[Keith Walters]

Exposing RED above 320 ASA, at 500 ASA or 1000 ASA, etc. just means you are underexposing to some degree; the ASA settings are just metadata which you can use when opening up and converting the RAW files.

 

The CMOS sensor itself or the A-D converters, etc. none of these are affected by the inputted ASA rating. There is no analog or digital gain feature in the camera.

 

The ASA rating does affect the signal levels of the live debayering to 720P for monitoring purposes though. But it has no affect on the RED RAW files other than, of course, the rating affects how you expose the image. REDCODE is just the wavelet compression scheme.

 

The RED does have histograms and waveform displays, but they are reading the 720P monitor signal, not the RAW files (not sure if you can read luminance information for RGB for a RAW file anyway without converting it first to RGB.) So if you have selected an ASA rating like 1000 ASA, for example, your onboard exposure displays are reading that boosted signal, so it is not a representation of the RAW recording. Setting the camera to 320 ASA comes the closest to matching the 720P monitoring signal to the RAW recording in terms of exposure values being measured by the camera's histogram and waveform.

[Keith Walters]

Exposing RED above 320 ASA, at 500 ASA or 1000 ASA, etc. just means you are underexposing to some degree; the ASA settings are just metadata which you can use when opening up and converting the RAW files.

 

The CMOS sensor itself or the A-D converters, etc. none of these are affected by the inputted ASA rating. There is no analog or digital gain feature in the camera.

Which is more or less what I just said.

The point that I think most people have missed is that the only "gain setting" available for the CMOS sensor itself is what the operator does with iris, NDs and so on. All the "ASA setting" does is tell Redcode or RedCine how to interpret what the sensor picks up; there is no way it can extract degraded, but still useable, information from overloaded photosites, the way a film scanner can from an overexposed negative.

 

Which begs the question: "How many of the 12 bits of the ADC actually get used by the system?"

[Guest Glen Alexander]

Which begs the question: "How many of the 12 bits of the ADC actually get used by the system?"

 

it's not only used but real question is how many are effective? it's probably somewhere around 10.5, if they had real engineers build it.

 

reminds me of that really old TV commercial, a kid asks a wise old owl, "How many licks does it take to get to the center of a Tootsie roll?", the owl licks "One", licks again "Two", bites/crushes the candy, "Three". owl, "It takes three". today 12-bits is being outdated, even sony's commercial eng is 14-bit.

[Brian Drysdale]

Applying a higher ASA is similar to closing your aperture to protect your highlights and a lower ASA, opening your iris so there's more information in black end, so you're always reducing the noise levels, rather than lifting up. However, in the end, it does require a judgement call as to where you wish to set your exposure for a consistent look throughout a scene. At low light levels you always have noise/grain issues, it's just how acceptable it is in the context of the film and the story.

 

I've had a couple of video productions, where I wasn't at the grading, when my deliberately dark shots have been lifted, creating poor looking blacks. All the had to do was leave them alone and they'd have been fine - a Gordon Willis moment there.

[John Sprung]

Well, no. Think about it.

Ah, so the setting on the camera isn't like loading a faster film stock, it's like keeping the same stock and just setting the meter to a pretend ASA to shift everything down. That makes sense. It's still counterintuitive that the camera is working like a light meter rather than a camera.... ;-)

 

 

 

-- J.S.

[DJ Joofa]

So yes, if you have highlight clippping, you set the the "ASA" to a higher figure, and (I presume) Redcode will reduce the gain of its digital amplifier. But that will only work if you do the same to your light meter and follow its instructions.

 

I think if you just up the ASA then the digital gain should increase and not decrease. Red's software has no way of knowing if corresponding metering was done to go with higher ASA or not. And, also to go with the notion of a "faster" film stock, higher ASA rating should mean higher gain for a digital system.