Exposure Index Settings both + and -

[tj williams]

Our Digital SLR camera allows a very wide variation of EI(Db) settings from extremely fast, light it only with a candle to very negative.

Of course we all hope the RED will have a good fast EI(Db) setting and enable us to shoot in lower light situations.

The other end of the exposure speed has not been much discussed. When desiring to shoot with the lens wide open for minimum depth of field or near the center of the lens'es stops for best resolution, we need to adjust the speed of the camera downward or put in ND filter packs.

So how much below 0DB can we go??? -30DB? -3DB???????

[jbeale]

It appears to me that no one from Red has been answering tech questions for the past several weeks- waiting for NAB no doubt.

I once had a Sony "prosumer" camera that had a -3 dB setting, but that's the extent of my experience of gain settings lower than normal. What low gain settings do the currently available pro video cameras offer?

[tj williams]

Hi Jbeale

NIkon DH 2 ISO 200 to 1600 (in 1/3, 1/2 or 1EV steps), 3200 and 6400
Nikon D40x Digital SLR Camera ISO Sensitivity: 100 to 1600 in steps of 1 EV
Kodak DCS Pro SLR ISO range 80-800.

These ranges are much beyond the 3 gain settings on a typical broadcast video camera. INMHO it would be enough to eliminate the need to change ND filters in front of the camera. Since the RED does not offer a filter wheel behind the lense As for instance the Sony Cinealta this might allow the control without the necessity of adding ND to the front of the lense during filming. The added advantage would be that in bright lighting the camera could make even cleaner more noise free pictures. A second advantage is that the Dp would be able to more closely select the stop preferred, than the use of ND would allow.

I think you are right about that. Perhaps as they write the handouts they will look at these questions to find the FAQ

[jbeale]

I agree that lower gains can be useful in a variety of situations- I've stacked up several NDs when I was shooting timelapse stuff. Unfortunately I suspect lower gains may not be possible.

I'm not a camera designer myself, but I had the impression that video cameras were generally designed so the "normal" or 0 dB gain position is one that uses the full dynamic range of the CCD/CMOS sensor. Most people want to maximize the available dynamic range. Any lower electronic gain setting than that would not be useful- you would just saturate at "grey" for example instead of "white", because any larger exposure would optically saturate the sensor (a "full well" condition, after which additional photons coming in do not increase the measured sensor output).

I believe this full-well exposure corresponds to the base sensitivity on a DSLR, that is the ISO 200, 100 or 80 settings on the cameras you mentioned. I've seen speculation on this board that the RED "Mysterium" sensor has a native sensitivity equivalent to ISO 320 or more. It seems possible, depending on the design details. If that is the point at which you reach optical saturation, then lowering the electronic gain will not help, unless the Mysterium has an adjustable quantum efficiency (ratio of photons in to electrons stored). That, I believe, would mean a radically new sensor design, different from what is now known.

I suspect we are going to be stuck with using ND filters for the forseeable future.

[Mark Pugh]

I agree that lower gains can be useful in a variety of situations- I've stacked up several NDs when I was shooting timelapse stuff. Unfortunately I suspect lower gains may not be possible.

I'm not a camera designer myself, but I had the impression that video cameras were generally designed so the "normal" or 0 dB gain position is one that uses the full dynamic range of the CCD/CMOS sensor. Most people want to maximize the available dynamic range. Any lower electronic gain setting than that would not be useful- you would just saturate at "grey" for example instead of "white", because any larger exposure would optically saturate the sensor (a "full well" condition, after which additional photons coming in do not increase the measured sensor output).

I believe this full-well exposure corresponds to the base sensitivity on a DSLR, that is the ISO 200, 100 or 80 settings on the cameras you mentioned. I've seen speculation on this board that the RED "Mysterium" sensor has a native sensitivity equivalent to ISO 320 or more. It seems possible, depending on the design details. If that is the point at which you reach optical saturation, then lowering the electronic gain will not help, unless the Mysterium has an adjustable quantum efficiency (ratio of photons in to electrons stored). That, I believe, would mean a radically new sensor design, different from what is now known.

I suspect we are going to be stuck with using ND filters for the forseeable future.
thanks for that

[Stephen Williams]

These ranges are much beyond the 3 gain settings on a typical broadcast video camera. INMHO it would be enough to eliminate the need to change ND filters in front of the camera. Since the RED does not offer a filter wheel behind the lense

Hi TJ,

Most broadcast cameras the gain is assignable for the 3 settings so there is usually a wide range.

Even 100asa is way to fast (F22 on a sunny day) to work without ND filters IMHO.

Stephen

[Gavin Greenwalt]

I believe this full-well exposure corresponds to the base sensitivity on a DSLR, that is the ISO 200, 100 or 80 settings on the cameras you mentioned. I've seen speculation on this board that the RED "Mysterium" sensor has a native sensitivity equivalent to ISO 320 or more. It seems possible, depending on the design details. If that is the point at which you reach optical saturation, then lowering the electronic gain will not help, unless the Mysterium has an adjustable quantum efficiency (ratio of photons in to electrons stored).

Your analysis is dead on, however since CMOS chips have no real sensitivity equivalent, just a maximum point at which noise is intolerable, the Mysterium may very well be a very noise free 80 ISO chip which by default is set to 320 ISO through gain.

I highly doubt that they would place their default setting at the very limit of its capacity I would be more inclined to believe it's at least one IE under its limit.

[jbeale]

Reportedly the Mysterium has 11.3 stops of dynamic range as presently configured. If its base sensitivity is really ISO 80 and the analog amps are set to give ISO 320, that means it is wasting two stops of dynamic range with that gain, and if set to ISO 80, it would really have 13.3 stops of range. That would be an impressive thing to claim! I was assuming if they could do that, they would have done so.

Note that to actually record data throughout that much range, you would need either 14 bit A/D converters, or some nonlinear element (for example a "knee" circuit or log converter) before the A/D.

[Brook Willard]

Remember, 12-bit lin colorspace can only record [correct me if I'm wrong] 12 stops of true latitude. There was a nice, juicy thread about all of this a while back. Much of it went over my head.

[GlennChan]

From my own tests with the Panasonic HDC-20 (similar to the Varicam, 3 CCDs I think):

There is the sensor, and there is the analog-digital convertor. There is a gain control you can manipulate in the camera menus (presumably this affects the analog gain). On the low end, the camera is limited by image noise. Lowering gain will reduce image noise. On the highlight end, the camera's colors start getting wacky... the highlights change in color. So you run into a balance between wacky colors and image noise.

From the images we've seen of Red so far, it does look like the highlights run into the wacky color problem. So if the Red does apply 'negative' gain, you may not have that much range (or the colors will start getting wacky in a way that can't be fixed.)
Side note: The wacky colors are likely due to non-linear response in the sensor; I think all sensors have this problem... the Dalsa does, one of their papers talks about this.

You can take a look at images at:
http://www.cinematography.net/largel-chip.htm
Note that the colors of the black and white chips change throughout the exposure range.

2- A filter wheel would be another way of approaching this problem... possibly something like
No filter
CC filter for 3200K scenes; no ND
CC filter for D50 scenes (i.e. not quite full daylight) + ND
CC filter for D65 scenes + stronger ND (i.e. full daylight)

It might also be neat to have a polarizer filter you can throw in for more ND and/or polarization. (Second filter wheel; I'm not sure how you'd adjust angle.)

[Gavin Greenwalt]

Remember, 12-bit lin colorspace can only record [correct me if I'm wrong] 12 stops of true latitude. There was a nice, juicy thread about all of this a while back. Much of it went over my head.

That doesn't make any sense. The relationship between latitude stops and bit depth is arbitrary.

You could have a 300 stop latitude sensor record to a 2 bit image and still have the "full" dynamic range.

[Ace]

That doesn't make any sense. The relationship between latitude stops and bit depth is arbitrary.

You could have a 300 stop latitude sensor record to a 2 bit image and still have the "full" dynamic range.

??

Please explain.

Are you talking about recording the 300 stops of latitude in the one 2 bit image?

[jbeale]

Hi Gavin,

It is true that the relation between absolute exposure level and the A/D output number is just an arbitrary gain setting. But if you specify that the exposure response is linear then the dynamic range is limited by the A/D. To be precise, if you have a X-bit A/D, then you have only X stops of dynamic range maximum, regardless of X. (In practice a bit less, due to noise in the A/D itself). To encode more than 12 stops with a 12-bit A/D means you must use a nonlinear process before the A/D.

[Brook Willard]

This is why I left my post so open-ended and uncertain. I read something to that effect, but I don't know enough about these things to start declaring facts. I'll search and see if I can find the aforementioned thread.

[Brook Willard]

Here we go: Starting with Jim's post on this page of this thread [clicky] (http://www.reduser.net/forum/showthread.php?t=586&page=2), keep reading to see what I was talking about.

[Adrian T.]

But if you specify that the exposure response is linear then the dynamic range is limited by the A/D. To be precise, if you have a X-bit A/D, then you have only X stops of dynamic range maximum, regardless of X.

Linearity doesn't imply that there's no linear factor involved. The transformation can still be linear if you apply a factor other than 1.0. With a factor of 0.5 you're able to transform 13 stops to 12 bits. Of course the factor needs to be applied before the A/D as an analog device - or simply use an A/D converter with a different input range.

Of course you'll loose accuracy by doing this. And the software using the digital data must know about it or you'll get funny results.

[Anders Holck]

The problem here is that if you lower the gain, you still loose a stop at the low end as this is lowered beneath the noise floor.
You can also not lower the gain beneath the full well output of the photosensors, or you'll start loosing dynamic range by making the AD unsaturated.