CCD Chips and different ASA values

[Christian]

Hi,

This might be the wrong forum to post a question like this but I didn't know where to put it else.

What i want to know is how a ccd or a digital camera works in terms of ASA/ISO. For example the Arri D20 has a fixes ISO of 300 in its TechSpecs, but in the menu of the camera you can set different values from 50 to 400. So is it in the CCD/CMOS Chip or later in some electronical parts or the A/D converter where the ASA/ISO comes from? I searched the web but couldn't find something easy to understand or complete.
help?

Greetz
Chris

[Gavin Greenwalt]

To my knowledge it usually happens before the ADC as a simple function of gain.

[Christian]

but after ccd?

[Billy Summers]

What will the Redcam, most likely, be rated at (ISO/speedwise)?

[Stephen Williams]

What will the Redcam, most likely, be rated at (ISO/speedwise)?

Hi,

From David Stumps tests I would say 320, but things may change.

Stephen

[Gavin Greenwalt]

but after ccd?

Uhhh not to sound like an ass but... how would it be before?

[Graeme Nattress]

Traditionally, amplification of the signal is applied both both in the analogue and digital domain.

ISO - doesn't really have one. You can figure an equivalent though. You've got 11.3 stops to work with, so you can place your middle grey anywhere in there. Rate ISO higher, and middle grey will go lower, giving you more highlight headroom, or the other way around.

Graeme

[Policar]

Traditionally, amplification of the signal is applied both both in the analogue and digital domain.

ISO - doesn't really have one. You can figure an equivalent though. You've got 11.3 stops to work with, so you can place your middle grey anywhere in there. Rate ISO higher, and middle grey will go lower, giving you more highlight headroom, or the other way around.

Graeme

This seems to contradict what you wrote earlier about exposing right (which makes a lot of sense, like exposing 500ISO film as 320ISO to get a denser negative.)

For instance, let's say I want to expose at 800ISO, despite having a 200ISO sensor. Assume there will be no blown highlights; we're pushing two stops because it's dark. If gain is applied at the analog level (to the sensor directly), I might lose a stop of dynamic range in the noisy shadows and I would have more noise throughout. I would, however, have a full 11 stops (minus noisy darks) of data to play with, and it would be spread evenly throughout the 12-bit image.

Let's say I did this digitally instead and assume a roughly flat characteristic curve. Almost all my image data is stuck in the brightest brights--which don't even exist, as per the theory of exposing right. So when I boost my image digitally, grain is more noticeable, and posterization occurs and I get an 8 bit image or something instead of a twelve bit one, and compression artifacts, etc. are more noticeable, too.

To me, it seems like analog gain is similar to using a higher ISO of film (blocky grain structure but pleasing, full image), whereas digital gain is like using slow film and pushing it (fine grain structure, but more grain, and a much less pleasing, higher contrast image).

When shooting Collateral, I think Dion Beebe always "exposed right," even if it meant using a full +12db of gain or whatever so that the image had the maximum data that could then be "pulled" down in post to darker image, if necessary.

Doesn't it make more sense to apply gain at the level of the sensor rather than in the DSP? Maybe I'm missing something but it seems rather foolish to make gain a digital function or, worse yet, to make the camera only shoot at one ISO speed and make gain something you add in post. (The accuracy of the viewfinder and histogram would be comprimised by this, too, making shooting harder.)

Anyhow, doesn't effect me since I'll never be able to afford one of these things, but since I may some day use one, I'm going to ask for gain at the level of the sensor.

[Christian]

Uhhh not to sound like an ass but... how would it be before?

:p

well I do know that the output of the CCD is amplified when an higher ASA Rating is applied, but does that mean, that something changes on the CCD so that it puts out more information OR is it always the same amount of output, which later is amplified.

(I would think of the last one like zooming into an JPEG Image, it has a certain resolution, you can zoom in and get a closer look with probably more details, but lack more and more image quality.).

[Graeme Nattress]

The pixels don't change. What changes is the analogue amplification of the voltage that pixel produces. This does not add more information. You're noise limited, and amplifying the voltage will also amplify the noise.

Policar - I don't see a contradiction, because I'd not need to know the assumed ISO to expose correctly, I'd use the histogram and ETTR. The ISO is only useful to someone used to working with film, exposing using light meters and the like.

So your question becomes, where do you apply gain?

Assume that sensor noise is < a-to-d quantisation noise, ie the 12bits of data are totally adequate to represent the analogue voltage generated by the pixels.

So, in idea a) we amplify the voltage. Does this alter SNR? It shouldn't, but being analogue gain, it might decrease it. Assuming we don't introduce any clipping, our 12bits are therefore still adequate.

So we could gain via b) which is digital gain. You could record the sensor RAW, and do so in post, where if you were to clip data, you'd know about it, and be able to do something about it, or ignore it, or whatever. You've got control here. You'd be working in float, so no precision loss either. You could do digital gain before recording. As the digital gain is just math, you're not loosing anything by doing this, but nor are you gaining any detail or whatever.

SO, I don't think it really matters whether you do a) or b) but I'd do b) as I could then deal with any possible clipping better.

With a more traditional camera and where the recording bit depth is less than the a-to-d bit depth, there could be logic in expanding the data out to fit all the code values. Doing so in the analogue domain would fill up all the code values, but that data in the gaps would be totally invented - ie be noise. So yes, it's going to look different, but I don't think it has, per se, any more actual information content.

Graeme