Mysterium-X dynamic range...

[Jeff Kilgroe]

Jeff, that info is shown in System Profiler. My MacBook Pro says "Pixel Depth: 32-Bit Color (ARGB8888)"

That is the graphics card output, not the display ability itself. We have already been over this topic in this thread...

Most all laptop screens are 6bits/channel (18bit total RGB) as have been all Macbook / Macbook Pro screens up to current. I have been told the latest 17" LED backlit screens are 8bit, but I'm trying to confirm.

 

[Noah Yuan-Vogel]

Yeah I cant seem to confirm anything either. I have sources that seem to say that at best the upgrades for macbook pro LCDs have come in the form of only TN+ panels (still 6bit) for better viewing angles and LED backlights for lower power consumption, but it is always possible that it doesn't refer to the very latest models. I've also read that all 13.3, 15, and 17 inch LCDs currently manufactured are only TN and TN+ (6bit only), but who knows what information you can trust online or even how current that information is (I havent found much information on the matter that is newer than 4months old for example). Im sure new LCD panels are coming out all the time.

 

[Rob Ruffo]

The thing with display panels is... You can't trust any of the numbers.
1 000 000:1 contrast ratio? Really? You want to tell me that with a straight face?
1 000 000:1 would mean that teh black woudl be absolute, like a piece of ink soaked paper in outer space, and the whites would be so bright they would be harmful to your eyes. Eye glases left in front of this display while it is showing a bright beach scene could start a fire through refraction. Same with bit count. Most displays have very low color counts at lower IRE ranges. That's why the darker area of gradients tend to band. So between 0-10 IRE the color count is the equivalent of maybe 4 bit, even in an LCD panel touted at 12 bit. Only exceptions to this, of course, are CRT, OLED and DLP, which allow for far more linear bit depths.

 

[Graeme Nattress]

Display contrast ratios have always amused me.

Graeme

 

[Noah Yuan-Vogel]

The thing with display panels is... You can't trust any of the numbers.
1 000 000:1 contrast ratio? Really? You want to tell me that with a straight face?
1 000 000:1 would mean that teh black woudl be absolute, like a piece of ink soaked paper in outer space, and the whites would be so bright they would be harmful to your eyes. Eye glases left in front of this display while it is showing a bright beach scene could start a fire through refraction. Same with bit count. Most displays have very low color counts at lower IRE ranges. That's why the darker area of gradients tend to band. So between 0-10 IRE the color count is the equivalent of maybe 4 bit, even in an LCD panel touted at 12 bit. Only exceptions to this, of course, are CRT, OLED and DLP, which allow for far more linear bit depths.

I believe those contrast ratio high numbers on LCDs only show up before the words "dynamic contrast ratio", and if they dont, they are probably lying. The real annoying part is that a lot of TV and monitor manufacturers have stopped posting static contrast ratios altogether and only listing dynamic contrast specs, since if you know what dynamic contrast is you'll probably turn off that function and leave it off as soon as you turn on the display for the first time. And don't even get me started on 60/120/240hz motion interpolation that seems to be all the rage in HDTVs for some reason.

 

[Graeme Nattress]

Static contrast ratios with the ANSI checkerboard pattern are the only ones worth talking about IMHO.

Graeme

 

[David Rasberry]

At least as far as actual field measurements in a typical viewing environment, I haven't measured anything over about 400 to 1 contrast ratio for a flat panel display properly calibrated for REC709.

 

[Steven Caesare]

Static contrast ratios with the ANSI checkerboard pattern are the only ones worth talking about IMHO.

Graeme

I agree those should be published.

However, I'll typically take a display that offers high static CR + dynamic modulation as opposed to the same display without such.

A well done dynamic system can work well for many types of material, and you can (typically) turn it off if you don't like it, and simply be back to the same static CR as the display without.

-sc

 

[JD Holloway]

Really? That's crazy! So there's exponentially more information in the brighter areas?

Weird eh? Thats why we try and ETTR
Its only truely useful in high key scenes where you need detail.
Clouds, "Heaven", Backlit smoke, White sand etc.

I have no idea why its this way (Graeme?) but it would be nice if it were flat/linear or selectable exponentially to highs or lows. I would think the best case scenario would be a Gaussian distribution with long tails out well beyond the DR of the sensor.

This way midtones (where we play most) have lots of meaty data available to them and the extreme ends (where we typically work less) are still well represented by bits and bobs.

Unless Im missing something...

Now if this "data weight" curve was selectable by scene....

 

[Graeme Nattress]

Sensor pixels just count photons. So they generally linearly represent the number of photons. To make a pixel intelligent enough to count them non-linearly is, as far as I understand not being a sensor designer, highly non-trivial and has other "costs" that might outweigh any benefit.

Graeme

 

[joshua csehak]

Wow, I always figured a sensor was analog with a charge of say 1 to 1,000 electron volts (or whatever unit it's measured in), and the ADC would map that value to an 8-bit number. So if a sensor read 1,000 eV, it'd assign it 256. If it read 33.8 eV, it'd assign it maybe 129 (33.8 being a little over half the perceptual brightness of 1,000). But it looks like a sensor is only as granular as its bit depth? IE an 8-bit sensor only has 256 levels of charge? Or equally, the ADC is only able to measure that charge linearly in 256 levels? I gotta say, it does make a lot of sense with the ETTR business.

So in that case, the bit depth of a sensor DOES dictate its dynamic range then, huh?

I also wonder, and I might have to run a test on this, if this is the case, if I took a picture of a gray scene, where the histogram looks like the washington monument, and exposed it all the way to the left, and another with it exposed all the way to the right (being careful to not clip at all in either case), and then matched them in post, there should be quite a difference in quality between the two.

I definitely make a practice of ETTR normally, but that's usually to keep the "negative" as think as possible. Shooting a gray scene, I would've, in the past, simply exposed properly. Now it sounds like it might be best to overexpose it and bring the levels down afterwards.

 

[Graeme Nattress]

Sensors are analogue in the sense that pixels measure light and represent that as an electrical charge. It's the noise in that signal that limits the effective bit depth of the analogue when it gets converted to digital, in that there's usually little benefit having an a to d with less quantization noise than the analogue noise in the signal it's digitizing.

We ETTR not because of higher precision in the a to d at brighter levels, but because in a sensor, the largest source of noise is usually read noise and that dominates at lower light levels. ETTR and you minimize the impact of read noise.

Graeme

 

[joshua csehak]

So noise is the real limiting factor, not bit rates? Should we care about a sensor's bit rate, then, or just how it handles noise?

I guess at this point it's just numbers, and the real thing is how the picture looks, like how an F23 looks a gajillion times better than an ex-1, even though they both record at 1080p.

 

[Jeff Kilgroe]

It's a noise issue. Sensors don't have bit rates, they're analog. Bits are not introduced until the analog to digital conversion takes place. At that point it's a question of the quality of conversion and how well the analog information is sampled and quantized into digital information. Bits come into play in the form of how precisely encoded the information is. 12 bits represents 4096 steps of precision... As for bit rates, that is a measure of bandwidth or transmission speed. Which is really more a concern for shuttling this data around within the camera, the compression algorithms and writing our data to storage media. As an example, 16 bits of data requires 33% more headroom than 12 bits of data.

As said here and in other threads, we expose to the right to avoid noise in the lower regions of our exposure. Pushing too much of the exposure to the right or into the higher values sacrifices dynamic range. So with the RED One we try to shoot a thicker negative, and light accordingly to accommodate this if we can. With Mysterium-X the benefits for exposure are gained mostly on the low end and better overall signal to noise ratio, so we can shoot a full histogram without as much concern for noise in the shadows.

 

[Hrvoje Simic]

In pro audio world multi-band compression and limiters are applied to audio signal to optimize usage of dynamic range. Can similar solution be applied to sensor's analog video signal - prior to A/D, to control the highlight range ?

http://www.tcelectronic.com/c400xl.asp

 

[JD Holloway]

@ Jeff and Graeme

Any reason why bit allocation isn't linear or other?
While low regions have less SNR (given), and higher bit allocation could be used to record it (more precise digitization of crap), is there any reason not to have a linear bit allocation distribution? I cant think of any...

 

[Daniel Browning]

Can similar solution be applied to sensor's analog video signal - prior to A/D, to control the highlight range?

Yes, there are many designs (similar and otherwise) that have been proposed for increasing dynamic range, but so far only the simplest ones have implemented commercially in these types of markets. As Graeme said, I'm sure there are tradeoffs involved. I think we'll see more of it in the future as they find ways to reduce the downsides. (For example, some of the HDR pixel designs reduce performance in low light.)

 

[Pawel Achtel]

Interesting.

OK, so, how about a simple solution to a complex problem:

Make two S35 (or/and F35) brains. One for low light applications and one for sunlight. The low light sensor would have large pixels and smaller resolution, intended for extremely low light levels: just as our vision is. And another, HDR type sensor designed to give us nice deep knee in the highlights. The low light sensor could be also used for super high speeds (say up to 1000 fps) and, because of its lower resolution, the data rates would be still managable.

It would make ND filtering easier too, hence better contrast on the high resolution sensor too.

Sooo, how about a 2.5k low-light/high-speed sensor (brain) and a 5k/6k sunlight brain? Sounds like simple, non-compromise and relatively economical solution. Am I missing something?

And, the best thing is that such solution would fully leverage the modularity of Epic/Scarlet. Arri Alexa or F35 can't do that as they are locked in one body!

Let's get some milage out of that modularity, guys! Jim? Graeme?

 

[jbeale]

Sooo, how about a 2.5k low-light/high-speed sensor (brain) and a 5k/6k sunlight brain? Sounds like simple, non-compromise and relatively economical solution. Am I missing something?

It may be that simply downsampling the image from the existing sensor would give you similar noise characteristics to a purpose-built sensor of lower resolution but otherwise the same technology.

Put another way, if the sensor readout noise is low compared with the photon shot noise, then there would be no point to making a custom low-light sensor with bigger pixels- you'd get the same image by just downsampling in post.

 

[Deanan]

Yes, there are many designs (similar and otherwise) that have been proposed for increasing dynamic range, but so far only the simplest ones have implemented commercially in these types of markets. As Graeme said, I'm sure there are tradeoffs involved. I think we'll see more of it in the future as they find ways to reduce the downsides. (For example, some of the HDR pixel designs reduce performance in low light.)

And others have kinks where the crossover points are between A/Ds.