What are you exited about (DSMC)?

[Daniel Browning]

Why does a larger sensor with equally large sensor sites demonstrate more DR than a smaller sensor?

Because is has more light from more area and less noise from scaling pixels for spatial frequency.

Compare RED ONE at 2K vs RED ONE at 4K. You should be able to find others on the forum here that will tell you the same thing, but here's the math:

When both are at the same field of view and T/stop: for every pixel in that gets 50,000 photons in 2K mode, the 4K has four pixels. When the 4K is resampled back down to 2K, it results in 200,000 photons, because signal adds linearly.

If the read noise of each 2K pixel is 11 electrons, then the four pixels in 4K mode will have 44 electrons read noise... *but* read noise adds in quadrature, so 4K mode actually results in only 22 electrons [sqrt(11^2 * (4K/2K)^2)] in the final 2K output.

So RED ONE in 4K mode has one stop more dynamic range than RED ONE in 2K mode.

I would think the reduced heat, cross talk etc etc etc would all contribute to a considerably cleaner signal on the smaller sensor with equal sized sensor wells.

Read noise is an important factor, and perhaps that's what you meant, but thermal noise (heat) is not a factor at room temperature and normal shutter speeds (1/48); electronic cross talk, too, is not a factor at 3+ µm pixel sizes, for any size sensor.

 

[Gavin Greenwalt]

I get it. So think of a group of 4 pixels as essentially one large pixel with increased DR.

Photons are photons so the combined area of those 4 pixels can be thought of as a 2x larger pixel.

 

[Roberto Lequeux]

Why does read noise not add linearly?

 

[Gavin Greenwalt]

This does add one level of confusion then.

Does that mean then that if you shoot cropped-s35 FF35-Monstro that you're back at MysteriumX DR? Even though Monstro has the 1+ stops of DR? Even though it has the lower density sensor?

In the case of the REDOne you're comparing 4k s35 to 2k s16. What about 4ks35 vs 2ks35. You're saying in such a case it would still be a wash? I guess I'm with Roberto. Why doesn't read noise add linearly? *puts on student cap*

 

[Daniel Browning]

Why does read noise not add linearly?

Because random noise follows a Gaussian distribution.

It's kind of like a guessing game: pick a number between 1 and 100 and tell me if I'm too high or too low. If you only have one guess (one pixel), you will probably be way off (noise). But if you have four guesses (4 pixels), you can get closer (less noise).

Here's another explanation. Let's say you're looking at a noisy part of the image in photoshop. Let's say that you happen to know the true luminance (brightness) is a value of "30" in photoshop. But because of the noise, the actual values are all over the place: 30, 31, 30, 29, etc. They are all centered around the mean (30) in a Gaussian distribution.

When you take four of those pixels, and add them together, and divide by four ((30+31+30+29)/4), you get closer to the true value, 30. The ones that were under 30 cancel out the ones that were over 30.

Does that mean then that if you shoot cropped-s35 FF35-Monstro that you're back at MysteriumX DR? Even though Monstro has the 1+ stops of DR? Even though it has the lower density sensor?

No; Monstro is newer and superior technology, so even with the same area it will have more dynamic range. (And when you're not cropping, you have the area advantage on top of the technology advantage.)

In the case of the REDOne you're comparing 4k s35 to 2k s16. What about 4ks35 vs 2ks35. You're saying in such a case it would still be a wash?

Yes, all other things being equal (AOTBE, i.e. same technology level, T/stop, etc.). For example, if EPIC S35 and Scarlet 2/3" have the same Mysterium-X technology, you could crop EPIC S35 down to the same size as Scarlet 2/3" (10.1mmx5.35mm) and noise would be the same. (Scarlet 2/3" would have higher resolution, though.)

[Personally, I hope that Scarlet 2/3" will have even less noise than a cropped EPIC.]

There is another factor which gives higher resolution an advantage: the OLPF. Native 2K must have an anti-alias filter to fight aliasing artifacts, so when 4K is resampled to 2K it would have noticeably higher detail and contrast. If both were downsampled to 720p, the difference is no longer important.

 

[Roberto Lequeux]

*puts on student cap*

Welcome to the classroom man :thumbsup:

It's kind of like a guessing game: pick a number between 1 and 100 and tell me if I'm too high or too low. If you only have one guess (one pixel), you will probably be way off (noise). But if you have four guesses (4 pixels), you can get closer (less noise).

Here's another explanation. Let's say you're looking at a noisy part of the image in photoshop. Let's say that you happen to know the true luminance (brightness) is a value of "30" in photoshop. But because of the noise, the actual values are all over the place: 30, 31, 30, 29, etc. They are all centered around the mean (30) in a Gaussian distribution.

When you take four of those pixels, and add them together, and divide by four ((30+31+30+29)/4), you get closer to the true value, 30. The ones that were under 30 cancel out the ones that were over 30.

I totally get this. I didn't think you were taking about the downsampling and thought you were talking about something else. I am sure Gavin is with me on this too. I guess your formula [sqrt(11^2 * (4K/2K)^2)] made me think you were talking about something else. This I understand [(30+31+30+29)/4]. :thumbsup:

 

[Raul Gonzo]

I think that some people are getting their hopes up too high regarding dynamic range.

Red has posted Monstro being 13+ stops for DR. Do you guys think that is too optimistic?

The extra DR is the main reason I have decided to get a Monstro sensor vs. the m-X. :undecided:

 

[Tom Lowe]

Red has posted Monstro being 13+ stops for DR. Do you guys think that is too optimistic?

The extra DR is the main reason I have decided to get a Monstro sensor vs. the m-X. :undecided:

Yes, I think it's too optimistic. The only way Red can get that kind of leap in DR would be through some type of real new breakthrough. Maybe some type of limited HDR taking place in-camera or whatever? If it was easy to do, others would be increasing their DR as well.

Then again, Jim has pulled many rabbits out of his hat already, so one can never say never.

 

[Daniel Browning]

Yes, I think it's too optimistic. The only way Red can get that kind of leap in DR would be through some type of real new breakthrough.

Agreed. It would pretty incredible.

It's possible that the 1 stop of dynamic range improvement is exactly due to the greater sensor area (just as RED ONE 4K has 1 stop more than RED ONE 2K). That is, maybe the 1 stop improvement goes away when S35 is cropped from FF35.

 

[David Mullen ASC]

It sounds like they are saying that Monstro will have two stops more dynamic range than Mysterium currently has -- since they claim 11+ stops for Mysterium, hence 13+ stops for Monstro.

It certainly is possible therefore... it's just that some people say that RED is counting stops of detail visible on charts at the very bottom of the noise floor -- some people would not count the last two stops of info in the blacks are being very usable, hence why there is some difference of opinion about how many stops of dynamic range you have to work with in the RED, hence why it becomes a political debate.

But it's safe to say that claims of increasing dynamic range by two stops over Mysterium are possible.

 

[Raul Gonzo]

hmmm- you guys have totally open my eyes to this. I may indeed end up getting the s35 after all.
Overall, I will just have to wait and see. Thankfully, my red will keep me happy for awhile. This time, waiting isn't as bad...

 

[Tom Lowe]

Two stops over Mysterium is possible, but don't most people peg Red One at 9 stops, maybe 10? That would be 11-12.

*Prepares to get his ass chewed out by Jannard* :couch:

 

[Gavin Greenwalt]

Well... the f23 has what 1+ stop more DR than REDOne? And its sensor size is 3/4 that of the REDOne. (~16mm x 3)

So I would say an additional 2 stops should be attainable while still remaining within the realm of 'common'. I would say it's also similar to a modern FF35 DSLR sensor.

 

[Daniel Browning]

I would say it's also similar to a modern FF35 DSLR sensor.

Agreed. The D3X has pretty amazing dynamic range: 12.8 stops per pixel using 1:1 S/N, and 15 stops at 720p resolution. The D3 is only 11.8 stops per pixel and 13.3 stops at 720p. Incidentally, the Monstro has the same pixel size as the D3X.

The design requirements (60 FPS vs. 5 FPS, for one) are stacked against RED, so it would be quite a feat to match a still camera. Or for a still camera to match RED's FPS without losing the dynamic range.

 

[Roberto Lequeux]

David, I am sure everyone here would love to hear what it is that you are most exited about for DSMC. Would you care to share a short answer with us?