Pixel fetish

[Mike Brent]

I have a question... what would you prefer:

1) a 6k camera over a 4.5k camera.

2) a 17 stops camera over a 12 stops camera.

If you can choose just one improvement, which one would you take?

 

[Vinit Borrison]

17 stops!

 

[Rick Burnett]

WITHOUT question I'd want more stops.

 

[Jonathan Stevenson]

This poll feels a little loaded, haha. "Would you rather have a super fast car, or a car that runs on a cup of water for a month." Maybe a better one would be "would you rather have a 13-stop, 6k camera or a 17 stop, 3k camera." I have a feeling most would still choose the 17 stop camera.

 

[Rick Burnett]

This poll feels a little loaded, haha. "Would you rather have a super fast car, or a car that runs on a cup of water for a month." Maybe a better one would be "would you rather have a 13-stop, 6k camera or a 17 stop, 3k camera." I have a feeling most would still choose the 17 stop camera.

No, I choose the car that runs on a cup of water for a month!

 

[Jonathan Stevenson]

No, I choose the car that runs on a cup of water for a month!

Exactly, that was my point. Pretty much everyone would!

 

[Mike Brent]

15 to 0...

OK, and here is the next question - if all the effort/time/money put into the Epic and Scarlet programs would go into improving dynamic range - would it be possible for Red to create a 17 stops Red One?

 

[Julio Quintana]

15 to 0...

OK, and here is the next question - if all the effort/time/money put into the Epic and Scarlet programs would go into improving dynamic range - would it be possible for Red to create a 17 stops Red One?

I think you incorrectly assume that focusing on resolution detracts from focusing on dynamic range. I think RED's strategy has been to increase resolution which leads to a cleaner image, allowing us to dig deeper into the shadows, effectively increasing the dynamic range.

 

[Mike Brent]

I think you incorrectly assume that focusing on resolution detracts from focusing on dynamic range. I think RED's strategy has been to increase resolution which leads to a cleaner image, allowing us to dig deeper into the shadows, effectively increasing the dynamic range.

How much the increased resolution expands the dynamic range? An extra stop?

 

[Rick Burnett]

How much the increased resolution expands the dynamic range? An extra stop?

A cleaner image comes from better pixel design, not smaller pixels. Pushing a pixel extremely too small will eventually get to the point the support circuity is going to be larger than the active area and at that point, things are going to get worse.

I feel RED has been pushing both boundaries to get that sweet spot where resolution AND dynamic range of the sensor are at their best.

 

[Mike Brent]

I don't know... all I see is talk about 4k 5k 6k 28k, 4k not enough etc etc. Imagine all that energy going into increasing dynamic range, that's all I'm saying. And the poll is pretty clear about it (26 to zero)

 

[David Rasberry]

Seems to me that Red is being pretty successful at both development goals, resolution and dynamic range. As much as good design, dynamic range is dependent on the current state of chip fabrication science. Dynamic range can't exceed the underlying signal to noise limits of the fab processes. That is a hard physical limitation.
There was a lengthy thread sometime back about the myths of small pixels vs big pixels. Essentially for the range of pixel sizes Red is using between about 3 microns and 7 microns, size alone has little impact on sensitivity and dynamic range.
One of the major trends at Infocomm this year was a move to much higher than HD resolution signage displays. Seamless displays of 10 megapixels and up were abundant on the show floor intended for medical imaging, digital signage, and event staging markets and viewing at walk up close distances.
There was one demo of a quad HD format 50" LCD display showing 4k video of a medical procedure that made a believer of me, even on this modest sized display.
RED has done their market research well and as far as I can see are leading the way in both resolution and dynamic range for practical reasonable cost production cameras.

 

[Rick Burnett]

Seems to me that Red is being pretty successful at both development goals, resolution and dynamic range. As much as good design, dynamic range is dependent on the current state of chip fabrication science. Dynamic range can't exceed the underlying signal to noise limits of the fab processes. That is a hard physical limitation.
There was a lengthy thread sometime back about the myths of small pixels vs big pixels. Essentially for the range of pixel sizes Red is using between about 3 microns and 7 microns, size alone has little impact on sensitivity and dynamic range.
One of the major trends at Infocomm this year was a move to much higher than HD resolution signage displays. Seamless displays of 10 megapixels and up were abundant on the show floor intended for medical imaging, digital signage, and event staging markets and viewing at walk up close distances.
There was one demo of a quad HD format 50" LCD display showing 4k video of a medical procedure that made a believer of me, even on this modest sized display.
RED has done their market research well and as far as I can see are leading the way in both resolution and dynamic range for practical reasonable cost production cameras.

It is not a hard physical limitation, I work in CMOS every day, I can tell you it is more complicated than that. It's not so much the size of the area, but the size of the devices in that area, and the size of the interconnect between not only the active/non-active area, but all the analog circuitry as well. Noise is created from heat, noise is created from wires at different potentials near each other that are changing value, and differences in pixel output can come from actual differences in the pixel fab layers since they are not consistent across the face of the sensor either.

The issue though is that it's not quantifiable to just the size of the pixel. There are plenty of CMOS designs with ALL kinds of pixel pitches that operated all over the spectrum in terms of sensitivity. Since everyone uses their own special techniques, a comparison of pixel performance really depends on what changes between the large pixel and the small pixel. If you have to shrink the area of the storage circuitry, you aren't going to be able to store as large a value. But that's not to say you do that.

Some designs use multiple active circuits at each pixel to deal with differences in the fab process so they can basically get the 'best' guess at the recorded value)

Jim already said the Scarlet 2/3 will not have the exact same DR as the Epic, it will be slightly less, and the difference between the two I can see is the pixel size.

That said, both the Epic and the Alexa have shown by some people to have 13.5 stops using two different pixel pitches AND two different techniques. I believe the Alexa uses the dual-gain path. Of course, once these are both out in the water they can be vetted further.

For me larger resolution is about getting rid of aliasing/moire down at your target resolution. The OLPF is not perfect with a bayer pattern given the density of green is twice that of both blue and red. Recording at a higher resolution to bring down into my target resolution is going to bring me an absolutely much sharper picture.

 

[David Rasberry]

There is a hard physical limitation at the fab process level for all kinds of sensors. Apart from the variables introduced by design, new fab processes introduced in the last 3 years have increased the maximum signal to noise potential by at least 10dB by increasing thermal efficiency and lowering power consumption. This is showing up in all kinds of silicon based chips from industrial temperature sensors to audio DSP's to computer processors as well as imaging sensors.
There are immediate market demands for higher than HD program delivery in commercial markets that will push trends faster than anything happening in the entertainment industry.
Show me any other practical production camera system at any price that can deliver today a native motion JPEG program at 2560x1620 pixel resolution or higher for these marketing and staging applications. RED 1 is still unique in this capability and Epic and Scarlet will raise the bar even higher. At 13.5 stops DR, R1-MX comfortably exceeds the dynamic range of any display system out there.

 

[Mike Brent]

I understand the technical limitation of more resolution but whats the technical limitation for better dynamic range? Where's the "problem"?

31 to ZERO - people want more dynamic range!!!

 

[Mike Brent]

comfortably exceeds the dynamic range of any display system out there.

What? Do displays also have dynamic range limitations?

 

[Rick Burnett]

There is a hard physical limitation at the fab process level for all kinds of sensors. Apart from the variables introduced by design, new fab processes introduced in the last 3 years have increased the maximum signal to noise potential by at least 10dB by increasing thermal efficiency and lowering power consumption. This is showing up in all kinds of silicon based chips from industrial temperature sensors to audio DSP's to computer processors as well as imaging sensors.
There are immediate market demands for higher than HD program delivery in commercial markets that will push trends faster than anything happening in the entertainment industry.
Show me any other practical production camera system at any price that can deliver today a native motion JPEG program at 2560x1620 pixel resolution or higher for these marketing and staging applications. RED 1 is still unique in this capability and Epic and Scarlet will raise the bar even higher. At 13.5 stops DR, R1-MX comfortably exceeds the dynamic range of any display system out there.

No, again, you are wrong, there is not a hard physical limitation. I'm telling you I work in silicon design EVERY day. There is a limitation, but it is caused by MANY different variables and also by the design characteristics by which even lowering power can cause more problems with noise.

Where do you get thermal efficiency has gone up? I've been doing chip design for 15 years, they are still using many of the same materials, many of the same packaging, and the characterization for chip design has remained pretty consistent over those 15 years, 125C to -40C for almost all applications. Wires are getting smaller and even though voltages drop, the current needed through those wires goes up. Smaller transistors leak A LOT more power which is a constant source of additional heat now. Some smaller technologies are so bad, power shut-off has to be employed so that NO power gets to the devices when idle as the power drain is so high. These smaller nodes are causing a lot more problems in heat/voltage drop than ever before.

It is true that we target lower power, but many times that comes at the sacrifice of slower speeds, and with slower speeds comes more noise, so we constantly battle that.

And where are these great significant changes in audio technology? I've been buying professional audio equipment for at least 7 years. I've not seen significant SNR gains in either my Apogee or Digidesign products. Most of the difference between audio gear comes down to the preamp and the accuracy of the A/D converter. These are not dependent on the process node, but on the circuitry design.

And where are you going with the question about any other device capable of doing what Red has done? I'm not comparing the M or MX sensor resolution, I'm just saying that there is at least one other camera with larger pixel size with the same dynamic range. The point is, a smaller pixel does not yield better DR response, and in fact, at some point of reduction, can degrade the SNR ratio as the circuitry in the design gets pushed closer together.

Even further, saying that 13.5 stops exceeds the dynamic range of any display system is NOT the point. The point is having the largest dynamic range for acquisition so that during scenes of extreme dynamic range, like outside in the sun pointing the camera to area with both full light and shade, you retain as much information as possible so that in post, you can design what you want to use instead of having to make that decision on set.

Is 13.5 enough? Until *I* use it and test it, I don't really know. But it doesn't matter, that is what we have and that is what we have to use. Additionally, for most cameras, it is A LOT more than what we've had.

The question was what do people want more moving forward, dynamic range or resolution. I think it is pretty clear people would rather have more dynamic range. I also contend from my professional opinion that at some point of pushing both those characteristics, if your target base wants one more than the other, there comes a point where you can pick one over the other when a sacrifice in performance has to be decided and this WILL happen if the pixels get smaller. The drive circuitry is not going to shrink because the sensor area is the same.

 

[David Rasberry]

The gist of the argument of large vs small pixels within the size range I indicated above is that for the similar design characteristics you would be trading off analog noise floor for read noise, i.e. quantization error. Near the analog noise floor margin where a single 6 micron sensor would have a 50% quantization error compared to it neighbors, a 3 micron sensor would give you 4 samples for the same physical area which properly interpolated by processing algorithms could result in lower overall image noise.
In actual camera tests the sensitivity and dynamic range scaled more predictably with total sensor area than with pixel size.
I don't expect 2/3" Scarlet to equal S35, but there have been a couple of posted shots of the same scene shot at 4k and 2k crop on MX and the sensitivity was essentially the same with a slight increase in apparent noise/grain. Much like the difference between 35mm and 16mm film stocks with the same ISO ratings. Not a huge difference.
I will point out that Sony's new 2/3" PDW-800 Cine Alta model has been tested at 800 iso with 13-14 stops DR.
I agree more DR is a very desirable characteristic for sure, but in view of emerging production markets the balance between resolution and DR needs to be carefully considered and one does not necessarily have to trade off resolution in order to increase DR, nor is decreasing resolution likely to yield huge gains in DR for a given size sensor.

 

[Radim Schreiber]

At the moment I would prefer higher dynamic range and lower noise levels.

 

[Mike Brent]

At the moment I would prefer higher dynamic range and lower noise levels.

I will add that I don't mind my camera not being a miniature the size of a cigarette box. All this effort to make everything smaller and smaller, and then you need to add weights just because it's TOO small and you can't hand hold it because it's too light... isn't that a little.... whats the word?

So you pay $2500 for a very advanced small camera and then you need to pay $400 for a balancing system, essentially paying for weights, because the camera is so small and light that you can't use it alone because it will shake like mad.