Graeme Nattress
RED Problem Solver
At that point you're trading dynamic range for resolution, and introducing a much more complex demosaic algorithm.
Graeme
Graeme
Welcome to our community
Be a part of something great, join today!
Large chip 3-chip designs?
- Thread starter Robert Jackson
- Start date
- Banned
- #22
I don't think you can make it easily work with a mirror as you show on the picture. At least I never saw such a setup. Maybe one reason for this is that the light coming out on the other side of the mirror should be quite perpendicular to the surface of the mirror glass. And the front should be perdendicular as well to make it symmetrical .So after all you have to use a gigantic cube shaped prism with a beam splitter surface inside similar to a 3CCD beam splitter. And now again you get all the disadvantages of the big prism.
Ace
Wedge In
- Joined
- Dec 28, 2006
- Messages
- 1,493
- Reaction score
- 0
- Points
- 0
- Age
- 42
- Location
- Sydney, Australia
- Website
- www.dlshs.net
You could have a 45 degree mounted mirror that rotates at a few thousand rpm, reflecting to each sensor as it spins a revoloution.. the speed of the mirror RPM would = mechanical shutter speed.. thats how i would do it anyway.
Andrew M.
Well-known member
Rotating mirror is very good solution but it adds to the complexity and cost.
Bit more of brain storming here and maybe we will come up with some good simple solution.
As to the alignment (just thinking loud here) what about pixel shifting.
I couldn’t believe that I said that.
Seriously though, if we shift second CMOS Bayer pattern sensor by one pixel down and one right, relative to the first sensor we are getting double the resolution. Are we getting ½ of luminosity doing this?
And then the absence of green in the picture will not influence the outcome as bad as usually does for Bayer pattern sensors.
We have to balance between getting 16 bit dynamic range and higher resolution.
Whatever we loose in terms of luma we gain it back by combaining this two sensors data in to the one from bit 5 to 12 (not from 1 to 4 and 13 to 16)
Even if we get 8K virtual sensor resolution out of this pixel shifting I will downres it down to 4K on the output sensor anyway.
The first 4 bits (1-4) of low light will not be coming from the second (high luma sensor) so no extra resolution here, also last 4 bits (13-16) of high luma will not come from the first sensor so no extra resolution on highlights.
Do we need extra resolution at very low and very high luma levels?
Though from bit 5 to 12 we will get double kick in terms of resolution.
16 bit dynamic range is the only thing that is inferior on the CMOS, as compare to the film now.
We beat resolution of the film.
We beat the focus sharpness capability over the film.
Now if we bit the dynamic range then film is dead.
Just to simplify the discussion.
FIRST SENSOR is the low light sensor, this is the one servicing bits 1 – 12 and it is the one plugged to the second sensor 12 bit input port.
SECOND SENSOR is the high light sensor servicing bits 5 – 16 and it is the one that has 12 bit input port to take data from the first sensor and it has 16 bit output port.
As to the connection between these two sensors, we do not want to have more of data processing added to this sensors, especially to the second one that have to do some bits ORs or ANDs on the data. We can connect these two sensors not on the final bit data bus coming out of them but just straight on the row and columns of the CMOS photosensors, so the second sensor will have ability to address the first sensors photo cells like its own.
Just a thought that may simplify the design.
On the other hand, splitting processing power between these two sensors may be desired, so pre massaging the data on the first sensor before sending it to the second one could be very beneficial.
This reminds me the human eye construction.
The picture data received by our eye is pre processed on the retina before it is sent to the brain.
http://en.wikipedia.org/wiki/Retina
Can we learn anything from the nature here?
But let’s focus on the mirror design. Any other ideas?
Bit more of brain storming here and maybe we will come up with some good simple solution.
As to the alignment (just thinking loud here) what about pixel shifting.
I couldn’t believe that I said that.
Seriously though, if we shift second CMOS Bayer pattern sensor by one pixel down and one right, relative to the first sensor we are getting double the resolution. Are we getting ½ of luminosity doing this?
And then the absence of green in the picture will not influence the outcome as bad as usually does for Bayer pattern sensors.
We have to balance between getting 16 bit dynamic range and higher resolution.
Whatever we loose in terms of luma we gain it back by combaining this two sensors data in to the one from bit 5 to 12 (not from 1 to 4 and 13 to 16)
Even if we get 8K virtual sensor resolution out of this pixel shifting I will downres it down to 4K on the output sensor anyway.
The first 4 bits (1-4) of low light will not be coming from the second (high luma sensor) so no extra resolution here, also last 4 bits (13-16) of high luma will not come from the first sensor so no extra resolution on highlights.
Do we need extra resolution at very low and very high luma levels?
Though from bit 5 to 12 we will get double kick in terms of resolution.
16 bit dynamic range is the only thing that is inferior on the CMOS, as compare to the film now.
We beat resolution of the film.
We beat the focus sharpness capability over the film.
Now if we bit the dynamic range then film is dead.
Just to simplify the discussion.
FIRST SENSOR is the low light sensor, this is the one servicing bits 1 – 12 and it is the one plugged to the second sensor 12 bit input port.
SECOND SENSOR is the high light sensor servicing bits 5 – 16 and it is the one that has 12 bit input port to take data from the first sensor and it has 16 bit output port.
As to the connection between these two sensors, we do not want to have more of data processing added to this sensors, especially to the second one that have to do some bits ORs or ANDs on the data. We can connect these two sensors not on the final bit data bus coming out of them but just straight on the row and columns of the CMOS photosensors, so the second sensor will have ability to address the first sensors photo cells like its own.
Just a thought that may simplify the design.
On the other hand, splitting processing power between these two sensors may be desired, so pre massaging the data on the first sensor before sending it to the second one could be very beneficial.
This reminds me the human eye construction.
The picture data received by our eye is pre processed on the retina before it is sent to the brain.
http://en.wikipedia.org/wiki/Retina
Can we learn anything from the nature here?
But let’s focus on the mirror design. Any other ideas?
- Banned
- #25
Hrvoje Simic
Banned
Acehole,
I like this idea.
In this solution the mirror would be reflective on both sides, and one revolution would cover 4 frames.
Andrew M.
Well-known member
And what is the speed equivalent on rotating mirror?
1/250 1/60?
We have to stop rotating when taking shoot of the frame.
This have to be some kind of electrically polarized plate rather then actual rotating mirror.
What is the speed on Misterium while reading ? 1/60 I presume at 60fps
1/250 1/60?
We have to stop rotating when taking shoot of the frame.
This have to be some kind of electrically polarized plate rather then actual rotating mirror.
What is the speed on Misterium while reading ? 1/60 I presume at 60fps
jbeale
Well-known member
Stopping the mirror and then restarting twice per exposure presents some mechanical difficulties. With the right kind of sensor you could imagine shifting the pixels electronically in sync with the mirror rotation to accumulate the image as the mirror rotates, but this may be impractical for several reasons.
I wonder if it is possible to have a two-layer sensor, sort of a simplified Foveon design, with the top layer being a regular Bayer-type array as is used now, and then beneath that is a low-sensitivity layer that picks up a dim trickle of photons not absorbed in the top layer, but enough to give you a few more bits (= stops) of dynamic range in the highlights.
You might even not have to actually run any more pixels per second through the A/D, if you have some switch at each pixel so only the top layer is read, unless it has reached full-well status, in which case you switch over and use the bottom pixel and left-shift the result by X bits. This would be a step towards a native "log-output" sensor (like film, and the human eye) as opposed to the linear sensors like all the CCD and CMOS chips I'm aware of so far.
I wonder if it is possible to have a two-layer sensor, sort of a simplified Foveon design, with the top layer being a regular Bayer-type array as is used now, and then beneath that is a low-sensitivity layer that picks up a dim trickle of photons not absorbed in the top layer, but enough to give you a few more bits (= stops) of dynamic range in the highlights.
You might even not have to actually run any more pixels per second through the A/D, if you have some switch at each pixel so only the top layer is read, unless it has reached full-well status, in which case you switch over and use the bottom pixel and left-shift the result by X bits. This would be a step towards a native "log-output" sensor (like film, and the human eye) as opposed to the linear sensors like all the CCD and CMOS chips I'm aware of so far.
Ace
Wedge In
- Joined
- Dec 28, 2006
- Messages
- 1,493
- Reaction score
- 0
- Points
- 0
- Age
- 42
- Location
- Sydney, Australia
- Website
- www.dlshs.net
Actually I was talking about rotating from a 45 degree angle, so that it doesnt flip at any time. It just spins on one axis and the tilt determins the direction of reflection.
Also, why would you need to stop the mirror? Do film shutters "stop" to expose a frame on camera?
Andrew M.
Well-known member
LCD shutter or electronically activated reflection mirror could be used instead of mechanical shutter.
But I think the best idea is to use half reflecting mirror.
This way the first sensor that needs more light is the one right behind the mirror.
The second sensor that works only for mid and high light intensities is gathering light from semi-reflective surface. I guess we could experiment with altering electrically the reflective factor of the mirror so for 1/60 of the second the mirror is fully reflective and for another 1/60 of the second mirror is not reflecting at all, but this will cut the light quantities to both sensors and it is good for the second sensor but not good at all for the first one.
Any other idea how to direct the low light levels to the first sensor and high levels to the second one at the same time?
LCD shutter, rotating parabolic mirror that flips from one sensor to the other without rotation distortion…more key words needed….
But I think the best idea is to use half reflecting mirror.
This way the first sensor that needs more light is the one right behind the mirror.
The second sensor that works only for mid and high light intensities is gathering light from semi-reflective surface. I guess we could experiment with altering electrically the reflective factor of the mirror so for 1/60 of the second the mirror is fully reflective and for another 1/60 of the second mirror is not reflecting at all, but this will cut the light quantities to both sensors and it is good for the second sensor but not good at all for the first one.
Any other idea how to direct the low light levels to the first sensor and high levels to the second one at the same time?
LCD shutter, rotating parabolic mirror that flips from one sensor to the other without rotation distortion…more key words needed….
Bruce Allen
Well-known member
Yes, exactly! I'd rather have a 3K camera with high dynamic range... Introducing a much more complex demosaic algorithm is okay too as long as computers keep getting faster.
Bruce Allen
www.boacinema.com
Graeme Nattress
RED Problem Solver
We picked the sweet spot for today, but all this stuff is interesting discussions for the future.
Graeme
Graeme
Andrew M.
Well-known member
Just talking to the other people around here about intrinsic characteristic of the CMOS sensors, looks like there is a blind time spot where sensor is shunted to clear the photocell charge for next frame exposure.
Each CMOS is different so I can’t speculate on the Misterium design, but if there is a blind time for Misterium, this time could be used for the second sensor to gather the high intensity light. So electronically triggered mirror would be a nice option for this.
Another option is to create light memory turbocharger that will extrapolate intensity of the light based on the current curve of the charge voltage increase. So once the first sensor is shut out of light and the second sensor is receiving the light, the first sensor will be extra turbo charging its photocell based on the ramp up of voltage.
Each CMOS is different so I can’t speculate on the Misterium design, but if there is a blind time for Misterium, this time could be used for the second sensor to gather the high intensity light. So electronically triggered mirror would be a nice option for this.
Another option is to create light memory turbocharger that will extrapolate intensity of the light based on the current curve of the charge voltage increase. So once the first sensor is shut out of light and the second sensor is receiving the light, the first sensor will be extra turbo charging its photocell based on the ramp up of voltage.
Bruce Allen
Well-known member
Agreed! It's so cool having talks about ideas for improving dynamic range and not feeling that our comments are disappearing into a corporate black hole. I like small companies that respond to their customers. It makes me want to live off of cans of soup for months so I can afford a share in a Red One. Actually, that, combined with working straight through NAB, might just get me there... Anyway, enjoy NAB and good luck.
Bruce Allen
www.boacinema.com
Thom Steinhoff
Well-known member
Very interesting idea, though as graeme said, they have the best solution to date.
Continuing your idea, though, if money was no object, I would think it would be cool to have the rotation axis of the mirror motor pointing directly down the barrel of the lens. Then you could have the top mirror (resting) be for the optical tap, then left right and bottom are the three chips or even get more drastic and have no optical and have six chips even spaced around the barrel, giving you the ability to have 2 sets of 3 (giving you more dynamic range maybe?)
Cool idea. I'll stick with my Red One, though, as it has no moving parts!
Thom
- Banned
- #35
Hrvoje Simic
Banned
E-xactly. Hence the community.
Thinking ahead is what made 4k.
RED's recognition of the power of idea input is another brilliant move. Disregarding the sea of ideas is like locking the countless doors to possible future. Personally, I can't stop thinking ahead and ideas come all the time. That's what got me to all this advanced multimedial expression tools in the first place.
I'm sure many people here also appreciate this open channel for that endless flow.
We'll keep the flow running as long as you guys are listening...
Michael Brennan
Well-known member
- Joined
- Mar 13, 2007
- Messages
- 724
- Reaction score
- 0
- Points
- 0
If we are talking about the same camera, the primary purpose of the 4th chip was to avoid a European import duty that was defined as "3 chip ccd cameras".
So Ikegami built a 4 chip ccd camera!
Mike
- Banned
- #37
Evangelos Achillopoulos
Well-known member
- Joined
- Apr 10, 2007
- Messages
- 558
- Reaction score
- 0
- Points
- 0
- Location
- Athens, Greece
- Website
- www.motionfx.gr
Hello to all Red forumers…
I’m new here, am just reading for months now…
I decided to register, I haven’t reserve any Camera but in Greece am the evangelist (no it’s not coming from my name) of the digital era of Cinema for many years so it’s coming…
Graeme I have tried in my FT 2K your demo plugins and they are quite nice.
I have played a little bit with a Foveon sensor the Fx17-78-F13D few months ago from Alternative Vision and it was awesome the results even though I use it only in 1280/720 in 24fps in a four pixels to one configuration and the images where stunning. We where experimenting in our Lab in order to build a scanner… This sensor doesn’t need to de-mosaic and their resolution is far better than typical CMOS one chip in the same pixel count.
I was wondering do you thing to use same technology like this in the future cameras.
I thing the quantities that you gone to order from Foveon they will justify such an R&D for them.
For further reading www.foveon.com read under technology.
I will soon be one of your RED users since I will order one after NAB for my Rental/2KDI business…
Best regards and remember a start is half of everything…
I’m new here, am just reading for months now…
I decided to register, I haven’t reserve any Camera but in Greece am the evangelist (no it’s not coming from my name) of the digital era of Cinema for many years so it’s coming…
Graeme I have tried in my FT 2K your demo plugins and they are quite nice.
I have played a little bit with a Foveon sensor the Fx17-78-F13D few months ago from Alternative Vision and it was awesome the results even though I use it only in 1280/720 in 24fps in a four pixels to one configuration and the images where stunning. We where experimenting in our Lab in order to build a scanner… This sensor doesn’t need to de-mosaic and their resolution is far better than typical CMOS one chip in the same pixel count.
I was wondering do you thing to use same technology like this in the future cameras.
I thing the quantities that you gone to order from Foveon they will justify such an R&D for them.
For further reading www.foveon.com read under technology.
I will soon be one of your RED users since I will order one after NAB for my Rental/2KDI business…
Best regards and remember a start is half of everything…
Graeme Nattress
RED Problem Solver
We had a discussion on the foveon stuff elsewhere, but there are real issues using it:
it doesn't have the fps for 4k.
Too low a resolution, just 2.5k at the moment.
Lots of colour issues / noise due to the aggressive matrix we need to get the colour right, so on and so forth.....
Graeme
it doesn't have the fps for 4k.
Too low a resolution, just 2.5k at the moment.
Lots of colour issues / noise due to the aggressive matrix we need to get the colour right, so on and so forth.....
Graeme
Andrew M.
Well-known member
Hello, Evangelos, welcome to the forum.
There is many articles that I posted links to it here on this forum, comparing CMOS, Fevon and CCD. One good one, you can find on the Dalsa website is here:
http://www.dalsa.com/dc/documents/I..._Whitepaper_Digital_Cinema_00218-00_03-70.pdf
In short, as per current state of technology, you can pack more photo-sensors per mm on CMOS sensor then on Fevon. All three technology developers are investing a lot of money in improving the characteristics of these sensors but as of now, if you want the speed and low heat dissipation the CMOS is bit more advanced. After all, if you pack 20 photo-sensor in place of one in Fevon you have 20 pixels to work with instead of one.
CMOS technology in the labs is pushed as far as 48MP and when Fevon will get even half the way there CMOS will be at 128MP. CMOS has its own problems to overcome as well but so far CCD and CMOS are technologies where the most R&D money is invested in.
There is many articles that I posted links to it here on this forum, comparing CMOS, Fevon and CCD. One good one, you can find on the Dalsa website is here:
http://www.dalsa.com/dc/documents/I..._Whitepaper_Digital_Cinema_00218-00_03-70.pdf
In short, as per current state of technology, you can pack more photo-sensors per mm on CMOS sensor then on Fevon. All three technology developers are investing a lot of money in improving the characteristics of these sensors but as of now, if you want the speed and low heat dissipation the CMOS is bit more advanced. After all, if you pack 20 photo-sensor in place of one in Fevon you have 20 pixels to work with instead of one.
CMOS technology in the labs is pushed as far as 48MP and when Fevon will get even half the way there CMOS will be at 128MP. CMOS has its own problems to overcome as well but so far CCD and CMOS are technologies where the most R&D money is invested in.