Large chip 3-chip designs?

[Robert Jackson]

I was wondering if anyone would like to talk about why large 3-chip designs are impractical? I've been really curious about why there are so many 2/3" 3-chip designs and that seems to be the point where people stop considering a multi-chip design as practical.

I'm sorry if this is obvious. Light falloff in large prisms or just a size-related issue or something, but I thought it was probably something that had been given significant consideration by the Red designers and wanted to ask.

 

[Gopher77]

I think I had read a post by Gramme that answered that. You just can't push that much data through.

 

[Mark B.]

Alignments get seriously critical when you're dealing with huge pixel counts on three different chips... trying to match them up perfectly would be a pain.

 

[Graeme Nattress]

Bigger prisms would lead to all kinds of problems with the lenses, and the distance between the back of the lens and the sensors would increase further and lead to all kinds of issues, not least the cost of 3 sensors rather than 1. You'd not be able to use standard lenses, so you'd need to buy new, and that would be a pain.

With our approach, you get to use standard lenses and get a great picture.

Graeme

 

[Michael Brennan]

large 3 chip

large 3 chip

Not large format, but there is a 4 chip ccd 4k camera head made by Olympus that is quite good, it uses pixel shifting and existing 2/3 inch HD lenses.
Olympus have not made the camera available for sale.
Panasonic have introduced pixel shifting to their new 1920x1080 camera head.

A multi chip non bayer mysterium with pixel shifting could double resolution but would need custom lenses. In theory and ignoring cost of lens design(!) cost, size and compexity of shooting would be probably akin to shooting 65mm IMAX?



Mike

 

[Graeme Nattress]

Unless those chips where 65mm dimensions, I'd say not. There's no substitute for size.

Pixelshifting is an interesting technique, but the only paper I read on it showed a luma increase of root(2) or about 1.4. It does seem to get used to try for a 2x factor increase in resolution, for which I've not seen a scientific justification for.

Graeme

 

[Andrew M.]

Talking about 3-chip design.
What about 2-chip design, to increase dynamic range.
Can we have semi transparent mirror in front of the first sensor and second sensor next to it, one responding to lower light levels and the second responding to the higher light levels?

Alternatively, is there a way to read only one CMOS sensor, first fast, for the intense light levels and then slow for the low light levels?

I was thinking to use 120fps in 4 intervals.
We use first ¾ to read low light levels and ¼ to read high levels.
Then to combine it as a one 30fps frame with extra two or 4 stops dynamic range?

 

[Jeff Kilgroe]

Pixelshifting is an interesting technique, but the only paper I read on it showed a luma increase of root(2) or about 1.4. It does seem to get used to try for a 2x factor increase in resolution, for which I've not seen a scientific justification for.

I'm not a big fan of pixel shifting. It does work, kinda, sorta in many situations. It essentially "doubles" the luma info for each pixel in the CCD array and can give increased resolution with midtone details and it's partial to neutral and grey colors. I often wonder if pixel shifting is more of a way of cheating when shooting those B&W and grey scale resolution charts.

Pixel shifting is easily defeated with certain types of camera motion or movement of the subject. CCD resolution of the HVX was originally confirmed by shooting a res chart with wobulation. ...I think it was Steve Mullen who did this. Any benefits to pixel shifting also break down when the green content of a source image is far lower than the red and blue components. So, if you're shooting a scene with lots of reds, purples, shadows, etc.., you're not gaining the benefits of a shifted green sensor layer. Which also leads to another method of testing true resolving power of shifted CCD cameras -- shoot a res chart under red light or shoot as normal and only analyze the red CCD layer. Although, this may not be as accurate as wobulation because you would be introducing a luminance handicap by only dealing with red light or CCD info.

 

[Andrew M.]

So, if you're shooting a scene with lots of reds, purples, shadows, etc.., you're not gaining the benefits of a shifted green sensor layer. Which also leads to another method of testing true resolving power of shifted CCD cameras -- shoot a res chart under red light or shoot as normal and only analyze the red CCD layer. Although, this may not be as accurate as wobulation because you would be introducing a luminance handicap by only dealing with red light or CCD info.

But isn’t that the same true for Bayer sensors?
Misterium is just 3MP Red or Blue sensor.
Once there is an absence of Green its resolutions drops far below 4K format standard?

 

[baro]

Talking about 3-chip design.
What about 2-chip design, to increase dynamic range.
Can we have semi transparent mirror in front of the first sensor and second sensor next to it, one responding to lower light levels and the second responding to the higher light levels?

At some forum somebody told Ikegami made a 4 chip camera once with that purpose (the fourth chip only received a small fraction of the incoming light). Ever since I hoped I would hear some more about it without success. I would be curious to hear if anybody has more information on this.

 

[Graeme Nattress]

Some of the earliest tube cameras had 4 tubes, with one dedicated to luma.

Graeme

 

[baro]

Maybe this is what the poster meant, thank you.

 

[Hrvoje Simic]

Talking about 3-chip design.
What about 2-chip design, to increase dynamic range.
Can we have semi transparent mirror in front of the first sensor and second sensor next to it, one responding to lower light levels and the second responding to the higher light levels?

...been thinking about the same thing, and glad I'm not alone.
Is this possible, Graeme ?
Both sensors would have the same distance to lense, just the second one would be on a 90 degree angle.

What are the obstacles in achieving that ?

 

[Robert Jackson]

...been thinking about the same thing, and glad I'm not alone.
Is this possible, Graeme ?
Both sensors would have the same distance to lense, just the second one would be on a 90 degree angle.

What are the obstacles in achieving that ?

Didn't you just cut the light reaching both sensors in half? Maybe there are some carbon arc lamps left over from the production of Birth of a Nation that you can pick up cheap. ;-)

 

[Anders Holck]

The obstacles in achieving that, would be alignment.
With the single sensor you just have to make sure the sensor is at the right distance from, and is perfectly flat to the flange. If it's shifted a bit up or down, left or right or even rotated (around the optical center) wont affect image quality that much.

In your setup both sensors and the mirror must be alligned to perfection in all 6 axis. Not to mention you's have to double the processing power.

But the setup could work I guess, 3-strip technicolor worked in a similar way.
Simply not worth it, even if you got it to work.

 

[Robert Jackson]

But the setup works, 3-strip technicolor worked in a similar way.

By the time light passed through the Technicolor beam-splitter and filters, the effective ASA of the Technicolor stock was only 5.

Interestingly enough, this is exactly why Citizen Kane was possible. Shooting at Selznick’s sound stages, Gregg Toland was able to make use of the Technicolor lighting the studio had purchased to film ‘Gone With the Wind.’ Shooting Kodak Panchromatic Super-XX with an ASA of 160 and push processing it to 320 he was able to use that Technicolor lighting rig to allow his lenses to be stopped down until he had sufficient depth of field.

Sorry. Wrote a paper about that in film school. ;-)

 

[Hrvoje Simic]

The obstacles in achieving that, would be alignment.

Yeah, I was thinking that alignment could be a problem...
Here's a quick idea of a assembly principle:

1. Sensors are mounted on the identical plates
2. 90 degree frame with mirror slot is made from a universal precise mold
3. Plates and mirror are mounted on the 90 degree frame
4. Additional sensor's x&y axis per-pixel alignment is achieved electronically through cropping (with 10-50 unused pixels tolerance)

Processing power will be no problem in the future.
Robots today are capable in achieving that type of precision, which leads to the next obstacle - cost of r&d.

 

[Robert Jackson]

Yeah, I was thinking that alignment could be a problem...

You know, Fuji's solution with dual 'S' and 'R' photosites seems to work pretty well without a beamsplitter. The second set of photosites are each really small, but their intended use is in situations with hot highlights, so it doesn't seem to matter much. It takes a lot of processing power to combine the data, though, and the file sizes are really huge.

 

[Andrew M.]

Processing power is not a problem if we do prepare data directly on the chip in the hardware.
Two sensors will be specially manufactured so the output side of the first sensor (first 12 bits) will expect the other sensor to be there starting from bit 5 to 16 so it will just combine it in to the 16 bit sensor output instead 12 bit one.
Actually the first sensor (low light level one) will connect to the second sensor with 12 bits and second sensor will have 16 bit output ready.

1 2 3 4 5 6 7 8 9 10 11 12 -- -- -- --
---------1 2 3 4 5 6 7 8 9 10 11 12
=========================================
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Alignment, hmm… we have more problems with alignment then that in 3CCD prism splitter and they manage to align 3 sensors there via prism.
Though I like Omens electronic self alignment idea.
Now, we cut the light by half but we combine two sensors data so we get it back at minimum in the middle luminosity range.
Also, one sensor will be manufactured to be optimized for low light levels the second for the high levels and two will be matched together. So once you have one of them defective you have to change two of them. Alternatively the optimization could be done by mirror splitter itself, sending less light to the bottom sensor and more light to the side one.

 

[Bruce Allen]

Totally agree, guys. A quick question though - why couldn't you also just make every second filter in the R, G, B Bayer matrix have a bit of ND / darker filter added to it as well? Sony did a R, G, B, E sensor, why can't someone make a filter that goes something like R, darkG, B, darkR, G, darkB? Or a R, G, B, ND filter, for that matter?

Bruce Allen
www.boacinema.com