hacking the ccd

[Matt Champagne]

I have a question that sort of concerns the over all nature of a ccd itself. It's my understanding that a ccd chip takes in light and converts it to an analog eletronic data on a pixel by pixel basis, and then each row of pixels is sequentially read and shifted from one side to the other at the speed of the clock.

My idea is to increase the clock speed of a still CCD, in order to make it a video CCD.

My theory is that therefore the limiting factors of a ccd capture are the clock speed, and the speed of the A/D converter and the related capture circuitry. So if you wanted to take a still CCD and bring it up to video frame rate, why not increase the clock speed of the ccd to some multiple of its original speed, place the data on a small register buffer, then create multiple A/D converter circuits which would each pull off of the buffer sequentially. Then after each image is captured the images themselves would have to be put on some sort of RAM buffer and sequenced properly.

What I would like to do with this is use something along the lines of the new canon digital camera with the full 35mm frame ccd, and get both the DOF natively as well as being ultra high definition.

I realize this is not simple...but more importantly anyone see a reason this wouldn't work?

And to answer the question before it comes up...I'm thinking gigabit ethernet would be the way to transfer that much data to a hard drive (provided the hard drive can capture that fast...i need to look into that).

[Rob Lohman]

A harddisk cannot store that fast enough. Personally I would look
into building your own camera if I where you (see the current
threads on that topic in this forum).

I don't know everything about CCD's etc., but I'm not sure what
you want to do is possible. From what I understand it the
components on the CCD itself also have speed limits in which
they can do a readout etc. Of course, I could be wrong on that....

You would probably destroy a very expensive camera in the
process as well. Keep in mind that true 35mm digital camera's
are very very expensive.

I'm doubting whether you understand all the numbers. For example,
the Canon 1Ds Mark II has a resolution of: 4992 x 3328

That's 16,613,376 bytes (or almost 16 MB per FRAME!). Just at
24 fps this creates a staggering 398,721,024 bytes per second
or 380 (!!) MB/s. You will at least (probably more) need a 10
harddisk based array (consider the power that this will draw).

And I'm not even talking about editing and working with that
footage, both in strain on your computer(s), time it will take to
process it and your ability to view such resolutions.

In my opinion it is a far better idea to work with 720p or 1080p
resolutions (which is proving challenging enough already!) which
already produce fast amounts of data and (1080p) is quite close
to the resolution a lot of films are processed at (not a lot are done
at 4K yet...)

If you are looking for true 35mm depth of field with higher resolutions
a self built system with a HD cmos or ccd chip with a 35mm adapter
sounds much more plausible...

[John Sandel]

Other than that, dude, you're in!

[Matt Champagne]

I realize all the data issues, and I imagine that substantial down res or compression would have to be applied. The key reasons I want to do this, however, are not so much the resolution as the native 35mm DOF without light loss, 4:4:4 sampling, and *i think* a higher color bit depth (its 16 bit isn't it?). To me the issue with the common dv camera isn't resolution but color depth and color sampling...so i'm willing to down res all the way to standard def...provided I could keep get the DOF and the colors.

The main thing I want to know is does anyone have any experience over clocking a ccd or cmos? I have only seen it done once...but it was on a cheap gameboy B&W camera, and they increased the frame rate from 0.8fps to 3fps.

On the other hand...if a 10 disk array did work at full res....why wouldn't you do it (editing issues aside of course)? $8,000 plus a new hack circuit design plus a $15,000 hard drive array...around $25,000 for a system that would put the Dalsa to shame. :0)

[Rai Orz]

I love crazy ideas.
Do you know the ARRI D20 technical data? And the data output? Its not crazy. It work. And sensors? There is a PDF on ARRIs side (something about bayering or so). In this PDF you find exact cmos sensor datas (dimension and pixels). Now do a google seach with this walues and see what you find... A Canon!... Coincidence?

[Rob Lohman]

I thought the current sensors do a max of 10 or 12 bits, but it is
all in the ANALOG -> DIGITAL samplers. These decide how many
bits you can use.

[Juan M. M. Fiebelkorn]

It could easily be a Fillfactory one too...
3000x2200 and 8.5 pixel pitch sounds too Fillfactory to me...

[Matt Champagne]

Rai: Very interesting indeed. I think I found the pdf you were refering to, and it does seem quite simular. It also alerted me to another potential problem which would be the shutter. There is no way to really be sure that a still cameras shutter can shutter at the right speed on a continuous basis. So perhaps an LCD shutter like in the vance cam would be necessary.

Rob: You may be right there...but does that mean that the limit of 12 bits is even there for the ccd, or is that only the current conventional limit of A/D (which would be awesome because using this parallel structure, you could use practically any A/D)

[Rai Orz]

The ARRI shutter have also a additional feature. There is a mirrow on the shutter wings. This is a old ARRI thing, from mechanical movie cameras, for optical viewfinders (If the SENSOR/Film is closed, the mirrow beams the picture to the viewfinder, whithout lightloss).

I think a LCD shutter is not a great idea. You lost light. But a simple mechanical rotation shutter can be build easy and sheap.

Most CCDs have anolge output. As Rob said, how many bits you get is just a A/D question. There are some chips (round $ 10-40,-) to do this job, form 8 to 24Bits.
CMOS sensors have a build in AD converter with fixed Bits (most 10 or 12), but some Sensors have also a analoge out pin, so you can also use external AD converters like before.

[Rob Lohman]

Matt: I would guess that this mainly depends on the A/D converter
(and your ability to store the bits, which is probably the main
reasons why everyone is still using 10/12 bits => digital photo
camera's still downsample to 8 bits unless you shoot in RAW mode).

However, I can imagine the CCD/CMOS sensors have a max where
you are just sampling noise. This probably has to do with how
light sensitive the thing is, but I'm really getting outside of my
knowledge here, so that's just a "logical" (??) guess...

[Juan McFarland]

Matt, I read your post with great interest and yes, I feel that the idea you propose is possible.

;)

[Marcelo Mazzariol]

Hi Matt., we belive it is possible and want to give it a try. Have you got your hands on that since the first post?


cheers
Marcelo

[David Heath]

Originally Posted by Matt Champagne

I realize this is not simple...but more importantly anyone see a reason this wouldn't work?

You seem to be proposing a huge increase in clock speed, and I suspect there would be all sorts of stray capacitance issues which would cause problems with reading non-digitised samples off in the manner you propose.

Even if you got over that, think about over-clocking computers. Beyond a small amount, the first necessity is to increase the cooling of the chip - faster=more power=hotter. If nothing else, I doubt you'd get much read off the chip at that speed before it melted.

[Ben Winter]

Originally Posted by Matt Champagne

My idea is to increase the clock speed of a still CCD, in order to make it a video CCD.

I realize this is not simple...but more importantly anyone see a reason this wouldn't work?

High-speed CCDs are manufactured with registers out of PMOS nwell transistors with short lengths to reduce switching speed. The amplifier has a higher unity gain frequency as it is designed for higher speed. As David mentioned, larger size means more capacitance which will royally screw you at high frequencies. As I'm taking a class on sensor design right now I can tell you, they are designed for specific applications. Smaller channels for faster speeds requires more accurate manufacturing methods which very simply comes down to price. Companies like AltaSens spend millions on R&D to develop sensors that match specific applications, if they could "hack" their still-image high-res sensors to work at high speeds, well, life would be awfully easy for us engineers.

If you want to hack anything, play around with CMOS sensors, they're much more responsive to being toyed around with.