Rai & Markus' "Drake" HD camera

Juan M. M. Fiebelkorn · December 11th, 2005, 10:40 PM

I guess he is trying to say that, seeing that most high end CMOS sensors are in fact some kind of SoC, there could be in theory one which had integrated memory to store every line at the same time on an internal memory line buffer.That way I guess there wouldn't be a need for having lots of pins to get the stored data out of the sensor.
Only problem I see is that for something like an Altasens you would need 2.8 kilobytes of memory per line....

Wayne Morellini · December 11th, 2005, 11:48 PM

Keith I mean digital memory per line or per sensor pad, so that the parallel operations can happen on chip and make for Global readout or very fast rolling read out, it then can be transmitted out of chip over a few pins while the charge for the next frame is accumulating (on the cheapest chips, the number of pins can become the major cost item).

I am pretty sure now that I remember sensors with on chip memory, but I don't know if they were used this way or not.

3MB of memory is relatively low (unless it is static, but static is not required here).

Keith Wakeham · December 12th, 2005, 06:01 AM

In theory it could be done, but in practice it gets more difficult. As of yet I haven't seen any cmos that has a full frame of memory to do what you want. 3mb isn't a lot but you have to consider that these sensors are made on a process much larger than current processor and ram chips so the physical size of rame gets a little larger then what we are used to.

To be honest the only real global shutter that is easy to operate is IT ccd and FT ccd. Both have charge storage areas that the build up charge is shifted to and then read out from. I think CCD's have gotten a bad rap here from people looking for "The Cmos Look" while ignoring the benefits available from CCD's and the fact that they don't suffer from the rolling shutter. Where as with a global shutter cmos you need to intergrate, readout, then shutter, so if you want long intergration times you need to read out very fast. Whereas the intergration in a ccd is takeing place during the readout of the previous frame.

Wayne Morellini · December 12th, 2005, 05:27 PM

I'm aware of the process issues, but 3MB, is long past history process wise (unless your using static memory that can use at least 6 transistors per cell (it can be done for less, but those schemes are not as good as the high transistor count schemes)). The problem is setting themselves up to have access to those processes, is difficult (maybe it has changed now days). There is Camera/Web/etc sensors with frame buffers, but they would be using a separate buffer/micron-controller/encoder on chip, not integrated in the sensor pads. So I doubt they are setup to solve rolling shutter issues.

It is interesting that the Rolling shutter issue was a CCD problem before it was a cmos problem, but the newer sensor technologies you mentioned solved that. The interesting thing is, does this memory effectively handle overcharge smear/streaking and does it still accumulate noise on it's travels to the A/D circuit?

Jason Rodriguez · December 12th, 2005, 11:14 PM

Actually there are CMOS sensor with memory on them, and that's how you get global shutter.

For instance, that's how the IBIS5A and the LUPA sensors from Fill Factory work. They use a 4T and a 6T (transistor) pixel structure that allows them to have "memory" to store the pixel charge so that the entire sensor can be reset at once. The only problem with the more transistors per pixel, is that you can have problems with column non-uniformity and fixed-pattern-noise. That is why all high-quality CMOS sensors you see are 3T pixel structures of some type (or at least they try to keep it very low, either 3T or 4T with non-global shutters), meaning that they will more-than-likely have rolling shutters.

Rolling shutter though is no longer an issue when the read-out of the sensor goes at 1/60th of a second or higher, and the shutter is not faster than 1/60th of a second. If you don't believe me, go get a Ikegami CMOS camera, set it to 1080/24p, and wiggle it around like mad-you're going to have a hard time to see any rollling shutter at the "normal" settings. Now those guys over on that Sony forum are setting their shutters at 1/200th or greater-and yes, at the read-out speeds of the Sony CMOS chip (which I'm assuming are at 1/60th of a second since it's NTSC compatable), when you "stop-motion" like that, you're going to see some bending on objects. But if he had his shutter set at a "normal" speed, then this wouldn't be an issue.

DSLR's get around the problem by having a mechanical shutter. Same thing with the Arri D20. But other cameras like the Kinetta, Ikegami, etc. are NOT using any mech shutters, they are simply clocking the chip higher, and then reading out every-other frame; or at least something along those lines. And I've seen both series of cameras, and tried to induce visible rolling shutter problems by shaking the cameras around wildly, and have been unsuccessful in producing anything "offensive" or even noticeable for that matter.

Frank Schoerner · December 13th, 2005, 05:20 AM

Quote:

Originally Posted by Jason Rodriguez

DSLR's get around the problem by having a mechanical shutter. Same thing with the Arri D20. But other cameras like the Kinetta, Ikegami, etc. are NOT using any mech shutters, they are simply clocking the chip higher, and then reading out every-other frame; or at least something along those lines. And I've seen both series of cameras, and tried to induce visible rolling shutter problems by shaking the cameras around wildly, and have been unsuccessful in producing anything "offensive" or even noticeable for that matter.

...And thats why it is not possible to get movie like motion blur with it. It may be god for sport but not for films.

PS: As i heard the other people from the drake team do there business like before. I dont know if drake ever will sell, but if, than like the fist version. All what i saw, the main parts, ideas and also solutions come from Rai, and without him there will be no new developments. But i think it was a god idea for Rai (and also for us) to search new partners, because now with bigger partners he have mutch more possibilities. I had a long conversation with him and he show me whats going on. I can not post it here, but i think i can qoute what Rai wrote here at dvinfo "...my goal is a real 35mm size camera..."

Levan Bakhia · December 13th, 2005, 07:31 AM

If you have contact with Rai, then tell him that if he wants, he can contact me, and I have partners here that can fund and invest in the development of the camera theoretically, if he want I would discuss this issue with him.

my e-mail is: levan@sarke.ge

Wayne Morellini · December 14th, 2005, 08:07 AM

I think that is a great idea, getting together on custom commercial cameras is good for development at this stage.

Tell him to contact me as well. But I really like to know what the direction of the two companies is, and website/business name for both.

Wayne Morellini · December 14th, 2005, 08:18 AM

Quote:

Originally Posted by Jason Rodriguez

For instance, that's how the IBIS5A and the LUPA sensors from Fill Factory work. They use a 4T and a 6T (transistor) pixel structure that allows them to have "memory" to store the pixel charge so that the entire sensor can be reset at once. The only problem with the more transistors per pixel, is that you can have problems with column non-uniformity and fixed-pattern-noise.

Has every pad on the IBIS5A have a AD converter and not a memory capacitor to store the charge, I thought it was a signal one that could be bypassed to an off chip AD? Explains a bit.

December 12th, 2005, 06:01 AM	#393
Keith Wakeham Major Player Join Date: Jan 2005 Location: St. John's, NL, Canada Posts: 416	In theory it could be done, but in practice it gets more difficult. As of yet I haven't seen any cmos that has a full frame of memory to do what you want. 3mb isn't a lot but you have to consider that these sensors are made on a process much larger than current processor and ram chips so the physical size of rame gets a little larger then what we are used to. To be honest the only real global shutter that is easy to operate is IT ccd and FT ccd. Both have charge storage areas that the build up charge is shifted to and then read out from. I think CCD's have gotten a bad rap here from people looking for "The Cmos Look" while ignoring the benefits available from CCD's and the fact that they don't suffer from the rolling shutter. Where as with a global shutter cmos you need to intergrate, readout, then shutter, so if you want long intergration times you need to read out very fast. Whereas the intergration in a ccd is takeing place during the readout of the previous frame. __________________ www.engr.mun.ca/~wakeham/index.htm

December 13th, 2005, 07:31 AM	#397
Levan Bakhia Major Player Join Date: Apr 2005 Posts: 230	@Frank If you have contact with Rai, then tell him that if he wants, he can contact me, and I have partners here that can fund and invest in the development of the camera theoretically, if he want I would discuss this issue with him. my e-mail is: levan@sarke.ge

December 11th, 2005, 10:40 PM	#391
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	I guess he is trying to say that, seeing that most high end CMOS sensors are in fact some kind of SoC, there could be in theory one which had integrated memory to store every line at the same time on an internal memory line buffer.That way I guess there wouldn't be a need for having lots of pins to get the stored data out of the sensor. Only problem I see is that for something like an Altasens you would need 2.8 kilobytes of memory per line....

December 11th, 2005, 11:48 PM	#392
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Keith I mean digital memory per line or per sensor pad, so that the parallel operations can happen on chip and make for Global readout or very fast rolling read out, it then can be transmitted out of chip over a few pins while the charge for the next frame is accumulating (on the cheapest chips, the number of pins can become the major cost item). I am pretty sure now that I remember sensors with on chip memory, but I don't know if they were used this way or not. 3MB of memory is relatively low (unless it is static, but static is not required here).

December 12th, 2005, 05:27 PM	#394
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	I'm aware of the process issues, but 3MB, is long past history process wise (unless your using static memory that can use at least 6 transistors per cell (it can be done for less, but those schemes are not as good as the high transistor count schemes)). The problem is setting themselves up to have access to those processes, is difficult (maybe it has changed now days). There is Camera/Web/etc sensors with frame buffers, but they would be using a separate buffer/micron-controller/encoder on chip, not integrated in the sensor pads. So I doubt they are setup to solve rolling shutter issues. It is interesting that the Rolling shutter issue was a CCD problem before it was a cmos problem, but the newer sensor technologies you mentioned solved that. The interesting thing is, does this memory effectively handle overcharge smear/streaking and does it still accumulate noise on it's travels to the A/D circuit?

December 12th, 2005, 11:14 PM	#395
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	Actually there are CMOS sensor with memory on them, and that's how you get global shutter. For instance, that's how the IBIS5A and the LUPA sensors from Fill Factory work. They use a 4T and a 6T (transistor) pixel structure that allows them to have "memory" to store the pixel charge so that the entire sensor can be reset at once. The only problem with the more transistors per pixel, is that you can have problems with column non-uniformity and fixed-pattern-noise. That is why all high-quality CMOS sensors you see are 3T pixel structures of some type (or at least they try to keep it very low, either 3T or 4T with non-global shutters), meaning that they will more-than-likely have rolling shutters. Rolling shutter though is no longer an issue when the read-out of the sensor goes at 1/60th of a second or higher, and the shutter is not faster than 1/60th of a second. If you don't believe me, go get a Ikegami CMOS camera, set it to 1080/24p, and wiggle it around like mad-you're going to have a hard time to see any rollling shutter at the "normal" settings. Now those guys over on that Sony forum are setting their shutters at 1/200th or greater-and yes, at the read-out speeds of the Sony CMOS chip (which I'm assuming are at 1/60th of a second since it's NTSC compatable), when you "stop-motion" like that, you're going to see some bending on objects. But if he had his shutter set at a "normal" speed, then this wouldn't be an issue. DSLR's get around the problem by having a mechanical shutter. Same thing with the Arri D20. But other cameras like the Kinetta, Ikegami, etc. are NOT using any mech shutters, they are simply clocking the chip higher, and then reading out every-other frame; or at least something along those lines. And I've seen both series of cameras, and tried to induce visible rolling shutter problems by shaking the cameras around wildly, and have been unsuccessful in producing anything "offensive" or even noticeable for that matter.

December 14th, 2005, 08:07 AM	#398
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	I think that is a great idea, getting together on custom commercial cameras is good for development at this stage. Tell him to contact me as well. But I really like to know what the direction of the two companies is, and website/business name for both.