Home Made HD Cinema Cameras - Technical Discussion

Steve Nordhauser · October 7th, 2004, 01:51 PM

I realize that a lot depends on what you expect from the CPU (resolution, preview, compression) but I was looking at laptops and a few are just coming out with the Mobile P4 processors which are different from the P4-M. They have HT and a fast FSB but are supposed to be less power hungry than the standard P4.

http://www.intel.com/products/notebo...obilepentium4/

The Mobile Intel Pentium 4 processor is called model 548. The novelty operates at 3.33GHz, features 533MHz Quad Pumped Bus, Hyper-Threading as well as SSE3 technologies. The chip can be plugged into platforms based on Intel’s 852GME and 852PM chipsets. The new Mobile Pentium 4 processors feature power management Enhanced Intel SpeedStep technology that enables lower thermals than its desktop counterpart providing more reliable system performance in a notebook. Thermal Design Power (TDP) of the Mobile Pentium 4 processor 548 is 88W, which is lower than that of a 90nm desktop chip running at similar clock-speed.

Yes, I know, 88W is much more than 25W, but if you do RT compression and drop some disk drives, it looks a bit better.

Juan M. M. Fiebelkorn · October 7th, 2004, 05:34 PM

Anyway, if I'm not too wrong, there is a Pentium-M with 533 MHZ bus too...

Jason Rodriguez · October 8th, 2004, 05:47 AM

No, there's no 533Mhz bus yet for the Pentium-M, but the Dothan core is much better than the P4 Mobile chips (Netburst). I posted a link on the other threat about some guys talking over the Pentium-M vs. P4-M over at Andantech.

BTW, the dual Via boards are absolutely no use if you're trying for high-speed 1080p because there's no 64-bit PCI-X slot for a frame-grabber. That and the Eden-N processors are horribly slow.

Also the chipsets that you can get with the Penitum-M (855GME with 6300ESB) are much better.

Ben Syverson · October 8th, 2004, 06:15 PM

Wayne wrote:
"I am eagerly waiting to see what they do, have they shared with you what capture solutions they are using (capture software in particular)?"

Sumix is writing all their own custom software, and are very receptive to cinema-oriented features...

Wayne Morellini · October 8th, 2004, 08:01 PM

Thanks Ben, I long suspected that was the case. It is a pity that different companies didn't co-operate with Rob to establish his software as a defacto standard to shift with their cameras. Si has been great in co-operating with Rob though. The problem we now face is probably format. I would imagine everybody has his own, and their is not a standard one third party NLE's can plug into as things grow in the future. So they will all have to use convert to which ever compatible format they have on their list of file types, but still, it is very eartly days still lots of time left.

Wayne Morellini · October 8th, 2004, 08:27 PM

Jason, I know that VIA has lost a lot of initative to the Israeli Pentium M, and the new low powered Transmeta Rob posted. But they are improving their game, unfortunately the next version (64bit and much processing power) seems to be in 2006, with no improved version inbetween but only a revision or so (I was hoping the increased processing functions would be in there, but still a chance, but that would be mnore than a revision).

Still on the multilevel scheeme of things, the Pent M's I have seen in the past were very expensive (I don't know what has happened in price since). And dare I say it, the Pent M might be the best high end solution.

I think with compressed camera data rate, that should be on camera buffered and not require PCI-X (I hope somebody has told them to include some preview ability/ stream in it), 2Ghz+ chips, onboard decompression engine (that hopefully will have some compatible arcitecture to partly assist decompress), and cheaper costs. Also the architechture of the latest series has improved a lot over the last, but not as good as Pent M. I think there is still hope for the old VIA on the bottom end at least. But I would like to know what Rob's latest sustained FPS (not shutter speed) is, that should give a good idea on the power required (probably with a possible 50-100% improvement possible throiugh more very specialised programming optimisation).

So I hope to maybe see them on the lower end cameras, and I'm glad everybody has dropped the Pent 4 stuff for the Pent-M. Maybe sub 2Ghz Pent-M's will drop a lot in price now.

Wayne Morellini · October 11th, 2004, 01:58 AM

Steve,

a) The MT9M413, 500fps 10tap, 1.3Mpix 12 micron sensor with Truesnap (I think fillfactory also has similar one) is a big sensor, why haven't you built a camera on one of these, maybe even a cut down 24/30fps version. What would the light sensitivity compared to the other microns be like with these large pixels?

b) It is interesting that when you look at it, the sensor chip manufacturers could build in memory circuits to buffer the image while a new one integrates (which this one does, analogue style), or even anneal a memory circuit with matching pad contact to the back of a sensor chip (which is done in all sorts of processing circuits nowadays). Then run a simpler industry standard high speed serial interface out of it. All the camera manufacturers have to do is to build analogue interface buffering components to attache real world interfaces to it. That would be a win win situation, as most processing is done internally (and the pad contact not requiring pin wiring) only pins for the external interface and power are required. As these, now, integrated circuits are required for all cameras, and are being manufactured at the most cost effective stage, cost is further reduced. As it is buffered, frame grabbers are only needed for the fastest capture devices (above 100MB/s data rate), so cheaper interfaces are available. Control and signals can be sent via interfaces, and control circuits would be on chip, so filtering compression dsp could eventually be added (which results in better use of interface for higher resolution/data rate). If compression upto 50:1 (as well as lossless) was included then many could be strung on network applications (remembering the standard security/production line markets first) with lower powered, cheaper computers, using their integrated video decompression DSP's to view it, and disk savings.

Until they develop their own, there are many cell designs that could be integrated. ARM is the often used one (usually with memory) with many extra dsp like cells to the arm core, and I'm sure there are other dsp cells suitable for this sort of work.

Even if cheap sensors ($200) landed up costing twice as much (the real cost difference should be negligible) that would be offset greatly by big savings in the rest of the work chain.

Thanks

Wayne.

Wayne Morellini · October 11th, 2004, 07:29 AM

Found an interesting article "Heat wave: FPGAs confront increasing, evolving power consumption" at the EDN site. I saw somewhere that they are giving away a free years subscription. I used to read EDN at Uni, they are a top Electronic Engineering Journal for professionals and is well worth the reading.

http://www.reed-electronics.com/ednmag/article/CA438310

Finaly found some information on format reading in that cmos sensor article above:

Quote:

Deciphering size

As you compare manufacturers’ image sensors, you’ll often find several measures of the sensors’ size in the data sheets. These measures include total package size, the dimensions of the active array, and the aspect ratio—typically 4-to-3 or, mimicking 35-mm film, 3-to-2. You’ll likely also encounter the “optical format,” an at-first-glance baffling number with values such as 1/4, 1/3, 1/2, 1/1.8 and 4/3 in. The term harks back to the standard sizes manufacturers applied to 1950s-era TV-camera tubes; the specification refers to the outer diameter of the tube’s long glass envelope. This designation, clearly long obsolete in practice, is still in wide use. To translate optical format to the lenses’ projected imaging plane and, therefore, to the required diagonal dimension, multiply the optical format by a two-thirds scaling factor and, if necessary, convert from US units to metric units.

Steve Nordhauser · October 12th, 2004, 05:14 PM

Lens formats:
http://www.siliconimaging.com/Lens%2...%20formats.htm

Wayne on the MT9M413:
Sure we have discussed it. It is a tricky design - 100 parallel digital lines into an FPGA to be multiplexed down. Lots of data generated. Fairly high cost camera. Of course you could design it to run slow but all the same hardware has to be there except a simpler interface. We cover a similar high speed market with the much cheaper SI-640HF VGA at 250fps. If we found an OEM to justify it, we would do the design, but probably won't on speculation.

Laurence Maher · October 13th, 2004, 05:22 AM

Can someone please give me a link to the summix web site?

Thanks!

P.S.

How's that SI camera coming Steve?

Wayne Morellini · October 13th, 2004, 11:44 AM

Didn't realise they didn't multiplex down on chip etc, that would add significant cost.

Laurence, it's http://www.sumix.com/ there is a camera section under optical.

Steve Nordhauser · October 13th, 2004, 11:56 AM

Wayne on that Micron:
I just checked my notes from when I looked into it. In moderate volume, the sensor alone is >$1K, my cost. Do you really want to buy a 720p camera for over $5K? It would be an excellent price for a 1.3Mpix 500fps camera.....but.....

Wayne Morellini · October 13th, 2004, 12:06 PM

Thats OK, some people want that sort of thing, I thought it had good machine vision advantages, but I can do without it myself. Somebody mentioned to me a price around $850.

Well, "Collateral" is out today, and it was shot on Viper (and another camera) so I'm hopefully getting to see what the quality is like.

Wayne.

Juan M. M. Fiebelkorn · October 13th, 2004, 02:12 PM

Wayne, if you see Collateral on digital projection, that´s okay.
I saw it in a 35mm transfer and the quality of it wasn´t very good.
I would say it is because of the transfer quality.

Wayne Morellini · October 13th, 2004, 11:18 PM

Typical, they spend all that money to use digital and then they skimp on the transfer for an a-list movie. I'll have to re-affirm light and range from the normal DVD. Saw Alien VS Predator last week, and the cinema had turned the brightness up, lots of grey shadows, so that didn't help it try to live up to its A grade trailer (note: a lot of the A grade bits of the movie hint hint, think the other 5 people in the audience agree). So stuffed again in less than 8 days.

October 7th, 2004, 01:51 PM	#196
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	I realize that a lot depends on what you expect from the CPU (resolution, preview, compression) but I was looking at laptops and a few are just coming out with the Mobile P4 processors which are different from the P4-M. They have HT and a fast FSB but are supposed to be less power hungry than the standard P4. http://www.intel.com/products/notebo...obilepentium4/ The Mobile Intel Pentium 4 processor is called model 548. The novelty operates at 3.33GHz, features 533MHz Quad Pumped Bus, Hyper-Threading as well as SSE3 technologies. The chip can be plugged into platforms based on Intel’s 852GME and 852PM chipsets. The new Mobile Pentium 4 processors feature power management Enhanced Intel SpeedStep technology that enables lower thermals than its desktop counterpart providing more reliable system performance in a notebook. Thermal Design Power (TDP) of the Mobile Pentium 4 processor 548 is 88W, which is lower than that of a 90nm desktop chip running at similar clock-speed. Yes, I know, 88W is much more than 25W, but if you do RT compression and drop some disk drives, it looks a bit better. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

October 12th, 2004, 05:14 PM	#204
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	Lens formats: http://www.siliconimaging.com/Lens%2...%20formats.htm Wayne on the MT9M413: Sure we have discussed it. It is a tricky design - 100 parallel digital lines into an FPGA to be multiplexed down. Lots of data generated. Fairly high cost camera. Of course you could design it to run slow but all the same hardware has to be there except a simpler interface. We cover a similar high speed market with the much cheaper SI-640HF VGA at 250fps. If we found an OEM to justify it, we would do the design, but probably won't on speculation. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

October 13th, 2004, 11:56 AM	#207
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	Wayne on that Micron: I just checked my notes from when I looked into it. In moderate volume, the sensor alone is >$1K, my cost. Do you really want to buy a 720p camera for over $5K? It would be an excellent price for a 1.3Mpix 500fps camera.....but..... __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

October 7th, 2004, 05:34 PM	#197
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	Anyway, if I'm not too wrong, there is a Pentium-M with 533 MHZ bus too...

October 8th, 2004, 05:47 AM	#198
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	No, there's no 533Mhz bus yet for the Pentium-M, but the Dothan core is much better than the P4 Mobile chips (Netburst). I posted a link on the other threat about some guys talking over the Pentium-M vs. P4-M over at Andantech. BTW, the dual Via boards are absolutely no use if you're trying for high-speed 1080p because there's no 64-bit PCI-X slot for a frame-grabber. That and the Eden-N processors are horribly slow. Also the chipsets that you can get with the Penitum-M (855GME with 6300ESB) are much better.

October 8th, 2004, 06:15 PM	#199
Ben Syverson Major Player Join Date: Mar 2004 Location: chicago Posts: 434	Wayne wrote: "I am eagerly waiting to see what they do, have they shared with you what capture solutions they are using (capture software in particular)?" Sumix is writing all their own custom software, and are very receptive to cinema-oriented features...

October 8th, 2004, 08:01 PM	#200
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Thanks Ben, I long suspected that was the case. It is a pity that different companies didn't co-operate with Rob to establish his software as a defacto standard to shift with their cameras. Si has been great in co-operating with Rob though. The problem we now face is probably format. I would imagine everybody has his own, and their is not a standard one third party NLE's can plug into as things grow in the future. So they will all have to use convert to which ever compatible format they have on their list of file types, but still, it is very eartly days still lots of time left.

October 8th, 2004, 08:27 PM	#201
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Jason, I know that VIA has lost a lot of initative to the Israeli Pentium M, and the new low powered Transmeta Rob posted. But they are improving their game, unfortunately the next version (64bit and much processing power) seems to be in 2006, with no improved version inbetween but only a revision or so (I was hoping the increased processing functions would be in there, but still a chance, but that would be mnore than a revision). Still on the multilevel scheeme of things, the Pent M's I have seen in the past were very expensive (I don't know what has happened in price since). And dare I say it, the Pent M might be the best high end solution. I think with compressed camera data rate, that should be on camera buffered and not require PCI-X (I hope somebody has told them to include some preview ability/ stream in it), 2Ghz+ chips, onboard decompression engine (that hopefully will have some compatible arcitecture to partly assist decompress), and cheaper costs. Also the architechture of the latest series has improved a lot over the last, but not as good as Pent M. I think there is still hope for the old VIA on the bottom end at least. But I would like to know what Rob's latest sustained FPS (not shutter speed) is, that should give a good idea on the power required (probably with a possible 50-100% improvement possible throiugh more very specialised programming optimisation). So I hope to maybe see them on the lower end cameras, and I'm glad everybody has dropped the Pent 4 stuff for the Pent-M. Maybe sub 2Ghz Pent-M's will drop a lot in price now.

October 11th, 2004, 01:58 AM	#202
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Steve, a) The MT9M413, 500fps 10tap, 1.3Mpix 12 micron sensor with Truesnap (I think fillfactory also has similar one) is a big sensor, why haven't you built a camera on one of these, maybe even a cut down 24/30fps version. What would the light sensitivity compared to the other microns be like with these large pixels? b) It is interesting that when you look at it, the sensor chip manufacturers could build in memory circuits to buffer the image while a new one integrates (which this one does, analogue style), or even anneal a memory circuit with matching pad contact to the back of a sensor chip (which is done in all sorts of processing circuits nowadays). Then run a simpler industry standard high speed serial interface out of it. All the camera manufacturers have to do is to build analogue interface buffering components to attache real world interfaces to it. That would be a win win situation, as most processing is done internally (and the pad contact not requiring pin wiring) only pins for the external interface and power are required. As these, now, integrated circuits are required for all cameras, and are being manufactured at the most cost effective stage, cost is further reduced. As it is buffered, frame grabbers are only needed for the fastest capture devices (above 100MB/s data rate), so cheaper interfaces are available. Control and signals can be sent via interfaces, and control circuits would be on chip, so filtering compression dsp could eventually be added (which results in better use of interface for higher resolution/data rate). If compression upto 50:1 (as well as lossless) was included then many could be strung on network applications (remembering the standard security/production line markets first) with lower powered, cheaper computers, using their integrated video decompression DSP's to view it, and disk savings. Until they develop their own, there are many cell designs that could be integrated. ARM is the often used one (usually with memory) with many extra dsp like cells to the arm core, and I'm sure there are other dsp cells suitable for this sort of work. Even if cheap sensors ($200) landed up costing twice as much (the real cost difference should be negligible) that would be offset greatly by big savings in the rest of the work chain. Thanks Wayne.

October 13th, 2004, 05:22 AM	#205
Laurence Maher Join Date: May 2004 Location: denton, texas, usa Posts: 416	Can someone please give me a link to the summix web site? Thanks! P.S. How's that SI camera coming Steve?

October 13th, 2004, 11:44 AM	#206
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Didn't realise they didn't multiplex down on chip etc, that would add significant cost. Laurence, it's http://www.sumix.com/ there is a camera section under optical.

October 13th, 2004, 12:06 PM	#208
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Thats OK, some people want that sort of thing, I thought it had good machine vision advantages, but I can do without it myself. Somebody mentioned to me a price around $850. Well, "Collateral" is out today, and it was shot on Viper (and another camera) so I'm hopefully getting to see what the quality is like. Wayne.

October 13th, 2004, 02:12 PM	#209
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	Wayne, if you see Collateral on digital projection, that´s okay. I saw it in a 35mm transfer and the quality of it wasn´t very good. I would say it is because of the transfer quality.

October 13th, 2004, 11:18 PM	#210
Wayne Morellini Inner Circle Join Date: May 2003 Location: Australia Posts: 2,762	Typical, they spend all that money to use digital and then they skimp on the transfer for an a-list movie. I'll have to re-affirm light and range from the normal DVD. Saw Alien VS Predator last week, and the cinema had turned the brightness up, lots of grey shadows, so that didn't help it try to live up to its A grade trailer (note: a lot of the A grade bits of the movie hint hint, think the other 5 people in the audience agree). So stuffed again in less than 8 days.