4:4:4 10bit single CMOS HD project

[Filip Kovcin]

with your permission, i wish to say a few words about big company HD and DIY HD.
i'm working in poland as film/tv director, and by chance also as a co-owner of film/video rental company.

on daily basis i'm working with film/tv/comercial crews, producers etc. we have dv,imx and HD cameras. (f900 and f750p)

yes - for now almost every producer will use $$$ HD camera. but in the nearest future? if it's cheaper and delivers the same quality? the producers really know the value of the money, so if it's worth - they will use it. even if it's not from known/respectable manufacturer(yet!).

but i beleive in this project - because it's fresh and pioneering.
how many simple guys on this planet works on DIY hd camera? not so many.
how many big $$$ companies are working on HD cameras? not so many.

you see - this is the ELITE! never mind what will happened next. victory or failure. let's try. because it's fun. and in my opinion possible.

see what jeff kreiness did - he is not big $$$$ company. he is just simple "with many quotas" guy. and he did it. and that's why we adore him.

sony is still looking for proper HD solutions, DALSA, ARRI is still looking, THOMSON, PHILIPS, PANASONIC, PANAVISON, OLYMPUS, JVC, even KODAK... is also around in this HD world, JEFF is still working... and...
the guys from DV comunity are also working on this very thing.

this is not so bad company, dont you think?

and hey - do you think that the guys from big $$$$ companies are sleeping? no - i can bet that they are also reading our posts - they KNOW!

so, at least - TRY!

OBIN and others - keep rockin'!

filip

[Jason Rodriguez]

Hey Obin,

Thanks for the reply.

And BTW, I'm not telling anyone to stop. Yes, please keep going, it's just that as of right now I felt we're getting the discussion polluted with some pretty advanced concepts/ideas when we haven't even got the basics-i.e., really good capture software and a really good bayer algorithm (you're not going to get any good greenscreen keys with a bad bayer algorithm).

Let's get this discussion back on making a useable product AS-IS, meaning I can buy an industrial camera, hook it up to a PC (maybe even a small PC), and have the real possibility of actually using it on a professional shoot, such as something on a set where I can afford to have a camera tied to a PC. Right now I love the potential for what this footage can look like, but frankly it's still not there yet for me to start bragging about this stuff to everyone else, especially a non-tech-savy producer who only understands the end product and is shelling out the $$$$'s for the production. He'll start asking questions about the noise banding, why it's so dark or clipped, the "dots" on the edges, etc.

So I'm glad the excitement's there for future production, but lets keep this right now to what it is and what it can be here in the next couple months-a very useable system where we can hook an industrial-style camera to a PC and get uncompressed-HD quality images for cheap $$'s.

How many Kinetta's are out in production right now? Any test productions using it?
What's the price? Delivery date?

The Kinetta is coming, as of right now I don't believe the Altasens chip it's using is in mass production yet, after all SI doesn't even have their Altasens camera for sale yet. Not absolutely sure on pricing, but very reliable sources have told me a good deal less than the Varicam. And yes, the NAB model was working (without an umbellical chord under the table :-), Jeff was shopping the camera around at the conference, and I got to play with it myself (although I was in a lounge/bar at the Stardust when I got to handle it, and there was no monitor-out available, so I didn't actually see any footage if that's what you're wondering), so it is a very real product, and it has many high-end people (i.e., they have some clout) in the production community pretty excited about what it can do. It should be shipping by this fall, say around October, November.

But this is a project I've known about for almost a year now, and the electronics designer on the team, Martin Snashall, is top notch and totally on-the-ball about what needs to be done with this camera. The guy's created some of the most ground-breaking stuff back at Abekas, such as the A64 and A84, and after picking his brain for hours both in email and at NAB, he knows what he's talking about whether it's demosaicing, color correction, etc. On top of that he's a top-notch FPGA programmer and has been designing video circuitry for years. At the helm you have Jeff who's been an award-winning filmmaker at Sundance, and has shot tons of films for other documentary filmmakers like D.A. Pennebaker (one of the fathers of the cinema-verite documentary), so he knows what needs to go into a camera design.

[Obin Olson]

I wonder if Jeff knows about what we are trying to do ?

I just rented some equipment to an Indy gig - it's this type of gig that would really shine with a homemade HD camera - they are using the XL1 - not to diss a good SD NTSC camera but that is NOT the right tool for a movie - yet I see people use it all the time because they can't afford better ...would they rent the "homemade" Hd camera? yes AND pay a little more then the xl1 for rental.....I think that is the best market we have.


- Indy filmmaker -

[Jason Rodriguez]

I think these cameras, with good software to make them run, will be a very nice replacement for the cheap way-to-compressed HDV cameras that are starting to hit the market.

I think that's where the sweet spot will be. And here's the thing-we don't need to make all the parts.

Here's the way I see it for this market:

1) We have the cameras already. We have the PC's to record them on. Right now we need good software. With good software we can record on commodity equipment and keep costs down, yet have a very high-quality product that will beat the pants off of DV and HDV, even HDCAM (but not HDCAM-SR, right now as a tape format it has many, many advantages in ease-of-use, especially for high-end production, and very low compression-as low as 2:1).

2) After having good software, people will see that these little cameras can really work. That produces sales. Sales produce $$$.

3) Other third-party manufacturers see that there's money to be made in these little cameras, and that there's some nice holes that can be filled. Like maybe even easier-to-use software. Or smaller hard-disk recorders. Or maybe even a company is making so much money on these little industrial cameras that they are willing to fork over the $100K+ necessary to develop their own little camera that combines the whole widget.

Right now I think we're in a little bit of a chicken-egg thing. Do you make the nicest camera you can and they will come, or do you make the little cameras and get a nice installed base that produces cash and whoos other third-party after-market manufacturers into the fold? Because Kinetta's got some cash for development, they've taken the first approach, and they're doing it all-out. We don't have the capital investment, and so we need to take the second approach-create a steady income market through a saleable product by introducing the one widget that is preventing this whole thing from being useable-good, easy-to-use, non-buggy software. That will make these little cameras possible to get more money to create the big stuff or at the very least get people thinking in another direction beyond making more HDV cameras.

If somebody sees that making the back-end instead of dragging around a PC will make them money, they will do it. But nobody's going to do anything if they feel that they're sinking their money into a product that's going nowhere, and/or won't work. When this stuff is rock-solid on a PC and works, then spending 5K on an Altasens that will go up to 60fps will be cheap, cheap, cheap, instead of being afraid that you're basically losing $5K on unproven, unsupported technology that's not designed for motion-picture production.

[Juan M. M. Fiebelkorn]

@ Rob

Don't know if I'm the right person, but from what I've been reading if you get a camera with camera link, your board would need a LVDS capable interface.If you go the GEthernet way, you'll need a board with Gigabit interface.Anyway I'll wait for Steve's answer.

Here I've found a nice Frame grabber with Camera link which has a Virtex II FPGA included and is supposed to be programable.
I don't know its pricing, but I'm sure will be above 1k.


http://www.datacube.com/Product/Data...Aprocessor.htm

[Rob Lohman]

Juan: I think the idea is to NOT have it on a PCI board? Or is that
just for programming?

What is LVDS?

All of this stuff is quite confusing and I'm hoping someone can
write up a small article which summarizes what FPGA is and how
to use it.

I've indeed been wondering myself how we hook up the camera
to something like this. Either directly or through GigE or CameraLink.
Not to mention viewfinder out and harddisk access....

[Juan M. M. Fiebelkorn]

@ Rob L.

Well, I didn't know sorry.
I'm pointing to PCI boards because of its simplicity to work with.
And because a Mini-itx with Eden processor (around $200) coupled with a PCI FPGA board with camera link has all you are asking for!! :).
VGA output (to connect a LCD display or viewfinder), IDE interfaces for attaching disks,simplicity of programing for Graphical Interfaces,etc,etc.
All in a relative small off-the-shelf package.

LVDS is what Camera link uses internally (If I'm not wrong).
Low Voltage Diferential Signal.
So if you have a board with support for LVDS, you could work with camera link, I guess.
Hope this helps.

Google search about what a FPGA is:


http://en.wikipedia.org/wiki/FPGA

[Rob Lohman]

No apologies necessary Juan. I was just wondering if PCI was
an option or not etc. Sounds pretty interesting at all.

So what does this Eden processor run? A custom "OS" or
something like unix?

[Juan M. M. Fiebelkorn]

Eden is X86.It is a normal PC.It runs any OS a PC Intel or Athlon runs.
The nice thing about Eden is that it only needs 7 watts for a 1 GHZ cpu (passive cooler, no fan!), against 100+ watts for a P4 and around 70 for an Athlon XP.

http://www.mini-itx.com/faq.asp

Here is something I think would become very useful, a Linux Bios replacement that lets you start an operating Linux Kernel without needing a hard disk.


The LinuxBIOS Project

LinuxBIOS is an Open Source project aimed at replacing the normal BIOS with a little bit of hardware initialization and a compressed Linux kernel that can be booted from a cold start. Other beneficial consequences of using LinuxBIOS include needing only two working motors to boot (cpu fan and power supply), fast boot times (current fastest is 3 seconds), and freedom from proprietary (buggy) BIOS code, to name a few. These secondary benefits are numerous and have helped gain support from many vendors in both the high performance computing as well as embedded computing markets.



http://www.linuxbios.org/

[Rob Scott]

Jason wrote:
I've got a great idea. Why don't we see how to get Obin's camera working as good as it can (new bayer de-mosaic filters, recording software, good lens, maybe an Altasens chip) before we jump onto this FPGA stuff. I know it sounds so nice, but I can guarantee you that you're not going to beat the Kinetta, Panavision, Arri, etc. with this home-made stuff.

Er ... dude, that's what I'm doing :-)

You hit the nail on the head, and it's exactly what I've been saying from day one. I don't see the need to add real-time HDI, FPGA (yet), and all this other stuff we've been talking about. If we try to compete with the Kinetta, we'd have to come up with a manufacturable product and we'd end up at the same price point. I like Kinetta and I have no desire to compete with them.

I think these cameras, with good software to make them run, will be a very nice replacement for the cheap way-to-compressed HDV cameras that are starting to hit the market.

Again, I think you hit the nail on the head. There is a sweet spot here that we can fill. As I've said before, an ObscuraCam (heh :-) is not for everyone; if you need the professional workflow, don't use an ObscuraCam.

[Steve Nordhauser]

Camera signals:
Internally the camera uses 3.3V levels for all the data. These are very succeptable to noise - they can go maybe 3" at high speed. LVDS or channel link (generic camera link) are differential signals - like RS-232. The idea is that you twist two wires together and drive them in opposite voltages. The receiver looks at the difference between them. Any noise on one is on the other since they are twisted together and does not influence the difference. Here is a link to the spec:

http://www.machinevisiononline.org/p...ink5_Specs.pdf

To use an external board for FPGA development, you would have to add a channel link transceiver to the board to get back to parallel digital lines.

This means that Juan is correct that the Datacube frame grabber would be an easy development plan. It starts at $1776 and rapidly moves up to about $4K with good sized FPGAs and memory. Plus $5K for their development environment.

[Rob Scott]

Steve Nordhauser wrote:
... Juan is correct that the Datacube frame grabber would be an easy development plan.

So, we could develop using an existing CameraLink camera and one of these boards; and then move the design to a custom board which connects the sensor directly to the FPGA using LVDS?

Presumably this custom board would have Gigabit Ethernet output. You could then output fully Bayer-filtered, 4:4:4, 10-bit lossy-but-good images at a very high frame rate.

[Steve Nordhauser]

Processors in the data path:
The Eden might be a nice tool for the uncompressed recording method if it can supply a PCI bus with RAID. Remember that for any real-time processing (compression, display), speed is king. While 7W makes it easy to battery pack, 1GHz vs 3GHz and a 133MHz vs 800MHz FSB will make a big difference.

FPGA development:
The typical path is to define your hardware in terms of schematic or HDL (hardware description language VHDL or Verilog), synthesize it (compile, kind of), simulate it, route it, do a timing simulation and test on hardware. I don't know much about the quality or cost of some of the C translation tools like Handel but here are some numbers to chew on:
1280x720@24fps = 37nsec per clock
1920x1080@24fps = 35nsec per clock - 24 bit transfers

I have found in the past (maybe they are better now) that FPGAs are alot like early assembler programming. If timing isn't critical, you can use higher level tools. In this design, the synchronous steps (what you do between register sets) must be less than those clock rates since the data is continuous. Real-time video is a tough first project.

Could someone post our Russian friend's website with his design? I'd like to take another look at it.

[Steve Nordhauser]

Rob:
Within the same family of FPGAs, designs are pretty portable. You need the same peripherals (I/Os, memory, clocks) but yes. What I would suggest if someone got that far is that we put the FPGA into the camera with a direct digital connection to the video path - no need for LVDS. Our gigabit interface has a Virtex II in it. Now you have your gigE output. We have discussed doing something simple in there - data packing or RLE since adjacent values in video tend to be similar if you separate the color planes.

Be sure to watch the tool prices. Xlinix gives away their low end tools but the ones you need for larger designs were >$5K.

It is your business, but I would suggest getting a workable tool chain going based on software first - this can get complicated and expensive fast. As it has been said, the final parts costs are low so it is attractive but the development costs and time are huge.

[Rob Scott]

Steve Nordhauser wrote:
It is your business, but I would suggest getting a workable tool chain going based on software first

No worries, I just find it very interesting. I won't get sidetracked onto this until I get the software-only system working. (And possibly never.)