[Theora-dev] FPGA implementation

Andrey Filippov theora at elphel.com
Sat Nov 27 22:29:32 PST 2004

>> Right now, what I'm thinking about is a video camera with hardware support
>> for theora compression that could be USB connected to a laptop
computer, or
>> could be built into a laptop/handheld computer.  This video camera
would be
>> used for both videoconferencing and as a camcorder.

Elphel 333 (as well as current model 313) uses Ethernet, not USB. Cameras
also support IEEE802.3af (PoE) so they can be connected with a single
data/power cable (standard CAT5) up to 100m from the power supply. Camera
needs approximately 3W of power (maybe Theora encoder will need a little
more) - it is more than allowed to drain from a single USB connector but
I'm still thinking of making a power converter for a laptop so it will be
connected to both Ethernet port and USB (just to get power).
As for USB - I'm only planning to add master USB port to simplify adding
stuff to the camera such as 801.11 USB adapter.
I'm also planning to use Gigabit Ethernet in the future camera - both for
production-quality video (such as 24p) and fast (1megapixel x 500fps or
more) cameras that perform on the fly compression and can provide
continuous data not limited to the internal memory. What stopped me so far
is that I could not find any Gigabit PHY with the documentation available
without NDA (even pinout seems secret so it will be impossible to have
open circuit diagrams).
>> How small could such a camera be?  What would be the costs per device
for the
>> materials, assembly, production, packaging, sales and support in a
>> of 100,000?  As long as I am pretending, what would be the costs per>
>> in a quantity of 1 million?
> I'm sure Andrey is better equipped to answer all this, but I'd guess
> something the size of an isight is reasonable. probably bigger if you
> have to attach a battery.

Both model 313 and 333 cameras have the main board 3.5" x 1.5" and the
square sensor board of 1.35"x1.35" (sensor board is separate to simplify
usage of different sensors). And they are designed to be network cameras
with external power supply, not the battery. But being the free design
they are subject to the same fate as FOSS - you never know what others
will use it for.

As for that high volume pricing - I don't have any figures. What I did on
my part - tried to simplify the circuitry, make it really compact - it
reduces signal integrity problems so I so just a 4-layer board is

> The limiting factor on cost in those quantities is the optics, so it
> boils down to how good an image quality you want and features like
> focus and zoom.

Yes, optics is a limiting factor here. Cameras use standard C/CS-mount
lenses but there are virtually no high resolution C/CS-mount lenses
probably just because nobody needed them so far - for analog video cameras
resolution was more than enough. I did not have any problems with
1280x1024 sensors, but 2048x1536 needs either very expensive optics
(~$1000) or can work with partially closed iris on regular ones. So we'll
probably have to order custom optics for the cameras - it is not really so
hard to make them - anyway cheap consumer cameras handle 3 megapixels

> I suspect there's not much difference between 100k an 1M
> units. At 1M you would fab a custom encoder chip rather than use an
> fgpa, which will cut the price of that part by a half or 3/4, but that
> won't be the only expensive component. Maybe if you put all the other
> electronics in the same custom chip?

I don't think it is a solution and I'm going to stay with the FPGAs. For
me as a manufacturer it dramatically reduces time to market and extends
the product life time. Model 313 was first released in December 2002 -
since that "hardware" algorithms (FPGA code) were improved many times,
bugs were found and fixed, new features were added. To say nothing that
initially I was able to do all the schematic, FPGA and software design
myself with minimal financial resources.

On the other hand - I'm trying to promote FPGA code design among software
developers. And the camera is not only a "camera" - it can be used as a
developer platform to play with/hack into. And it is not just a sand box
"development board" - it is still a network camera (with sensor, optics,
network, CPU running GNU/Linux) with all the internals GPL-ed. So without
touching a soldering iron you can make a virtually new useful hardware of

> If they had a few million in venture capital they certainly could. To
> sell a million units will require as much sales and marketing as
> engineering.

I do not have investors and try to finance projects with the income from
other ones. And it does work - maybe not as fast as with additional
resources - but leaves more freedom.


More information about the Theora-dev mailing list