[MPlayer-dev-eng] HW for open standard video and audio formats

[Timothy B. Terriberry]

Luca Barbato wrote:

> RC wrote:
>> 
>> You've said that before, but I haven't yet seen your list of patents
>> that you were supposedly putting together.
> 
> MDCT
> http://www.freepatentsonline.com/5867819.html
> 
> Other vorbis related patents http://www.mp3licensing.com/patents/index.html
> (pick whatever you want, those are a bit too broad IMHO)
> 
> Something else that come up from a quick search
> 
> http://www.freepatentsonline.com/6882685.html
> 
> 
> Feel free to have a look yourself....

I'm sorry, I'm going to have to be the one to ask you to stop spreading
blatant lies.

I don't know nearly as much about the Vorbis side of things, but I do
know that Real sponsored a patent search against the MPEG patent lists,
done by actual patent attorneys, and found that Vorbis did not infringe.

Does this mean that there is no chance Vorbis infringes on any patents?
Absolutely not. I fully agree with you that the current US patent system
makes such a determination impossible. However, numerous major game
companies ship games using Vorbis (including Microsoft), as well a
number of consumer electronic manufacturers. Anyone else using Vorbis is
at no more risk of being sued than they are. The same cannot be said of
formats which not only have patents specifically designed to cover
aspects of their algorithms, but a licensing authority set up for the
express purpose of collecting royalties for use of that codec.

On the Theora side: Theora does _not_ use the exact same algorithms as
MPEG1/2. It does use DCT+MC like MPEG1/2 does, but so does H.264, and no
one in their right mind would claim those algorithms are "exactly the
same". Half-pel motion interpolation is done differently, coded block
flags and macro block modes are done differently, DCT coefficients are
quantized differently, and tokenized in a completely different manner,
and stored in the bitstream differently as well.

VP3, on which Theora is largely based, _is_ in fact covered by patents.
It was distributed as a commercial product by On2 Technologies before
they donated to Xiph. However, it comes with an irrevocable license to
any patents On2 might hold that cover it. So again, you assume no more
risk of liability than On2 ever did.

Has Xiph released some binding legal guarantee that Vorbis and Theora do
not infringe on any patents? Of course not, because under the current US
legal system, such a thing is impossible. We have always been very open
about our inability to make such a declaration. However, we have done
everything our lawyers tell us we can do, and even were we a large
multinational corporation, there is little else that could be done.


So, to address your specific claims:

The MP3 patents were reviewed by an actual patent lawyer, who gave us an
opinion that Vorbis was not infringing. Feel free to fund your own legal
opinion if this is not good enough for you.

I'd question if you even read the 6,882,685 patent. It covers the very
specific 4x4 multiplierless DCT used in H.264, and was authored by the
people who proposed the algorithm to the JVT. What relation this has to
Theora, I have not the slightest idea.

In any case, none of this has anything to do with mplayer development.
It has all been discussed before and in more detail on the xiph.org
mailing lists. Please search them if you have further questsions, and
please stop spreading baseless FUD. I now consider this portion of the
conversation closed, and will not respond further on the subject.


Now, to respond to the original poster:
Some work has already been done for hardware Vorbis and Theora
implementations.

For Vorbis, there is the "Ogg-on-a-Chip" project, which implemented a
hardware decoder. See http://oggonachip.sourceforge.net/ for more
details. For an emebedded software decoder, there is of course the
Tremor library.

For Theora, Andrey Filippov has implemented an encoder using FPGAs for
the Ephel 333 camera. It doesn't support motion compensation, just
simple prediction with (0,0) motion vectors, but can stream Theora at
1280x1024 at over 30 fps. We actually had to do some optimization work
to get a decoder that was fast enough to play them back (only a small
amount of assembly optimization work has been done for Theora so far).
See http://www.linuxdevices.com/articles/AT3888835064.html for further
details.