Patents with partitioned convolution

[EvilDragon]

I was also wondering if I should write a program that counts from 0 to 2^800000. It would therefore produce any sequence of bytes possible for 100 kilobytes of memory. This would naturally include all algorithms that can be stuffed into 100k, including pretty much all those that have ever been and ever will be patented.

Hahaha, that's... that's.... PURE GENIUS!

[llze]

Patents can only be made for inventions that are new. New means everything that is not the current state of the technic. So the theoretical frame of all calculations itself is not new, its just the current range to work in. Also it is not a new scientific discovery or new mathematical discovery.

[Deisss]

I don't exactly know this by myself.

But some time ago I got a definitely genius software to patent (I start a company for it), at least in europe it appears that patent on code almost never works (it's close to impossible to prove anything), so you patent the idea of the code instead... And this is what happens for this piece of code, we patent it's underlying idea instead of the code itself.

This is my 2 cents ^^

[llze]

yea, quite possible..if it is based on your inventional process, has a commercial application and is developed in ways that are not obvious. Sure this leaves a lot of room for arguments.

[Miles1981]

I was wondering that myself…

I was also wondering if I should write a program that counts from 0 to 2^800000. It would therefore produce any sequence of bytes possible for 100 kilobytes of memory. This would naturally include all algorithms that can be stuffed into 100k, including pretty much all those that have ever been and ever will be patented.

That won't work to have a patent on all algorithms

The matter is tackled on the claims of a patent. If the patent consists of 4 elements, you have to implement these 4 elements to violate the patent. Sometimes, there are different claims in the patent, and each of them has several elements. They need all to be present to violate a claim.
The problem is more subtle IMHO: you can't know which patent on convolution exists or is being processed (you would spend more time reading patent than coding), so there is always a risk of someone saying that you have to violate their patent and they can sue you, hoping you would settle out of court.

[hugoderwolf]

Is the burden of proof on the patent holder (to prove infringement) or the developer (that none occurred)?

The burden of proof is of course on the patent holder. That also means a patent is worth nothing if you can't prove its infringement outside the black box. Regarding the convolution patent: if it is really only the optimization algorithm for choosing the block sizes that's actually covered by the patent, they'll have a very hard time proving such a kind of infringement.

If you talk to a patent lawyer, one of the first questions he'll ask you is "How will you find out someone is infringing your patent?". Along with "Can a capable guy find out how your thing works by looking at the black box?" If not, you shouldn't give him the information for free by publishing it in a patent.

[valhallasound]

IIRC, the original Lake DSP patent was granted in 1994 or so. It would be worth checking to see if it has expired (it has been 21 years, and US patents are limited to 20 years).

Sean Costello

[Ivan_C]

Is it this one ?

https://www.google.com/patents/US6574649

Looks like it is possible to update it a few times. If the limit is 20 years in the US, then this one is valid up to 2018 at least

[cron]

edit: Info didn't bear out upon refreshing my memory on a few things.

[valhallasound]

Is it this one ?

https://www.google.com/patents/US6574649

Looks like it is possible to update it a few times. If the limit is 20 years in the US, then this one is valid up to 2018 at least

I was thinking of the original Lake DSP patent:

http://www.google.com/patents/US5502747?cl=en

Filed in 1993, granted in 1996. I just Googled how to determine the expiration date of a patent, and found this:

Utility Patents and Plant Patents based on applications which were pending on June 8, 1995 (actual filing date, not priority date), and any Utility or Plant Patent which was issued on or after June 7, 1978 and had not expired before June 8, 1995, have a term measured by the longer of 17 years from the date of issue or 20 years from the date of US filing* of the earliest non-provisional** application in its chain of parentage.

So, it looks like this patent expired sometime in 2013. More recent Lake DSP patents would cover, well, whatever was talked about in those patents, that wasn't part of this original patent.

I'm old enough to remember all the discussion on music-dsp about this patent, as people were starting to implement the ideas in the Bill Gardner paper, but were stopped by Lake DSP's fairly aggressive behavior. It turns out that the passage of time has resolved this issue, at least from my understanding of how patents work.

Sean Costello

[Ivan_C]

That would be great if the patent has really expired !

[Miles1981]

It would have been even greater if the patent was not allowed in the first place. It's freakishly trivial.

[Richard_Synapse]

I think there is enough prior art to perform basic convolution, the Kulp paper is rather clear. If you plan to do something fancier than that, be sure to take a look at the (non-expired) patents about convolution, as there is quite a few, iirc.

Richard

[Ivan_C]

Well, if I want to perform convolution on quite large impulse responses, it would be difficult to do so without partitioned convolution algorithms and with acceptable performance unfortunately...

[Richard_Synapse]

This is what I meant, basic partitioned convolution without special tricks. Get yourself a copy of the Kulp paper and see for yourself, although I'm not even sure this is necessary now if the Lake patent has expired as Sean suggests.

Richard