Subject: [linux-audio-dev] Patents again
From: Juhana Sadeharju (kouhia_AT_nic.funet.fi)
Date: ti maalis 07 2000 - 10:05:30 EST
Hello. Yet again found an irritating patent:
US Patent 5317104, David R. Frost, E-mu Systems, Filed: Dec 1992
(continuation of the patent filed at Nov 1991).
The patent seems to patent the generation of reverberation impulse
responses from white noise by filtering.
"The simulated impulse response is generated by combining frequency bands
of white noise. The power of each band decays exponentially in time.
The time constants are generated from the characteristics of a real or
imaginary listening space."
The figure in the patent shows following system:
white noise --> linear phase band split --> envelopes for each band
--> exponential mix --> impulse response
Lets read what Dr. J. A. Moorer wrote in his classical About This
Reverberation Business (CMJ, 3(2) 1979) paper:
"We made a somewhat striking discovery in simulating natural-sounding
impulse responses. While digitizing the impulse responses from concert
halls around the world, we kept noticing that the impulse responses
in the finest concert halls sounded remarkably similar to white noise
with an exponential amplitude envelope. To test this observation, we
generated synthetic impulse responses by shaping unit-variance
Gaussian pseudorandom sequences with an exponential of the desired length.
[ ... ] With this method of creating synthetic impulse responses, one
can "tailor" the characteristics to one's needs. By selectively filtering
the impulse response before convolution, we can get any desired rolloff
rates at various frequencies. It was the most natural-sounding of all
the techniques we tried, [ ... ] Stereo can then be produced by simply
generating two synthetic impulse-response sequences and calculating
the left and right ears separately. [ ... ] Frequency dependence in the
decay can easily be introduced by filtering the synthetic impulse response
before doing convolution."
What is the difference between two methods? Nothing, Moorer just don't
tell the obvious facts on how the filtering can be done with well-known
methods. Band-splitting is well-known technique (phase vocoder) to alter
individual frequency bands in time domain. Even Moorer himself has
pioneered such systems in his phase vocoder paper(s) and wrote a nice
paper on linear-phase bandsplitting done at Lucasfilm at 1986.
What else? Can we now follow Moorer's advices and implement the
filtering a slightly different way? No! ...if we believe the patent:
"While the invention will be described in conjuction with the preferred
embodiments, it will be understood that they are not intended to limit
the invention to those embodiments. On the contrary, the invention is
intended to cover alternatives, modifications and equivalents, which may
be included within the spirit and scope of the invention as defined
by the appended claims."
Now who _really_ made the simple modifications? I really wish patent
office could itself follow this kind of idiotic "agreements" written
in these patents and not give patents to everyone who asks. But money
talks!
Sure we can implement the Moorer's system and even with the linear-phase
bandsplitting because it is only an obvious tool to perform the
time-varying filterings (e.g., phase vocoder). It has used extensively
for any filtering.
I so far thought that David Rossum at E-mu could set up high standards
to them but I sure were wrong. Discusting.
-*-
A side note: both Moorer's system and the patented system are based on
substractive filtering of white noise. A totally different concept is
based on additive synthesis where we sum up many band limited time-varying
noise signals. This kind of systems are used extensively in computer
graphics for making textures. This different method doesn't involve
the linear-phase filtering which is essential in the patented system.
-*-
Now that I mentioned again the Lucasfilm system: It assembled and hardwired
the assembler opcodes as one executable to dsp processors. There is also
ptolemy which hardwires flow system to dsp processors. If the same is done
in Linux (say) (from application to audio engine), does it violate a patent
(199* something, I will check that later). What an invent!
We could sure(?) circumvent it with double (or more) processor systems by
keeping one (or more) processor as dedicated DSP processor.
Juhana
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:23:28 EST