sTeeVee
July 13th 09, 05:41 PM
On Jul 12, 4:56*pm, Krid > wrote:
> >> In audio what this normally means is
> >> performing a frequency domain multiplication (as you would for a
> >> filter, say) in the time domain. *The effect is the same, but it can be
> >> done in near real time, rather than having to gather up a heap of
> >> samples, perform an FFT, do the multiplication and then an inverse FFT
> >> to bring it back to time domain.
>
> > Actually a convoultion is a very time-consuming calculation.
> > Mathematically you're mainly right with your statement, but ironically
> > for speeding up a convolution algorithm, it is transformed into the
> > frequency domain (via FFT) where convolution turns into mere
> > multiplication. After this multiplication is done, the result has to be
> > transformed back to the time domain via inverse FFT.
>
> > So a convolution algorithm is *more* hassle than an FFT, because the
> > sped up version of a "fast convolution" actually requires two FFTs. ;-)
>
> Damn! One really shouldn't post, when in a hurry.
>
> Not sure, if I understood you right. What you describe as *not* necessary
> actually *is* the classical method of a "fast convolution".
>
> What other, more clever, convolution algorithms are there, without the
> need for FFTs?
>
> Bye,
> Krid.
>
> --www.dirk-music.de- Hide quoted text -
>
> - Show quoted text -
As I recall, John Chowning from Stamford created "digital" FM that was
workable for musical applications. He tried to shop it around. Alonso
from Dartmouth (later became the Synclavier from New England Digital)
was the first one to "grab" it. Then, the forward-thinking engineers
from Yamaha came in and signed an exclusive deal with Chowning, with
the exemption of NED, for the use of digital multioperator frequency
modulation synthesis. See the following link, however it does not
contain the relationship with NED
http://en.wikipedia.org/wiki/John_Chowning
Have fun,
sTEEVEE
> >> In audio what this normally means is
> >> performing a frequency domain multiplication (as you would for a
> >> filter, say) in the time domain. *The effect is the same, but it can be
> >> done in near real time, rather than having to gather up a heap of
> >> samples, perform an FFT, do the multiplication and then an inverse FFT
> >> to bring it back to time domain.
>
> > Actually a convoultion is a very time-consuming calculation.
> > Mathematically you're mainly right with your statement, but ironically
> > for speeding up a convolution algorithm, it is transformed into the
> > frequency domain (via FFT) where convolution turns into mere
> > multiplication. After this multiplication is done, the result has to be
> > transformed back to the time domain via inverse FFT.
>
> > So a convolution algorithm is *more* hassle than an FFT, because the
> > sped up version of a "fast convolution" actually requires two FFTs. ;-)
>
> Damn! One really shouldn't post, when in a hurry.
>
> Not sure, if I understood you right. What you describe as *not* necessary
> actually *is* the classical method of a "fast convolution".
>
> What other, more clever, convolution algorithms are there, without the
> need for FFTs?
>
> Bye,
> Krid.
>
> --www.dirk-music.de- Hide quoted text -
>
> - Show quoted text -
As I recall, John Chowning from Stamford created "digital" FM that was
workable for musical applications. He tried to shop it around. Alonso
from Dartmouth (later became the Synclavier from New England Digital)
was the first one to "grab" it. Then, the forward-thinking engineers
from Yamaha came in and signed an exclusive deal with Chowning, with
the exemption of NED, for the use of digital multioperator frequency
modulation synthesis. See the following link, however it does not
contain the relationship with NED
http://en.wikipedia.org/wiki/John_Chowning
Have fun,
sTEEVEE