[music-dsp] Decimation by two

[Muon Software Ltd - Dave]

> Poly-phase all-pass filters are very efficient for this.
> Have a look at
> http://www.cmsa.wmin.ac.uk/~artur/Poly.html

I got it up and running now and have passed the code onto Bram to post into
the archive (but you can mail me if you're desperate to get a copy before it
gets posted).

I have to say that this class of filters works incredibly well for Fs/4
bandlimiting! my own experiments show that I can get a stereo total
brickwall up and running realtime for the cost of around 6% of my 466
processor (in an unoptimised VST plugin I built for testing).

> Since each of the two all-pass filters is a function of z^(-2), rather
than
> z^(-1), you can downsample before the all-pass filtering.

I'm having trouble with this next bit - I saw something in the paper
published on the site you mentioned about flipping even samples into the
A-filter cascade and odd samples into the B-filter cascade, thus processing
only half the data. I couldn't really get my around this optimisation, and
my attempts to implement it crumbled - can anyone offer further insight into
this?

> And there's even have a Java applet that gives you the filter coefficients
> from the desired filter response!

Without this applet I'd never have got these things off the ground - the
coefficient calculations are outrageously difficult!!

> koen
Who's da man, You da man!


dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
http://shoko.calarts.edu/musicdsp/