[sdiy] Multi-output comb filter?

rsdio at audiobanshee.com rsdio at audiobanshee.com
Mon May 21 06:24:17 CEST 2018

The only process that I’m aware of where substantial work is shared instead of duplicated is the FFT, but you excluded that in your opening statement.

It’s a bit distracting to attempt to think of a comb filter as having “outputs" that are “mixed" together. You’ll get a lot further realizing that a comb filter has only a single input and a single output, where the constructive and destructive interference ends up favoring some frequencies and penalizing others. It’s not like a typical filter where you can control bandwidth or slope; although feedback does allow further enhancement of the peaks and valleys, you can’t really move any part of the total response without moving everything else.

In contrast, a state variable filter does have multiple outputs where you can choose how to mix them together. In some sense, the SVF shares some work because the three outputs depend upon each other to work. Each of the outputs is a combination of the others in a cyclic relationship. Again, you can’t drastically change one without also affecting the others.


On May 20, 2018, at 8:08 PM, cheater00 cheater00 <cheater00 at gmail.com> wrote:
> Note I'm not adhering to any specific definition of comb filter. I just wonder if there's a medium between a typical, delay based comb filter (you get all outputs mixed together, but work is very little) and a bank of filters (you get each output separately, but a lot of work is duplicated).
> On Tue, 15 May 2018 04:06 , <rsdio at audiobanshee.com> wrote:
> On May 14, 2018, at 11:02 AM, Gordonjcp <gordonjcp at gjcp.net> wrote:
> > On Mon, May 14, 2018 at 02:03:13PM +0200, cheater00 cheater00 wrote:
> >> Hi, I was wondering if anyone knew of a (digital? analog?) comb filter
> >> where the computation is done once, but you get separate outputs out of
> >> each peak. It's fine if it's just "present" or "not present" according to
> >> some threshold. I am looking for implementations in the time domain, not
> >> FFT based.
> > 
> > You don't get separate outputs of each peak.  A "comb filter" isn't
> > really a filter in the sense that say a bandpass filter is.  It's a very
> > short delay line, in which frequencies that are delayed by an odd number
> > of half-cycles cancel out and frequencies that are delayed by a whole
> > number of half-cycles add, giving the "comb teeth" response.
> What Gordon said!
> The beauty of the comb filter is that it’s cheap to implement (“just” a delay) and you get all of those peaks from one process.
> Creating what you ask for - separate outputs for each peak - would be a totally different process. Most importantly, it wouldn’t be cheap at all.
> Brian

More information about the Synth-diy mailing list