Hi!

I have been playing with mixers, creating some feedback loops ... those of you who have tried probably know the huge variety of sounds you can get with this method.

Now I am trying to do the same with Pd. For the moment I have implemented a simple model of a mixer in feedback :

[delread~] -> [gain] -> [filter bank] -> [clip -1 1] -> [delwrite~]
-> [dac~]

In my filterbank I have 4 band-pass filter in parallel.

I do have some feedback sounds, but it always sounds the same. Cannot even get close to what I get with my mixer. Any ideas how to enhance this a bit?

a simple mixer is far more complex than this patch, it's a full bag of non-linearities, phase distortion, tiny delays, temperature, behaviour when overdriven, etc.
For example there is no such thing as a [clip~] in a mixer : it's closer to a [tanh~].

yeah ... I know it's not that simple
That's why I'm asking if any ideas to make this more interesting!!!

You might also want a blocksize smaller than 64 to get higher pitches.

See 3.audio.examples/G04.control.blocksize.pd in the help browser.

I always put in some effects like reverbs, filter-delays, allpass filters, becaus this effects have already buil in functionalities that make the sound more interesting

Some years ago I've experimented a lot with feedback within Pd, and found that a rather specific non-linear mapping function was needed to make it work. I've never been able to realize this with regular Pd(-extended) objects, so I ended up with a new object [krunch~]. It is in the recent version of Slice//Jockey on http://www.katjaas.nl/slicejockey/slicejockey.html. Without feedback it sounds not much different from [cyclone/overdrive~], but in a feedback loop together with a resonating filter (allpass, lopass, bandpass or hipass) it works best. You can try it out in Slice//Jockey by setting delay, filter and krunch to high gain values. The loop needs a small input from the beat slicer to start.

The non-linear function is a modified arctangent. Source code is included with Slice//Jockey, I'm afraid that's the only documentation in this case.

Katja

I think such "instantaneous" feedback loops are almost exclusively the domain of analog electronics, unless you want to solve equations iteratively (like a circuit simulator; Pd isn't going to do this and it's computationally expensive anyway). For anything that uses a signal flow graph, the smallest feedback delay you can get is one sample (and in Pd you can only get that by setting the block size to 1). This delay makes a big difference. Higher sample rates will help, but it still won't be the same as mixer feedback.

In other words, Pd can only do y[n] = f(y[n-1]). Mixer feedback is basically a nonlinear differential equation. A good discrete time simulation would require y[n] = f(y[n]). You can only do that by solving iteratively.

If you're trying to do something like a digital reverb inside a feedback loop, Pd can do that just fine.

One thing to try, anyway, is to distribute multiple [tanh~] (or other) nonlinearities at different points. You could put one after each filter, for example.