Preventing unpredictable howl-round when feeding back into a delay

So I've made this patch which feeds short pulses of noise into a delay, the output is then fed back into the delay again, with some attenuation. It works pretty well, but every now and then it just blows up in a noisy howl-round squeal and stops the audio running.

Is there any strategy for preventing this? I was thinking of just measuring the output level (not sure how?) and if it goes above a certain level drop some of the feedback parameters momentarily to stop the feedback. I think I just described something like a compressor? But not sure that's the right way to do it.

I wanted to play with an idea I saw at a synth festival in Bristol at the weekend, there was a demo of an Echon 6 which used bucket brigade delay chips to make plucked string sounds from short pulses from white noise. It goes through the BBD delay and gets sent back into itself with some attenuation, but the delay is so short that the pulses of white noise make a tone.

This recreates the general idea (but doesn't behave like I would guess analog BBDs do) but I'd like to improve it. But right now it very unpredictable and keeps blowing up. Anyone got any suggestions?

☝️ This is a macro that's in the main ensemble below (twice) called delay:

BBD-strings.ens

Find more posts tagged with

Comments

silvermoats

Besides making sure your feedback control is always <1, the typical thing to do here seems to be putting a saturator in the feedback chain so the signal never exceed a certain level - less extreme than a hard clipper but more reliable than a compressor. I think there are a few different ways around it though, look up Karplus Strong synthesis for some examples maybe?

colB

Compressor works really well. It's a loooong time since I built a feedback delay that doesn't include some form of compression. Saturation is really useful, but I like that more for texture rather than for preventing the feedback explosion.

The trick with compression is tuning the balance of input vs feedback in the side chain, and also tuning the balance between compressing the feedback level vs compressing the input level. So it performs feedback limiting, and also ducking, making multi-layered looping delay effects easier to achieve…

colB

I couldn't get it to blow up, but it's probably one of these rare things that's still possible due to numerical error resulting in just grater than 1, then boom…

Here is one way to create a simple feedback compressor. You can test this with a feedback coefficient greater than 1, and it limits it. Try varying the release time on the peak detector and/or the constant subtracted after the peak detector (that's the threshold level). It's a pretty simple implementation, think it through and you'll get it pretty quick.

Better though is to calculate a 'correct' feedback coefficient based on a desired decay time for your plucked note.

…I had a muck about…

Here's a version with decay time calculation for the feedback coefficient. It also takes the filter out of the feedback loop, and adds it at the output, but using the decay time to control the cutoff to get a similar natural fade of the highs quicker than the lows…

The nasty here was calculating the correct decay value for the primary decy envelope… nasty.

I tried to use as much Primary here because that's what you are usin. Don't like it though, and ended up wrapping the core filter, just to make life easier.

Have fun :)

BBD-strings colb edit.ens

I did get one 'blow up' while working on this… not sure if it will happen now that it's done. If so, just add that compressor in the feedback, or just a -1..1 clipper.

gentleclockdivider

Just put a tanh in the feedback loop ( found in core math ) ,

Quick links

Unanswered questions