Filter feedback

[k brown]

I searched the forums and nothing really came up - has anyone figured out a way to do filter feedback that has some sort of limiter in the chain so no matter how high it's turned up it won't 'choke' the filter and 'crash' the synth for a few seconds? From my own playing around, only the SVF LP really sounds good with FB (and can sound really good), but I've never figured out a way to make it 'choke-proof'. I've put it in a few of my projects and I don't mind dealing with the chokes, but many end-users would think the synth was just crap because of it.

[Spogg]

I’m not a filter guy but I have experimented with limiters in echo feedback situations.

The problem is that gain/output level limiters always introduce more distortion the harder they have to work. This is due to the evaluation of the signal by envelope followers which are never perfect or ideal. Slow ones will be smoother but not respond well to transients and fast ones will have more ripple on the output which infects the signal more.

Cheers

Spogg

[k brown]

Bummer, mannn -

Maybe Martin will weigh in; perhaps he's experimented with filter feedback - I hope there's at least a partial solution, because at the right levels it really fattens the sound.

[tulamide]

It has been done before, so I'm sure somebody will know it. Here's an example from Reason's Thor synth, featuring a Lowpass Ladder Filter with a shaper in the feedback loop.

Lowpass Ladder with feedback shaper

[martinvicanek]

Vast topic.
From control theory, we know that if the transfer function has no poles outside the unit circle (in z-space) the system will be stable (won't choke). Unfortunately, this theorem won't help much unless you are familiar with the math and know how to determine the transfer function for a given filter topology in the first place. But even then it only covers so-called linear time independent (LTI) systems, i.e. static filters with constant coefficients - that would pretty much rule out every single synth!
With fast envelopes a filter may choke even if it is stable at any instance in time - a phenomenon known as parametric resonance. There is no simple general stability criterion for non-LTI systems (i.e. dynamic filters) because you have so many possibilities to consider, but it is known that certain topologies (like thisor this) are less prone to "choking" than others.
My practical recommendation woul be to include a "Reset Audio" button to avoid the hassle of having to reload a plugin when it chokes.

[k brown]

It's been my experience that a synth never needs to be reloaded when it chokes from filter feedback; it always restores itself after a few seconds. I wonder if there's something that will rest it immediately, rather than having to wait or manually hit a Reset button.

P.S. - I just did a little experimenting and found that your Complex Low Pass seems far more tolerant of feedback than the SVF. Care to share a little about that filter - I've been using it a lot lately. Is it one of the topologies you mentioned?

[trogluddite]

It's been my experience that a synth never needs to be reloaded when it chokes from filter feedback; it always restores itself after a few seconds

It often depends upon how long you leave the code in that state. The real killer condition is when the numbers continue climbing until they get bigger than a 32-bit float can represent, so become marked as +/- infinity - quite easy if the feedback is exponential. Since most DSP equations have no mathematically defined result for infinity, processing them usually results in a NaN (not a number). The result of absolutely any calculation with a NaN is always another NaN, so this will usually lead to a total lock-up - in downstream code too if NaN is the final filter output.

Though it probably wouldn't sound very nice, just hard-clipping the feedback with min/max so that +/- infinity aren't possible would prevent that in most cases. The filter could still "lock-up" in the sense of the output being stuck with a massively distorted signal, but it should avoid the kind of fatal lockups that require a reset of the audio streams.

[martinvicanek]

{...] your Complex Low Pass seems far more tolerant of feedback than the SVF. Care to share a little about that filter - I've been using it a lot lately. IS it one of the topologies you mentioned?

Although it is not one of the above topologies, it has been designed with the objective to allow fast modulation, yes. The idea was to use a complex oscillator with some damping as a filter. The key property of such an arrangement is that the internal fiter states are always orthogonal no matter what, hence changing the filter parameters wouldn't blow the filter up.

[k brown]

Very interesting - had no idea an osc could be used that way.

@trog - I'll try a min/max (I assume with float 1s) in the loop and see what effect.

[k brown]

Put a min/max 1- to 1 in the input to the feedback amp (0-1.5) - works great! - big time fattness with guard rail; no choke!

Thanks trog! - you know I could have sworn that I'd tried that before, but I must not have had it in the right place in the signal path. Takes things a giant step toward that elusive 'analog warmth' we all get so sick of hearing about.