Roman Thilenius

is there a way how to predict/precalculate when a biquad explodes?

biquad-explosion

As with any filter, there is interaction between the input signal and the coefficiants. So unless you can predict/precalcuate the input, you can't predict the output and subsequent change of explosion.

However, anytime your b coefficients are greater than 1., there's a potential for trouble. I'd say that if b1 + b2 > 1., then the filter is in a potentially explosive state.

As with any filter, there is interaction between the input signal and the coefficiants.  So unless you can predict/precalcuate the input, you can't predict the output and subsequent change of explosion.  

However, anytime your b coefficients are greater than 1., there's a potential for trouble.  I'd say that if b1 + b2 > 1., then the filter is in a potentially explosive state.


well it is clear that it depends on the input, but i am trying to make up a set of rules, is that possible?

a set of rules which is better than "avoid loud subbasses at the input and dont modulate the filter frequency too fast", that is.

interestingly it is not possible to find anything on the net about this.

well it is clear that it depends on the input, but i am trying to make up a set of rules, is that possible?
a set of rules which is better than "avoid loud subbasses at the input and dont modulate the filter frequency too fast", that is.

interestingly it is not possible to find anything on the net about this.


There is a "triangle of stability" IIRC that can help predict this, though it's still pretty limited.

Biquads don't modulate as well as other filters. You can improve things with filtercoeff~ by increasing the frequency of its calculations (which are done once per signal vector by default).

KVRAudio's forums are also a good place to check out on this.

Biquads don't modulate as well as other filters.  You can improve things with filtercoeff~ by increasing the frequency of its calculations (which are done once per signal vector by default). 

KVRAudio's forums are also a good place to check out on this.


i was hoping to find it on musicdsp, but there is nothing on the topic.

a biquad filter external which updates every sample would be great ... but i dont know if that would make a difference ... or of it would be worth the CPU.

what i was using until now is a simple mechanism which raises the interpolation time of the incoming coeff signals the lower the freq input is. that works, more or less, but is not perfect for any situation.

somehow i was hoping for an industry standard for running the filter safely using interpolated input. :)

a biquad filter external which updates every sample would be great ... but i dont  know if that would make a difference ... or of it would be worth the CPU.

what i was using until now is a simple mechanism which raises the interpolation time of the incoming coeff signals the lower the freq input is. that works, more or less, but is not perfect for any situation. 

There's always smooth-biquad~. If you update the coefficients every sixteen samples or so you should get a decently smooth sweep.

How fast are you trying to modulate the filter? You could use a lowpass filter on the frequency input (so you can't accidentally try to modulate it at 200 Hz). Cross~ might be worth a look there, since it's got a pretty steep rolloff.

There's always smooth-biquad~.  If you update the coefficients every sixteen samples or so you should get a decently smooth sweep.

How fast are you trying to modulate the filter?  You could use a lowpass filter on the frequency input (so you can't accidentally try to modulate it at 200 Hz).  Cross~ might be worth a look there, since it's got a pretty steep rolloff.


biquad explosion