hi Tim,
while coding cyclone's filters, I was mainly concerned about matching impulse responses of msp counterparts (I believe, that currently most of them match 100% for the entire domain of control parameters... with svf~, however, there is still some difference at high Q -- this one is tricky to clone, due to a nonlinear component thrown into the formula).
In theory, a plain svf performs slightly better than four separate filters (lowpass, highpass, bandpass, notch). Cyclone's svf is not a plain svf, but since it does no oversampling, and calculates the coefficients once per block, the hope is it does perform well (although the only test I ever did, apart from checking that it is useable, was estimating the oversampling costs).
Comparing to the biquad~ internal, svf~ does have an overhead, obviously -- there is always a price for allowing to control a filter with signal input (both q and f may be modulated).
Krzysztof
Tim Blechmann wrote: ...
i'm currently using biquad filters (i'm getting the settings from iemlib's lowpass, highpass (...) externals). i consider to change my patch to use svf~ from the cyclone library.
does anyone know, which algorithm is more expensive in terms of cpu time? and are there any mayor differences in the characteristics of these two filter types?