A question about digital filter (impulse response)

[dundel]

Hi

Is there any posibilty to create a filter using impulse response snapshots of real analog filter (like in hydratone). I am no dsp enginer but could it work

[pitchdrifter]

It would 'work' but there wouldn't be much of a point to it. Filters sound great when the cutoff frequency moves, when both resonance and cutoff react with an envelope follower to the incoming audio (like some hardware filterbanks do it). With an impulse you'd get a single static response file.. you need countless response files to properly copy a filter's sound. There's some hardware unit that uses multisampled responses.. forgot it's name.
Ultimately it's better to use a good algorhythmic filter plugin, like the ohmforce quadfrohmage or u-he's filterscape.

Markus

[Christian Budde]

Is there any posibilty to create a filter using impulse response snapshots of real analog filter (like in hydratone). I am no dsp enginer but could it work

One method called 'impulse invariance method'. But as xRAVENx mentioned it will only get you a single set of coefficients. If you are about to recreate a variable, analog filter, then it is usually better to use bilinear transformation. It is possible to fit these curves to meet the original.
I've downloaded hydratone and valvetone demo some days ago. Havn't checked hydratone in detail, but valvetone has several impulse responses stored internally (i guess), these are overlayed depending on the actual settings.
Recording some impulse responses of valvetone shows me, that it uses internally an impulse response of about 148 samples. The latency of 64 samples indicates, that there is most probably some FFT convolution algorithm behind. Although 64 samples is usually know as the limit of time domain and frequency domain convolution. Below this time domain convolution is better, above frequency domain convolution. Without decompilation, which is not legal (!), i can only guess further details here. Most probably i've already gone too far in detail, but i'm always so curious about how things work.
At the end i was a little bit disappointed, because i expected some sort of dynamic convolution, which i couldn't found.
If my thesis about how it works is correct, then i'm also a little bit disapointed about the CPU performance. At least it does sound very good to my ears.

Sorry for being a bit OT,

Christian

[martian]

no they dont use dynamic convolution like a sintefex, i think there's some troublesome patents around that technology anyway?

instead for things like hydratone, its just thousands of impulses. yes its quite a cpu hog on pc, but they havent finished the SSE optimisations for it yet, which will reduce cpu load by up to 4x apparently.

[autloc]

It would 'work' but there wouldn't be much of a point to it. Filters sound great when the cutoff frequency moves, when both resonance and cutoff react with an envelope follower to the incoming audio (like some hardware filterbanks do it). With an impulse you'd get a single static response file..

If you used the impulse response as the coefficients of an FIR filter where the delay elements in the FIR filter are replaced with first-order allpass filter blocks, you could adjust the allpass filter coefficient to sweep the filter up and down (musicdsp.org has the code for a warped-FIR filter for anyone curious). It wouldn't be an accurate emulation or anything, but at least the filter wouldn't be static (and it wouldn't require a huge library of impulse responses). An allpass filter chain greater than 100 elements long could be pretty processor intensive, mind you..

[pitchdrifter]

It would 'work' but there wouldn't be much of a point to it. Filters sound great when the cutoff frequency moves, when both resonance and cutoff react with an envelope follower to the incoming audio (like some hardware filterbanks do it). With an impulse you'd get a single static response file..

[Lunch Money]

xRAVENx - He's just saying that you point the phasers at the dilithium crystals.

[dundel]

bumpy be bump ;]

Maybe Aleksey has to say something about that.