Overlapping filters

Bump!

I've tried using multiple filters and testing how a sine wave's amplitude changes over the spectrum [particularly between overlapping filters] but haven't fine tuned it perfectly.

I've also tried using multiple cross~ objects but I don't like the idea of the signal being put through multiple filters that might colour or alter the phase of the added signal. Anybody got a better idea or patch to share?

Maybe look to [fffb~] which is like multiple [reson~] objects. Or dive into the world of Fourier transforms with [pfft~].

lh

if you choose biquad for a frequency cross, i would suggest
a "Q setting" of 12 db.

everything higher will only add more phase problems, and
everything lower will split less.

depending on what you need to do, pfft might be also a good
idea for a 2-band or 4-band.

-110

bump!

Roman Thilenius, I'm not sure what you mean by a Q setting of 12db, thought Q was unitless.

Anyway, for anyone seeing this thread for the first time, I want to split a signal into bands preferably using bandpass filters (or any other method you might recommend). I think about 5 bands would be fine. The most important think is that when each band is added it would reconstruct the signal perfectly or as close as possible with everything including the phase left intact.

I suppose I really need to know what would be suitable bandpass parameters (e.g. the Q) or if another method would be more suitable.

I'm guessing an fft (like the forbidden planet patch) would be best to very sharply delimited bands?

I had been using multiple biquad~ objects (before I discovered cascade~) but it was hard to find the right parameters.

First of all I wasn't able to find how much attenuation per octave a given Q value would give! I had to try different Q values and overlaps and then test the amplitude of a sweeping sine tone or white noise. It took forever and I never got it quite right.

I should say that using multiple cross~ objects looks fine in terms of the spectrum but I'm not sure if it maintains the original phase and if the filter is as steep as it could be with multiple bandpass filters.

I'm assuming there is a standard way of going about this. What do multiband compressors use for example?

Thanks raja.

I was hoping to avoid any maths but I suppose I'll have to look into it.

I was reading about Logic's multiband compressor and someone in the forum said that the filters used were Linkwitz-Reilly filters.

I don't see any implementation of these for max/msp, oh well.

I found this: http://www.audiohelix.com/web-docs/bw-q-q-bw.html which will give you the bandwidth of a given Q. Presumably the bandwidth in this context is defined as the width between where it dips below -3dB.

Well thats a step in the right direction anyway.

http://maxobjects.com/?v=objects&id_objet=3054&requested=crossover&operateur=AND&id_plateforme=0&id_format=0

Found a Linkwitz-Riley filter but I don't think it'll work on an intel mac : (