digiology

I want to split up the spectrum into seperate bands something like a few bandpass filters that overlap so that there are no 'dips' in the spectrum.

This would be like cross~ where adding all the bands yield approximately the original signal.

I've been making my own using multiple biquads with filtergraphs. I've been testing various Q values and band widths with a sine wave to test how the amplitude changes over the spectrum.

I'm sure someone has something coded in max that will do this or even better the formulae I need to create filters of any width across the spectrum and know the Q values are set appropriately so that the added output signals will be almost the same as the original.

I want to split up the spectrum into seperate bands something like a few bandpass filters that overlap so that there are no 'dips' in the spectrum.
This would be like cross~ where adding all the bands yield approximately the original signal.

I've been making my own using multiple biquads with filtergraphs. I've been testing various Q values and band widths with a sine wave to test how the amplitude changes over the spectrum.
I'm sure someone has something coded in max that will do this or even better the formulae I need to create filters of any width across the spectrum and know the Q values are set appropriately so that the added output signals will be almost the same as the original.

overlapping-filters

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~].

if you choose biquad for a frequency cross, i would suggest

everything higher will only add more phase problems, and

depending on what you need to do, pfft might be also a good

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.

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?

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.

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.

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 : (

The Linkwitz-Riley filter is supposed to keep the phase intact, I'm pretty sure its perfect for the job.

I read a post by Trond that many of the objects were implemented for Intel in Jamoma. I'm only on 10.4 so I can't confirm this.

I'm getting "unknown external object type". Boo!

Yeah I'm on 10.4. Unfortunately Jamoma is also for 10.5+

This seems like a good excuse to upgrade to 10.6! Thanks for your help.

If I am correct, Jamoma do support 10.4 on Intel machine so the external should work provided that the Jamoma DSP framework is also built.

[edit] Do not hesitate to drop a note on the Jamoma board if needed : 

[edit] Do not hesitate to drop a note on the Jamoma board if needed : http://jamoma.org/forum.html


The package type will only open on 10.5 and yes I'm on an Intel mac. I just did a search on the forum, they dropped support for 10.4 for version 0.5

I wrote some mistakes. Indeed 10.4 is not supported anymore by Jamoma 0.5.1. Sorry about this.

According to Tim Place, *in theory* Jamoma 0.5 should work on Tiger but it was not extensively tested and we did not receive some feedback from users. See : 

http://jamoma-forums-mailing-lists.3076123.n2.nabble.com/Trouble-with-installer-on-10-4-tp5226454p5226454.html

According to Tim Place, *in theory* Jamoma 0.5 should work on Tiger but it was not extensively tested and we did not receive some feedback from users. See : http://jamoma-forums-mailing-lists.3076123.n2.nabble.com/Trouble-with-installer-on-10-4-tp5226454p5226454.html

Thanks raja but you forgot the mxo file!


Thanks again. I think I had actually gotten the last attachment's folder mixed up with an old one. Anyway the important thing is that it works now!

Its actually a nice filter, I wonder why its not (I think) usually used as an EQ.

Overlapping filters