[sdiy] Filter bank design

[Ian Fritz]

JH --

At 04:05 PM 12/15/2002, jhaible wrote:

>Filter bank spacing of 2.1**(1/5) : This is to get roughly
>5 filters per octave (quite dense spacing), while avoiding
>two filters being spaced exactly an octave.
>(5 filters per octave would be 2**(1/5). )
>
>I wonder why we need such tight spacing in the low
>range, however. If you take a violin, the upper modes
>of the wood resonance are very close to each other
>(thus using a filter bank to get a "string filter").
>But in the lower range, you have just a couple of
>singular resonant peaks (air resonance and fundamental
>of wood resonance), haven't you?

Correct.  The Burhans bank was used for electrifying a clavichord.  He was 
just trying to juice up the sound a bit, I think, and he wanted to cover 
all the frequencies.  It's quite important to also understand that his 
frequency set has no overlap with the musical scale over 10 octaves, so you 
don't get notes popping out at you.

For violin simulation something closer to what you suggest would be more 
appropriate.

That is why I built my bank with 6x6 subbanks with variable Q on each bank, 
and, additionally, why I put a fader on each of the 36 channels.  I can 
reduce the density anywhere by just turning some of the channels down.  I 
can also turn down the Q at low frequencies and set up some broader low 
frequency formant-type resonances.  Then I can use the higher subbanks for 
the high-Q timbral enrichment.

A cool design if you are mainly interested in string simulation might be to 
have a few parametric-EQ-type filters for low frequencies combined with a 
dense fixed-frequency bank above something like 500 Hz or so.

   Ian