ODP: Filters

[Brigman, Corley]

<harmonic distortion>
ok, this makes sense, but isn't totally easy to avoid, since you don't 
really know the frequency content of the bin, and must redistribute its 
energy among multiple bins. dividing it evenly may not be correct. suppose 
we have a 512 element FFT, and a sine wave at 440 Hz. Bin 6 contains the 
info for ~430 to ~516. all the energy for this waveform shows up in this 
bin? at least that's the general idea, since it's basically an impulse it 
probably doesn't actually work that way. anyway to pitch shift up would 
distributing the energy equally between the bins that correspond to ~860 to 
~1032 (4 or 5 bins likely). However, only 1 of these bins should actually 
have any energy in it, the one at 880. Or am I doing this wrong? Even so 
I'm sure there are standard ways to deal with this, i guess this is where 
you need the phase information? to give additional info about the frequency 
content of the bin? i'm really curious about this, now that I am thinking 
about it...i know the frequencies when I do it additive but this is a 
different creature....

>i'd think an ifft would be better for more complicated stuff.. you'd
>acutally need more than one multiplication per harmonic for the filter; the
>frequency response of the filter could be achieved this way, but with the
>FFT method you are using an array of complex numbers that represents  both
>the frequency and phase response of the filter in question (although if you
>are looking to implement a perfectly phase-distortion-free filter, this
>isn't a problem :])

in my very limited empirical analysis :), phase doesn't matter that much
when 
you deal with additive. see my other message for a possible explanation. or 
maybe my sample was too small (very easy, i only used variations on 2
waveforms,
single-note)....

>-uh-oh, we're dangerously close to wavelet analysis here :]
>and isn't the k5000's formant filter FFT-based (or rather iFFT; i'm not
sure
>the data has ever left the frequency domain at that point)?

that's kind of the idea, although I have only conceptual knowledge of
wavelets, 
no experience or even theory. the k5000's filter is described as a 128-band 
graphic eq, which afaik can be modeled as a 128-point FFT, so it wouldn't 
surprise me....and also AFAIK even though the k5000 DOES have some sampled 
transients, i don't think these go through the formant filter, so it sounds
like
it probably is just an IFFT...128 points wouldn't be so bad after all...

enough for now. i'll break out my dsp book tonight and after a couple months
to 
make sense of it :) i'll correct all the mistakes i'm sure i made above....

corley brigman
intel corp.
corley.brigman at intel.com

speaking for me, not for intel.