posted 12 November 2005 07:12            

---

Hi,
because of my current interest in partial transformation,
I also looked again at this sound.
What I don`t understand completely is how !BW is build in
here.
I could understand if a combination of a PulseTrain and a
Delay would result in a kind of harmonic filter, where the
Delay determines the lowest, 'fundamental' frequency to be
filtered. But probably I don't understand it correctly, can
somebody explain it to me?
Best,
Florian

IP: Logged

posted 12 November 2005 11:09            

---

Think of the pulse waveform as a spectral envelope. The delay determines the lowest partial (as you have already pointed out). The duty cycle of the pulse corresponds to the bandwidth of the filter.

IP: Logged

posted 13 November 2005 12:02            

---

But this not comparable with a normal band-pass,
since a fundamental freauency and its partials
with a step-size of 256 samples are filtered.
I am reasking because I am interested in filtering
just one band via the FFT-way.
What would be the best solution for this?

IP: Logged

posted 13 November 2005 15:00            

---

The delay moves the center frequency of the pass band. It should be a fraction of the total number of partials (for example the MaxDelay could be 256 samp and the Delay could be a hot value so you could move the window up or down in frequency). The duty cycle controls the bandwidth. Duty cycle should be a fraction of the full frequency range, for example: {1.0/256}

[This message has been edited by SSC (edited 13 November 2005).]

IP: Logged

posted 16 November 2005 02:19            

---

I first thought that the period-length of the PulseTrain was a kind
of subdivision of the spectral envelope and that way filters not only
at one frequency but at several, depending on the period-length of the
PulseTrain.
Could this be an interesting variation to this sound?
For example: paste the expression "!Period * 256 Samples" into the Period field of the PulseTrain (so that the product results in subdiv.
of 256).
Wouldn't that result in a kind of "harmonic filter"?

IP: Logged

posted 16 November 2005 08:46            

---

Well, if you were multiplying the 256 samp period by 1/2, you would get 2 copies of the spectral envelope shape per spectral period. So it would be like getting two bandpass filters. And if you multiplied the period by 1/64, you would get 64 bandpass filters, and so on.

Is that what you had in mind?

IP: Logged

posted 17 November 2005 06:43            

---

If this results in equally over the spectrum spread filters, in case the
!CF equals zero (otherwise they are equally spread with an offset), then
it corresponds to that what I had in my mind. But I'm not completely
sure that this can be accomplished by the method/change mentioned
earlier.
Or is the result of this change that you just create multiples of this
very same filter, that is, the frequency of the filters stays always the
same? (that is definitely not what I had in mind)
As a possible next step: is it possible to build an entire spectral
filterbank, that gives you controllable freq, bw and scale per band,
based on this principle? Possibly somebody already has done something
like this before?
Btw: thanks for the tips regarding the shifting partials issue!

IP: Logged

All times are CT (US)