13 Jul
2013
13 Jul
'13
4:04 p.m.
Hi all,
I'm working on a wavetable oscillator class, and I'm wondering about how to
best go about bandlimiting. I see two ways to achieve bandlimiting, i'll
detail as A and B.
A) Create different wavetables for each octave. Base octave includes all
harmonics. Octave 1 has the top half of the harmonics removed by FFT. Oct 2
has the harmonic content halved again.
B) Oversample the single waveform x8. Play the oversampled audio back, and
lowpass with a steep rolloff just below the nyquist of the output
samplerate. Removal of the otherwise aliasing harmonics is done at the
higher samplerate, so its not aliased yet at that stage.
I'm asking in terms of quality and CPU usage: this is for a synth which has
3 oscillators per voice, and 16 / 32 voices..
Cheers, -Harry
13 Jul
13 Jul
4:32 p.m.
On Sat, Jul 13, 2013 at 05:03:59PM +0100, Harry van Haaren wrote:
A) Create different wavetables for each octave. Base
octave includes all
harmonics. Octave 1 has the top half of the harmonics removed by FFT. Oct 2
has the harmonic content halved again.
Fine for fixed frequencies (aeolus works this way but not per octave,
all notes are precomputed separately), but becomes problematic if the
you allow frequency modulation over a wide range.
Can be combined with oversampling to make interpolation easier.
B) Oversample the single waveform x8. Play the
oversampled audio back, and
lowpass with a steep rolloff just below the nyquist of the output
samplerate. Removal of the otherwise aliasing harmonics is done at the
higher samplerate, so its not aliased yet at that stage.
This could be done very efficiently, in particular if you accept some
compromises for the lowpass filter, and even more if the waveforms can
be summed before downsampling.
There are other methods, depending on what sort of waveforms
you want to generate.
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
4:56 p.m.
On Sat, Jul 13, 2013 at 5:32 PM, Fons Adriaensen <fons(a)linuxaudio.org>
wrote:
This could be done very efficiently, in particular if
you accept some
compromises for the lowpass filter, and even more if the waveforms can
be summed before downsampling.
Good idea, I wouldn't have thought of summing before downsampling.
In terms of compromises for the lowpass filter, what is it you were
thinking of?
A bit of reading tells me that most analog devices use elliptic filters,
with 4/8 poles or so for this purpose..
Seems a little overkill? I've no experience in this area... yet!
-Harry
3972
days inactive
3972
days old
linux-audio-dev@lists.linuxaudio.org
2 comments
2 participants
participants (2)
-
Fons Adriaensen -
Harry van Haaren