30 Nov
2010
30 Nov
'10
2:51 p.m.
How does Steve Harris' Harmonic generator LADSPA plugin work? I mean relation
between controls (concrete harmonics levels) and transfer function
Vout = Vin(1 + a*Vin + b*Vin^2 + ...)
30 Nov
30 Nov
4:35 p.m.
On Tuesday, November 30, 2010 16:51:21 Andrew Gaydenko wrote:
How does Steve Harris' Harmonic generator LADSPA
plugin work? I mean
relation between controls (concrete harmonics levels) and transfer
function
Vout = Vin(1 + a*Vin + b*Vin^2 + ...)
..or, saying about my aim, is there software to process sound stream (JACK) or
file the way shown above (with given a, b, c,.. coefficients)? You see, I say
about amplitude transformation only.
1 Dec
1 Dec
12:33 a.m.
Andrew Gaydenko wrote:
How does Steve Harris' Harmonic generator LADSPA
plugin work? I mean relation
between controls (concrete harmonics levels) and transfer function
Vout = Vin(1 + a*Vin + b*Vin^2 + ...)
This can be done with waveshaping.
If your source is a sinewave you can run it's samples, x [n],
through a polynomial function:
y [n] = P (x [n])
and there is a way to calculate the required polynomial P
using Chebyshev polynomials.
I suspect that if you google for "waveshaping chebyshev" you
will find a more complete exposition than what I can drag up
from the depths of my memory.
Erik
--
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/
12:46 a.m.
On Wed, Dec 01, 2010 at 10:33:17AM +1100, Erik de Castro Lopo wrote:
and there is a way to calculate the required
polynomial P
using Chebyshev polynomials.
Or even better: evaluate the Chebyshev polynomials directly on
the signal (this avoids loss of precision due to cancellation of
large nearly equal terms).
Ciao,
--
FA
There are three of them, and Alleline.
11:16 a.m.
On Wednesday, December 01, 2010 02:46:10 fons(a)kokkinizita.net wrote:
On Wed, Dec 01, 2010 at 10:33:17AM +1100, Erik de
Castro Lopo wrote:
I need just
Vout = Vin(a0 + a1*Vin + a2*Vin^2 + ... + aN*Vin^N)
for given a0,..aN with limited N (a1..aN abs will be < 0.1). Say, we just
recalculat samples' amplitudes directly. Is there any caveats in such direct
transforming? And why do I need to use another (Chebyshev polynomials) way?
Will I get HF hum? Please, direct my nose to an approriate reading! :-)
Andrew
and there is a way to calculate the required
polynomial P
using Chebyshev polynomials.
Or even better: evaluate the Chebyshev polynomials directly on
the signal (this avoids loss of precision due to cancellation of
large nearly equal terms).
1 p.m.
On Wed, Dec 01, 2010 at 01:16:07PM +0300, Andrew Gaydenko wrote:
I need just
Vout = Vin(a0 + a1*Vin + a2*Vin^2 + ... + aN*Vin^N)
for given a0,..aN with limited N (a1..aN abs will be < 0.1). Say, we just
recalculat samples' amplitudes directly. Is there any caveats in such direct
transforming? And why do I need to use another (Chebyshev polynomials) way?
If that is what you need it's simple enough to compute...
But it won't give you a_1 second harmonic distortion, a_2 third
etc (although for small values of 'a_i' you could get near).
The Chebyshev way allows you to control each harmonic independently
(provided the input is a sine).
This is because for the n-th Chebyshev polynomial T_n ()
T_n (cos (x)) = cos (n * x)
The polynomials are very easy to compute using recursion:
T_0(x) = 1
T_1(x) = x
T_N(x) = 2 * x * T_{N-1}(x) - T_{N-2}(x)
etc.
This is simple enough to do directly on the signal, no need
to compute the explicit polynomials.
Will I get HF hum? Please, direct my nose to an
approriate reading! :-)
You will get aliasing if you try to generate anything above Fs/2.
Ciao,
--
FA
There are three of them, and Alleline.
1:13 p.m.
On Wednesday, December 01, 2010 15:00:49 fons(a)kokkinizita.net wrote:
On Wed, Dec 01, 2010 at 01:16:07PM +0300, Andrew
Gaydenko wrote:
Fons, I want just polinomial transformation regardless harmonics at all. I
have mentioned harmonics as a result of searching a LADSPA plugin related to
my needings.
I'm afraid direct amplitude caculation in accorance with a0...aN will produce
"too fast" signal deviation, resulting in exceeding fs/2 band.
Andrew
I need just
Vout = Vin(a0 + a1*Vin + a2*Vin^2 + ... + aN*Vin^N)
for given a0,..aN with limited N (a1..aN abs will be < 0.1). Say, we just
recalculat samples' amplitudes directly. Is there any caveats in such
direct transforming? And why do I need to use another (Chebyshev
polynomials) way?
If that is what you need it's simple enough to compute...
2:33 p.m.
On Wed, Dec 01, 2010 at 03:13:03PM +0300, Andrew Gaydenko wrote:
Fons, I want just polinomial transformation regardless
harmonics at all. I
have mentioned harmonics as a result of searching a LADSPA plugin related to
my needings.
Then what is your question ? All you need to do is compute the
polynomial for each input sample.
I'm afraid direct amplitude caculation in
accorance with a0...aN will produce
"too fast" signal deviation, resulting in exceeding fs/2 band.
Any non-linear transfer function will do this for some inputs.
If you have an x^N term, anything above Fs/(2*N) will alias.
If you want the nonlinearity without aliasing, you have to
upsample (by a factor N, with N = highest power of x used),
apply the nonlinear function, then downsample.
Ciao,
--
FA
There are three of them, and Alleline.
3:21 p.m.
On Wednesday, December 01, 2010 16:33:35 fons(a)kokkinizita.net wrote:
Fons, thanks! It is *that* trick I needed to eliminate my doubts wrt fs/2
products :-) I think SRC is an approriate lib for up/down-sampling.
Andrew
I'm afraid
direct amplitude caculation in accorance with a0...aN will
produce "too fast" signal deviation, resulting in exceeding fs/2 band.
Any non-linear transfer function will do this for some inputs.
If you have an x^N term, anything above Fs/(2*N) will alias.
If you want the nonlinearity without aliasing, you have to
upsample (by a factor N, with N = highest power of x used),
apply the nonlinear function, then downsample.
4:11 p.m.
On Wed, Dec 01, 2010 at 05:21:09PM +0300, Andrew Gaydenko wrote:
On Wednesday, December 01, 2010 16:33:35
fons(a)kokkinizita.net wrote:
Fons, thanks! It is *that* trick I needed to eliminate my doubts wrt fs/2
products :-) I think SRC is an approriate lib for up/down-sampling.
For the fixed ratios you need zita-resampler will probably
somewhat faster.
--
FA
There are three of them, and Alleline.
I'm
afraid direct amplitude caculation in accorance with a0...aN will
produce "too fast" signal deviation, resulting in exceeding fs/2 band.
Any non-linear transfer function will do this for some inputs.
If you have an x^N term, anything above Fs/(2*N) will alias.
If you want the nonlinearity without aliasing, you have to
upsample (by a factor N, with N = highest power of x used),
apply the nonlinear function, then downsample.
5165
days inactive
5166
days old
linux-audio-dev@lists.linuxaudio.org
9 comments
3 participants
participants (3)
-
Andrew Gaydenko -
Erik de Castro Lopo -
fons＠kokkinizita.net