SuperCollider CLASSES (extension)

LTI
ExtensionExtension

Linear Time Invariant General Filter Equation
Inherits from: UGen : AbstractFunction : Object
 

SLUGens released under the GNU GPL as extensions for SuperCollider 3, by Nick Collins http://composerprogrammer.com/index.html

Description

General Linear Time-Invariant (LTI) filter UGen where you specify any coefficient set via a Buffer Represents the general LTI filter difference equation in the time domain:

y(n) = b0x(n) + b1x(n-1) + ... + b(Nb)x(n-Nb) + a1y(n-1) + ... + a(Na)y(n-Na)

This is not a pole/zero view, so you'd need to calculate time domain coefficients yourself if you want to work from z-plane backwards. A corollary is, stability is not guaranteed. This is part of the fun?

You need to pass in the coefficients via two buffers, of arbitrary size.

Class Methods

*ar (input, bufnuma: 0, bufnumb: 1, mul: 1, add: 0)

Arguments:

input

What do you want to filter?

bufnuma

Feedback filter coefficients, from previous outputs

bufnumb

Feedforward filter coefficients, from previous inputs

Inherited class methods

Instance Methods

Inherited instance methods

Examples

(
a = [0.02, -0.01]; // feedback coefficients
b = [1, 0.7, 0, 0, 0, 0, -0.8, 0, 0, 0, 0, 0.9, 0, 0, 0, -0.5, 0, 0, 0, 0, 0, 0, 0.25, 0.1, 0.25]; // feedforward coefficients
c = Buffer.sendCollection(s, a, 1);
d = Buffer.sendCollection(s, b, 1);
)

{ LTI.ar(SoundIn.ar([0, 1]), c.bufnum, d.bufnum) }.play


// Note- you cannot update buffers during playback unless you stay within the initially allocated sizes

(
a = Array.fill(100, { 0.0 }); // feedback coefficients
b = Array.rand(100, -0.5, 0.5); // feedforward coefficients
b[0] = 1;
c = Buffer.sendCollection(s, a, 1);
d = Buffer.sendCollection(s, b, 1);
)

{ LTI.ar(SoundIn.ar([0, 1]), c.bufnum, d.bufnum) }.play


(
b = Array.rand(100, -0.5, 0.5); // feedforward coefficients
b[0] = 1;
d.sendCollection(b);
)

// may explode...

(
10.do({ arg i; a[100.rand] = rrand(-0.1, 0.1) }); // feedforward coefficients
c.sendCollection(a);
)

// from a routine
(
e = {
    inf.do {
        b = Array.rand(100, -0.5, 0.5); // feedforward coefficients
        b[0] = 1;
        d.sendCollection(b);
        0.1.wait;
    }

}.fork
)

e.stop;




// Code for testing and trying coefficients:

// given two arrays of filter coefficients, calculate an impulse response over 1024 samples, then the fft gives approximate frequency gain and phase response


(
var size = 1024, real, imag, cosTable, complex;
var a, b;
var lastn, lastindex, num;
var y, max;

a = [0.02, 0.05, 0, 0, 0.01]; // feedback coefficients

b = [1, 1, -0.5, 0, 0, 0, -0.6, 0.7]; // feedforward coefficients

// check poles of a are inside the unit circle by factorising the complex polynomial?
// this procedure uses only a finite impulse response so may give fallacious results of stability

num = a.size;
lastn = Array.fill(num, { 0 });
lastindex = 0;

real = Signal.fill(size, { arg i;
    y = if(i < b.size) { b[i] } { 0 };
    y = y + ((a.collect({ arg val, j; val*(lastn.wrapAt(lastindex + num-1-j)); })).sum);
    lastn[lastindex] = y;
    lastindex = (lastindex + 1) % num;
    y
});

imag = Signal.newClear(size);

cosTable = Signal.fftCosTable(size);

complex = fft(real, imag, cosTable);

a = complex.postln;
[real, (complex.magnitude), (complex.phase) ].flop.flat
.plot("fft", Rect(0, 0, 1024 + 8, 500), numChannels: 3);

max = 0;
y = complex.magnitude;
y.do { arg val; if(val > max, { max = val }) };
max
)


// how to create the arbitrary filter from its difference equation coefficients? Need a new UGen (LTI)- or use Csound

(
a = [0.02, 0.05, 0, 0, 0.01]; // feedback coefficients
b = [1, 1, -0.5, 0, 0, 0, -0.6, 0.7]; // feedforward coefficients
c = Buffer.sendCollection(s, a, 1);
d = Buffer.sendCollection(s, b, 1);
)

{ Impulse.ar(1) }.play

{ LTI.ar(Impulse.ar(1), c.bufnum, d.bufnum) }.play

{ LTI.ar(SoundIn.ar([0, 1]), c.bufnum, d.bufnum) }.play


(
a = [0.01, -0.01]; // Array.fill(10, { rrand(0.001, 0.01) }); // feedback coefficients
b = [1] ++ Array.fill(100, { exprand(0.1, 1) }); // feedforward coefficients
c = Buffer.sendCollection(s, a, 1);
d = Buffer.sendCollection(s, b, 1);
)


// piercing, careful!
{ Saw.ar(LFNoise0.kr(10, 4000, 5000)) }.play

{ LTI.ar(Saw.ar(LFNoise0.kr(10, 4000, 5000)), c.bufnum, d.bufnum, 0.1) }.play

// Also see [Convolution]