/****************************************************** * * Adaptive Systems for PD * * copyleft (c) Gerda Strobl, Georg Holzmann * 2005 * * for complaints, suggestions: grh@mur.at * ****************************************************** * * license: GNU General Public License v.2 * ******************************************************/ #include "adaptive.h" /* ------------------------ nlms~ ------------------------- */ static t_class *nlms_tilde_class; typedef struct _nlms { t_object x_obj; t_float f; t_sample *buf; t_sample *tmp; t_int bufsize; int adapt; // enable/disable adaptation t_int N; //number of coefficients of the adaptive system t_float *c; // coefficients of the system t_float mu; // step-size parameter (learning rate) t_float alpha; // small constant to avoid division by zero t_canvas *x_canvas; } t_nlms_tilde; static void nlms_tilde_a(t_nlms_tilde *x, t_floatarg f) { x->adapt = (f==0) ? 0 : 1; } static void nlms_tilde_geta(t_nlms_tilde *x) { if(x->adapt==0) post("nlms~: adaptation is currently OFF"); else post("nlms~: adaptation is currently ON"); } static void nlms_tilde_mu(t_nlms_tilde *x, t_floatarg f) { x->mu = f; } static void nlms_tilde_getmu(t_nlms_tilde *x) { post("mu (step-size parameter): %f", x->mu); } static void nlms_tilde_alpha(t_nlms_tilde *x, t_floatarg f) { x->alpha = f; } static void nlms_tilde_getalpha(t_nlms_tilde *x) { post("alpha: %f", x->alpha); } static void nlms_tilde_getN(t_nlms_tilde *x) { post("N (number of coefficients): %d", x->N); } static void nlms_tilde_clear(t_nlms_tilde *x) { int i; // clear coefficients for(i=0; iN; i++) x->c[i] = 0; // clear temp buffer for(i=0; iN-1; i++) x->buf[i] = 0; } static void nlms_tilde_init(t_nlms_tilde *x) { int i; // set the first coefficient to 1, all others to 0 // so this is a delay free transmission x->c[0] = 1; for(i=1; iN; i++) x->c[i] = 0; // clear temp buffers for(i=0; iN-1; i++) x->buf[i] = 0; } static void nlms_tilde_print(t_nlms_tilde *x) { int i; // print coefficients post("\nNr. of coefficients: %d",x->N); post("coefficients:"); for(i=0; iN; i++) post("\t%d: %f",i,x->c[i]); } static void nlms_tilde_write(t_nlms_tilde *x, t_symbol *s) { // make correct path char filnam[MAXPDSTRING]; char filename[MAXPDSTRING]; canvas_makefilename(x->x_canvas, s->s_name, filnam, MAXPDSTRING); sys_bashfilename(filnam, filename); // save to file adaptation_write(filename, x->N, x->mu, x->c); } static void nlms_tilde_read(t_nlms_tilde *x, t_symbol *s) { // make correct path char filnam[MAXPDSTRING]; char filename[MAXPDSTRING]; canvas_makefilename(x->x_canvas, s->s_name, filnam, MAXPDSTRING); sys_bashfilename(filnam, filename); // read file adaptation_read(filename, &x->N, &x->mu, x->c, x->buf); } static t_int *nlms_tilde_perform(t_int *w) { t_nlms_tilde *x = (t_nlms_tilde *)(w[1]); t_sample *x_ = (t_sample *)(w[2]); t_sample *d_ = (t_sample *)(w[3]); t_sample *y_ = (t_sample *)(w[4]); int n = (int)(w[5]); int i, j, tmp; t_sample e=0, x_2; for(i=0; itmp[i]=0; // y_[i] += x->c[j] * x_[i-j]; // so lets split in two halfs, so that // negative indezes get samples from the // last audioblock (x->buf) ... tmp = (i+1 - x->N)*(-1); tmp = tmp<0 ? 0 : tmp; for(j=0; jN-tmp; j++) x->tmp[i] += x->c[j] * x_[i-j]; for(j=x->N-tmp; jN; j++) x->tmp[i] += x->c[j] * x->buf[(i-j)*(-1)-1]; if(x->adapt) { x_2=0; // error computation e =d_[i] - x->tmp[i]; // Normalized LMS Adaptmsation Algorithm // (split in the same way as above) // // c[n] = c[n-1] + mu/(alpha + x'[n]*x[n])*e[n]*x[n] // calc x'[n]*x[n] // TODO: Performance Optimization: save results from the past // so that this for loop should be obsolet ... for(j=0; jN-tmp; j++) x_2 += x_[i-j] * x_[i-j]; for(j=x->N-tmp; jN; j++) x_2 += x->buf[(i-j)*(-1)-1] * x->buf[(i-j)*(-1)-1]; for(j=0; jN-tmp; j++) x->c[j] = x->c[j] + x->mu/(x->alpha+x_2) * x_[i-j] * e; for(j=x->N-tmp; jN; j++) x->c[j] = x->c[j] + x->mu/(x->alpha+x_2) * x->buf[(i-j)*(-1)-1] * e; } //post("%d: in %f, d: %f, out: %f, error: %f, c1:%f, c2:%f", i, x_[i], d_[i], x->tmp[i], e, x->c[0], x->c[1]); } // store last samples for next audiobuffer for(i=0; iN-1; i++) x->buf[i] = x_[n-1-i]; // now write tmp to outlet while(n--) y_[n] = x->tmp[n]; return (w+6); } static void nlms_tilde_dsp(t_nlms_tilde *x, t_signal **sp) { // allocate new temp buffer if buffersize changes if(x->bufsize != sp[0]->s_n) { if(sp[0]->s_n < x->N) post("nlms~ WARNING: buffersize must be bigger than N, you will get wrong results !!!"); if(x->tmp) freebytes(x->tmp, sizeof(t_sample) * x->bufsize); x->tmp = (t_sample *)getbytes(sizeof(t_sample) * sp[0]->s_n); x->bufsize = sp[0]->s_n; } dsp_add(nlms_tilde_perform, 5, x, sp[0]->s_vec, sp[1]->s_vec, sp[2]->s_vec, sp[0]->s_n); } static void nlms_tilde_helper(void) { post("\nnlms~: Adaptive transversal filter using normalized LMS"); post("INPUT:"); post("\tinlet1: input signal x[n]"); post("\tinlet2: desired output signal d[n]"); post("\tinit_arg1: number of coefficients of the adaptive system"); post("\tinit_arg2, mu: step-size parameter (learning rate)"); post("OUTPUT:"); post("\toutlet1: output signal\n"); } static void *nlms_tilde_new(t_symbol *s, int argc, t_atom *argv) { t_nlms_tilde *x = (t_nlms_tilde *)pd_new(nlms_tilde_class); int i; // default values: x->N = 8; x->mu = 0.05; x->alpha = 0.0001; x->adapt = 0; x->tmp = NULL; x->bufsize = 0; switch(argc) { case 2: x->mu = atom_getfloat(argv+1); case 1: x->N = atom_getint(argv); x->N = (x->N<=0) ? 1 : x->N; } // allocate mem and init coefficients x->c = (t_float *)getbytes(sizeof(t_float) * x->N); for(i=0; iN; i++) x->c[i] = 0; // allocate mem for temp buffer x->buf = (t_sample *)getbytes(sizeof(t_sample) * x->N-1); for(i=0; iN-1; i++) x->buf[i] = 0; inlet_new(&x->x_obj, &x->x_obj.ob_pd, &s_signal, &s_signal); outlet_new(&x->x_obj, &s_signal); x->x_canvas = canvas_getcurrent(); return (x); } static void nlms_tilde_free(t_nlms_tilde *x) { if(x->c) freebytes(x->c, sizeof(t_float) * x->N); if(x->buf) freebytes(x->buf, sizeof(t_sample) * x->N-1); if(x->tmp) freebytes(x->tmp, sizeof(t_sample) * x->bufsize); } void nlms_tilde_setup(void) { nlms_tilde_class = class_new(gensym("nlms~"), (t_newmethod)nlms_tilde_new, (t_method)nlms_tilde_free, sizeof(t_nlms_tilde), CLASS_DEFAULT, A_GIMME, 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_a, gensym("adaptation"), A_DEFFLOAT, 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_geta, gensym("getadaptation"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_mu, gensym("mu"), A_DEFFLOAT, 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_getmu, gensym("getmu"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_alpha, gensym("alpha"), A_DEFFLOAT, 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_getalpha, gensym("getalpha"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_getN, gensym("getN"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_init, gensym("init_unity"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_clear, gensym("clear"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_print, gensym("print"), 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_write, gensym("write"), A_DEFSYMBOL, 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_read, gensym("read"), A_DEFSYMBOL, 0); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_dsp, gensym("dsp"), 0); CLASS_MAINSIGNALIN(nlms_tilde_class, t_nlms_tilde, f); class_addmethod(nlms_tilde_class, (t_method)nlms_tilde_helper, gensym("help"), 0); }