aboutsummaryrefslogtreecommitdiff
path: root/src/NLMS~.c
blob: b065afaa6e6bf22b737f4815b71863e6425cec55 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
/* For information on usage and redistribution, and for a DISCLAIMER OF ALL
* WARRANTIES, see the file, "LICENSE.txt," in this distribution.

NLMS normalized least mean square (LMS) algorithm
lib iem_adaptfilt written by Markus Noisternig & Thomas Musil 
noisternig_AT_iem.at; musil_AT_iem.at
(c) Institute of Electronic Music and Acoustics, Graz Austria 2005 */

#ifdef NT
#pragma warning( disable : 4244 )
#pragma warning( disable : 4305 )
#endif


#include "m_pd.h"
#include "iemlib.h"
#include <math.h>
#include <stdio.h>
#include <string.h>


/* ----------------------- NLMS~ ------------------------------ */
/* -- Normalized Least Mean Square (linear adaptive FIR-filter) -- */
/* -- first input:  reference signal -- */
/* -- second input: desired signal -- */
/* --  -- */

/* for further information on adaptive filter design we refer to */
/* [1] Haykin, "Adaptive Filter Theory", 4th ed, Prentice Hall */
/* [2] Benesty, "Adaptive Signal Processing", Springer */


typedef struct NLMS_tilde
{
    t_object            x_obj;
    t_symbol            *x_w_array_sym_name;
    t_float             *x_w_array_mem_beg;
    t_float             *x_io_ptr_beg[4];// memory: 2 sig-in and 2 sig-out vectors
    t_float             *x_in_hist;// start point double buffer for sig-in history
    t_int               x_rw_index;// read-write-index
    t_int               x_n_order;// order of filter
    t_int               x_update;// 2^n rounded value, downsampling of update speed
    t_float             x_beta;// learn rate [0 .. 2]
    t_float             x_gamma;// regularization
    t_float             x_msi;
} t_NLMS_tilde;

t_class *NLMS_tilde_class;

static t_float *NLMS_tilde_check_array(t_symbol *array_sym_name, t_int length)
{
    t_int n_points;
    t_garray *a;
    t_float *vec;

    if(!(a = (t_garray *)pd_findbyclass(array_sym_name, garray_class)))
    {
        error("%s: no such array for NLMS~", array_sym_name->s_name);
        return((t_float *)0);
    }
    else if(!garray_getfloatarray(a, &n_points, &vec))
    {
        error("%s: bad template for NLMS~", array_sym_name->s_name);
        return((t_float *)0);
    }
    else if(n_points < length)
    {
        error("%s: bad array-size for NLMS~: %d", array_sym_name->s_name, n_points);
        return((t_float *)0);
    }
    else
    {
        return(vec);
    }
}

static void NLMS_tilde_beta(t_NLMS_tilde *x, t_floatarg f) // learn rate
{
    if(f < 0.0f)
        f = 0.0f;
    if(f > 2.0f)
        f = 2.0f;
    
    x->x_beta = f;
}

static void NLMS_tilde_gamma(t_NLMS_tilde *x, t_floatarg f) // regularization factor (dither)
{
    if(f < 0.0f)
        f = 0.0f;
    if(f > 1.0f)
        f = 1.0f;
    
    x->x_gamma = f;
}


static void NLMS_tilde_update(t_NLMS_tilde *x, t_floatarg f) // downsample learn-rate
{
    t_int i=1, u = (t_int)f;
    
    if(u < 0)
        u = 0;
    else
    {
        while(i <= u)   // convert u for 2^N
            i *= 2;     // round downwards
        i /= 2;
        u = i;
    }
    x->x_update = u;
}

/* ============== DSP ======================= */

static t_int *NLMS_tilde_perform_zero(t_int *w)
{
    t_NLMS_tilde *x = (t_NLMS_tilde *)(w[1]);
    t_int n = (t_int)(w[2]);
    
    t_float **io = x->x_io_ptr_beg;
    t_float *out;
    t_int i, j;
    
    for(j=0; j<2; j++)/* output-vector-row */
    {
        out = io[j+2];
        for(i=0; i<n; i++)
        {
            *out++ = 0.0f;
        }
    }
    return (w+3);
}

static t_int *NLMS_tilde_perform(t_int *w)
{
    t_NLMS_tilde *x = (t_NLMS_tilde *)(w[1]);
    t_int n = (t_int)(w[2]);
    t_int n_order = x->x_n_order;   /* number of filter-order */
    t_int rw_index = x->x_rw_index;
    t_float *in = x->x_io_ptr_beg[0];// first sig in
    t_float *desired_in = x->x_io_ptr_beg[1], din;// second sig in
    t_float *filt_out = x->x_io_ptr_beg[2];// first sig out
    t_float *err_out = x->x_io_ptr_beg[3], eout;// second sig out
    t_float *write_in_hist1 = x->x_in_hist;
    t_float *write_in_hist2 = write_in_hist1+n_order;
    t_float *read_in_hist = write_in_hist2;
    t_float *w_filt_coeff = x->x_w_array_mem_beg;
    t_float my, my_err, sum;
    t_float beta = x->x_beta;
    t_float gamma = x->x_gamma;
    t_int i, j, update_counter;
    t_int update = x->x_update;
    t_int ord8=n_order&0xfffffff8;
    t_int ord_residual=n_order&0x7;
    
    if(!w_filt_coeff)
        goto NLMS_tildeperfzero;// this is quick&dirty Musil/Miller style
    
    for(i=0, update_counter=0; i<n; i++)// store history and convolve
    {
        write_in_hist1[rw_index] = in[i]; // save inputs to variable & history
        write_in_hist2[rw_index] = in[i];
        din = desired_in[i];
        
		// begin convolution
        sum = 0.0f;
        w_filt_coeff = x->x_w_array_mem_beg; // Musil's special convolution buffer struct
        read_in_hist = &write_in_hist2[rw_index];
        for(j=0; j<ord8; j+=8)	// loop unroll 8 taps
        {
            sum += w_filt_coeff[0] * read_in_hist[0];
            sum += w_filt_coeff[1] * read_in_hist[-1];
            sum += w_filt_coeff[2] * read_in_hist[-2];
            sum += w_filt_coeff[3] * read_in_hist[-3];
            sum += w_filt_coeff[4] * read_in_hist[-4];
            sum += w_filt_coeff[5] * read_in_hist[-5];
            sum += w_filt_coeff[6] * read_in_hist[-6];
            sum += w_filt_coeff[7] * read_in_hist[-7];
            w_filt_coeff += 8;
            read_in_hist -= 8;
        }
        for(j=0; j<ord_residual; j++)	// for filter order < 2^N
            sum += w_filt_coeff[j] * read_in_hist[-j];
        
        filt_out[i] = sum;
        eout = din - filt_out[i];	// buffer-struct for further use
        err_out[i] = eout;
        
        if(update)	// downsampling for learn rate
        {
            update_counter++;
            if(update_counter >= update)
            {
                update_counter = 0;
                
                sum = 0.0f;// calculate energy for last n-order samples in filter
                read_in_hist = &write_in_hist2[rw_index];
                for(j=0; j<ord8; j+=8)	// unrolling quadrature calc
                {
                    sum += read_in_hist[0] * read_in_hist[0];
                    sum += read_in_hist[-1] * read_in_hist[-1];
                    sum += read_in_hist[-2] * read_in_hist[-2];
                    sum += read_in_hist[-3] * read_in_hist[-3];
                    sum += read_in_hist[-4] * read_in_hist[-4];
                    sum += read_in_hist[-5] * read_in_hist[-5];
                    sum += read_in_hist[-6] * read_in_hist[-6];
                    sum += read_in_hist[-7] * read_in_hist[-7];
                    read_in_hist -= 8;
                }
                for(j=0; j<ord_residual; j++)	// residual
                    sum += read_in_hist[-j] * read_in_hist[-j]; // [-j] only valid for Musil's double buffer structure
                sum += gamma * gamma * (float)n_order; // convert gamma corresponding to filter order
                my = beta / sum;// calculate mue
                
                
                my_err = my * eout;
                w_filt_coeff = x->x_w_array_mem_beg; // coefficient constraints
                read_in_hist = &write_in_hist2[rw_index];
                for(j=0; j<n_order; j++) // without unroll
                    w_filt_coeff[j] += read_in_hist[-j] * my_err;
            }
        }
        rw_index++;
        if(rw_index >= n_order)
            rw_index -= n_order;
    }

    x->x_rw_index = rw_index; // back to start

    return(w+3);
    
NLMS_tildeperfzero:
    
    while(n--)
    {
        *filt_out++ = 0.0f;
        *err_out++ = 0.0f;
    }
    return(w+3);
}

static void NLMS_tilde_dsp(t_NLMS_tilde *x, t_signal **sp)
{
    t_int i, n = sp[0]->s_n;
    
    for(i=0; i<4; i++) // store io_vec
        x->x_io_ptr_beg[i] = sp[i]->s_vec;
    
    x->x_w_array_mem_beg = NLMS_tilde_check_array(x->x_w_array_sym_name, x->x_n_order);

    if(!x->x_w_array_mem_beg)
        dsp_add(NLMS_tilde_perform_zero, 2, x, n);
    else
        dsp_add(NLMS_tilde_perform, 2, x, n);
}


/* setup/setdown things */

static void NLMS_tilde_free(t_NLMS_tilde *x)
{
    freebytes(x->x_in_hist, 2*x->x_n_order*sizeof(t_float));
}

static void *NLMS_tilde_new(t_symbol *s, t_int argc, t_atom *argv)
{
    t_NLMS_tilde *x = (t_NLMS_tilde *)pd_new(NLMS_tilde_class);
    t_int i, n_order=39;
    t_symbol    *w_name;
    t_float beta=0.1f;
    t_float gamma=0.00001f;
    
    if((argc >= 4) &&
        IS_A_FLOAT(argv,0) &&   //IS_A_FLOAT/SYMBOL from iemlib.h
        IS_A_FLOAT(argv,1) &&
        IS_A_FLOAT(argv,2) &&
        IS_A_SYMBOL(argv,3))
    {
        n_order = (t_int)atom_getintarg(0, argc, argv);
        beta    = (t_float)atom_getfloatarg(1, argc, argv);
        gamma   = (t_float)atom_getfloatarg(2, argc, argv);
        w_name  = (t_symbol *)atom_getsymbolarg(3, argc, argv);
        
        if(beta < 0.0f)
            beta = 0.0f;
        if(beta > 2.0f)
            beta = 2.0f;
        
        if(gamma < 0.0f)
            gamma = 0.0f;
        if(gamma > 1.0f)
            gamma = 1.0f;
        
        if(n_order < 2)
            n_order = 2;
        if(n_order > 11111)
            n_order = 11111;
        
        inlet_new(&x->x_obj, &x->x_obj.ob_pd, &s_signal, &s_signal);
        outlet_new(&x->x_obj, &s_signal);
        outlet_new(&x->x_obj, &s_signal);
        
        x->x_msi = 0;
        x->x_n_order = n_order;
        x->x_update = 0;
        x->x_beta = beta;
        x->x_gamma = gamma;
        // 2 times in and one time desired_in memory allocation (history)
        x->x_in_hist = (t_float *)getbytes(2*x->x_n_order*sizeof(t_float));
        
        // table-symbols will be linked to their memory in future (dsp_routine)
        x->x_w_array_sym_name = gensym(w_name->s_name);
        x->x_w_array_mem_beg = (t_float *)0;
        
        return(x);
    }
    else
    {
        post("NLMS~-ERROR: need 3 float- + 1 symbol-arguments:");
        post("  order_of_filter + learnrate_beta + security_value + array_name_taps");
        return(0);
    }
}

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), 0, A_GIMME, 0);
    CLASS_MAINSIGNALIN(NLMS_tilde_class, t_NLMS_tilde, x_msi);
    class_addmethod(NLMS_tilde_class, (t_method)NLMS_tilde_dsp, gensym("dsp"), 0);
    class_addmethod(NLMS_tilde_class, (t_method)NLMS_tilde_update, gensym("update"), A_FLOAT, 0); // method: downsampling factor of learning (multiple of 2^N)
    class_addmethod(NLMS_tilde_class, (t_method)NLMS_tilde_beta, gensym("beta"), A_FLOAT, 0); //method: normalized learning rate
    class_addmethod(NLMS_tilde_class, (t_method)NLMS_tilde_gamma, gensym("gamma"), A_FLOAT, 0);   // method: dithering noise related to signal

    //class_sethelpsymbol(NLMS_tilde_class, gensym("iemhelp2/NLMS~"));
}