Use MSGT_DECVIDEO in a video decoder.
[mplayer/glamo.git] / libaf / filter.c
blobc5ab0391307221f888988dba2bc7ed6b48c8d029
1 /*
2 * design and implementation of different types of digital filters
4 * Copyright (C) 2001 Anders Johansson ajh@atri.curtin.edu.au
6 * This file is part of MPlayer.
8 * MPlayer is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * MPlayer is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License along
19 * with MPlayer; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 #include <string.h>
24 #include <math.h>
25 #include "dsp.h"
27 /******************************************************************************
28 * FIR filter implementations
29 ******************************************************************************/
31 /* C implementation of FIR filter y=w*x
33 n number of filter taps, where mod(n,4)==0
34 w filter taps
35 x input signal must be a circular buffer which is indexed backwards
37 inline FLOAT_TYPE af_filter_fir(register unsigned int n, const FLOAT_TYPE* w,
38 const FLOAT_TYPE* x)
40 register FLOAT_TYPE y; // Output
41 y = 0.0;
42 do{
43 n--;
44 y+=w[n]*x[n];
45 }while(n != 0);
46 return y;
49 /* C implementation of parallel FIR filter y(k)=w(k) * x(k) (where * denotes convolution)
51 n number of filter taps, where mod(n,4)==0
52 d number of filters
53 xi current index in xq
54 w filter taps k by n big
55 x input signal must be a circular buffers which are indexed backwards
56 y output buffer
57 s output buffer stride
59 FLOAT_TYPE* af_filter_pfir(unsigned int n, unsigned int d, unsigned int xi,
60 const FLOAT_TYPE** w, const FLOAT_TYPE** x, FLOAT_TYPE* y,
61 unsigned int s)
63 register const FLOAT_TYPE* xt = *x + xi;
64 register const FLOAT_TYPE* wt = *w;
65 register int nt = 2*n;
66 while(d-- > 0){
67 *y = af_filter_fir(n,wt,xt);
68 wt+=n;
69 xt+=nt;
70 y+=s;
72 return y;
75 /* Add new data to circular queue designed to be used with a parallel
76 FIR filter, with d filters. xq is the circular queue, in pointing
77 at the new samples, xi current index in xq and n the length of the
78 filter. xq must be n*2 by k big, s is the index for in.
80 int af_filter_updatepq(unsigned int n, unsigned int d, unsigned int xi,
81 FLOAT_TYPE** xq, const FLOAT_TYPE* in, unsigned int s)
83 register FLOAT_TYPE* txq = *xq + xi;
84 register int nt = n*2;
86 while(d-- >0){
87 *txq= *(txq+n) = *in;
88 txq+=nt;
89 in+=s;
91 return (++xi)&(n-1);
94 /******************************************************************************
95 * FIR filter design
96 ******************************************************************************/
98 /* Design FIR filter using the Window method
100 n filter length must be odd for HP and BS filters
101 w buffer for the filter taps (must be n long)
102 fc cutoff frequencies (1 for LP and HP, 2 for BP and BS)
103 0 < fc < 1 where 1 <=> Fs/2
104 flags window and filter type as defined in filter.h
105 variables are ored together: i.e. LP|HAMMING will give a
106 low pass filter designed using a hamming window
107 opt beta constant used only when designing using kaiser windows
109 returns 0 if OK, -1 if fail
111 int af_filter_design_fir(unsigned int n, FLOAT_TYPE* w, const FLOAT_TYPE* fc,
112 unsigned int flags, FLOAT_TYPE opt)
114 unsigned int o = n & 1; // Indicator for odd filter length
115 unsigned int end = ((n + 1) >> 1) - o; // Loop end
116 unsigned int i; // Loop index
118 FLOAT_TYPE k1 = 2 * M_PI; // 2*pi*fc1
119 FLOAT_TYPE k2 = 0.5 * (FLOAT_TYPE)(1 - o);// Constant used if the filter has even length
120 FLOAT_TYPE k3; // 2*pi*fc2 Constant used in BP and BS design
121 FLOAT_TYPE g = 0.0; // Gain
122 FLOAT_TYPE t1,t2,t3; // Temporary variables
123 FLOAT_TYPE fc1,fc2; // Cutoff frequencies
125 // Sanity check
126 if(!w || (n == 0)) return -1;
128 // Get window coefficients
129 switch(flags & WINDOW_MASK){
130 case(BOXCAR):
131 af_window_boxcar(n,w); break;
132 case(TRIANG):
133 af_window_triang(n,w); break;
134 case(HAMMING):
135 af_window_hamming(n,w); break;
136 case(HANNING):
137 af_window_hanning(n,w); break;
138 case(BLACKMAN):
139 af_window_blackman(n,w); break;
140 case(FLATTOP):
141 af_window_flattop(n,w); break;
142 case(KAISER):
143 af_window_kaiser(n,w,opt); break;
144 default:
145 return -1;
148 if(flags & (LP | HP)){
149 fc1=*fc;
150 // Cutoff frequency must be < 0.5 where 0.5 <=> Fs/2
151 fc1 = ((fc1 <= 1.0) && (fc1 > 0.0)) ? fc1/2 : 0.25;
152 k1 *= fc1;
154 if(flags & LP){ // Low pass filter
156 // If the filter length is odd, there is one point which is exactly
157 // in the middle. The value at this point is 2*fCutoff*sin(x)/x,
158 // where x is zero. To make sure nothing strange happens, we set this
159 // value separately.
160 if (o){
161 w[end] = fc1 * w[end] * 2.0;
162 g=w[end];
165 // Create filter
166 for (i=0 ; i<end ; i++){
167 t1 = (FLOAT_TYPE)(i+1) - k2;
168 w[end-i-1] = w[n-end+i] = w[end-i-1] * sin(k1 * t1)/(M_PI * t1); // Sinc
169 g += 2*w[end-i-1]; // Total gain in filter
172 else{ // High pass filter
173 if (!o) // High pass filters must have odd length
174 return -1;
175 w[end] = 1.0 - (fc1 * w[end] * 2.0);
176 g= w[end];
178 // Create filter
179 for (i=0 ; i<end ; i++){
180 t1 = (FLOAT_TYPE)(i+1);
181 w[end-i-1] = w[n-end+i] = -1 * w[end-i-1] * sin(k1 * t1)/(M_PI * t1); // Sinc
182 g += ((i&1) ? (2*w[end-i-1]) : (-2*w[end-i-1])); // Total gain in filter
187 if(flags & (BP | BS)){
188 fc1=fc[0];
189 fc2=fc[1];
190 // Cutoff frequencies must be < 1.0 where 1.0 <=> Fs/2
191 fc1 = ((fc1 <= 1.0) && (fc1 > 0.0)) ? fc1/2 : 0.25;
192 fc2 = ((fc2 <= 1.0) && (fc2 > 0.0)) ? fc2/2 : 0.25;
193 k3 = k1 * fc2; // 2*pi*fc2
194 k1 *= fc1; // 2*pi*fc1
196 if(flags & BP){ // Band pass
197 // Calculate center tap
198 if (o){
199 g=w[end]*(fc1+fc2);
200 w[end] = (fc2 - fc1) * w[end] * 2.0;
203 // Create filter
204 for (i=0 ; i<end ; i++){
205 t1 = (FLOAT_TYPE)(i+1) - k2;
206 t2 = sin(k3 * t1)/(M_PI * t1); // Sinc fc2
207 t3 = sin(k1 * t1)/(M_PI * t1); // Sinc fc1
208 g += w[end-i-1] * (t3 + t2); // Total gain in filter
209 w[end-i-1] = w[n-end+i] = w[end-i-1] * (t2 - t3);
212 else{ // Band stop
213 if (!o) // Band stop filters must have odd length
214 return -1;
215 w[end] = 1.0 - (fc2 - fc1) * w[end] * 2.0;
216 g= w[end];
218 // Create filter
219 for (i=0 ; i<end ; i++){
220 t1 = (FLOAT_TYPE)(i+1);
221 t2 = sin(k1 * t1)/(M_PI * t1); // Sinc fc1
222 t3 = sin(k3 * t1)/(M_PI * t1); // Sinc fc2
223 w[end-i-1] = w[n-end+i] = w[end-i-1] * (t2 - t3);
224 g += 2*w[end-i-1]; // Total gain in filter
229 // Normalize gain
230 g=1/g;
231 for (i=0; i<n; i++)
232 w[i] *= g;
234 return 0;
237 /* Design polyphase FIR filter from prototype filter
239 n length of prototype filter
240 k number of polyphase components
241 w prototype filter taps
242 pw Parallel FIR filter
243 g Filter gain
244 flags FWD forward indexing
245 REW reverse indexing
246 ODD multiply every 2nd filter tap by -1 => HP filter
248 returns 0 if OK, -1 if fail
250 int af_filter_design_pfir(unsigned int n, unsigned int k, const FLOAT_TYPE* w,
251 FLOAT_TYPE** pw, FLOAT_TYPE g, unsigned int flags)
253 int l = (int)n/k; // Length of individual FIR filters
254 int i; // Counters
255 int j;
256 FLOAT_TYPE t; // g * w[i]
258 // Sanity check
259 if(l<1 || k<1 || !w || !pw)
260 return -1;
262 // Do the stuff
263 if(flags&REW){
264 for(j=l-1;j>-1;j--){//Columns
265 for(i=0;i<(int)k;i++){//Rows
266 t=g * *w++;
267 pw[i][j]=t * ((flags & ODD) ? ((j & 1) ? -1 : 1) : 1);
271 else{
272 for(j=0;j<l;j++){//Columns
273 for(i=0;i<(int)k;i++){//Rows
274 t=g * *w++;
275 pw[i][j]=t * ((flags & ODD) ? ((j & 1) ? 1 : -1) : 1);
279 return -1;
282 /******************************************************************************
283 * IIR filter design
284 ******************************************************************************/
286 /* Helper functions for the bilinear transform */
288 /* Pre-warp the coefficients of a numerator or denominator.
289 Note that a0 is assumed to be 1, so there is no wrapping
290 of it.
292 static void af_filter_prewarp(FLOAT_TYPE* a, FLOAT_TYPE fc, FLOAT_TYPE fs)
294 FLOAT_TYPE wp;
295 wp = 2.0 * fs * tan(M_PI * fc / fs);
296 a[2] = a[2]/(wp * wp);
297 a[1] = a[1]/wp;
300 /* Transform the numerator and denominator coefficients of s-domain
301 biquad section into corresponding z-domain coefficients.
303 The transfer function for z-domain is:
305 1 + alpha1 * z^(-1) + alpha2 * z^(-2)
306 H(z) = -------------------------------------
307 1 + beta1 * z^(-1) + beta2 * z^(-2)
309 Store the 4 IIR coefficients in array pointed by coef in following
310 order:
311 beta1, beta2 (denominator)
312 alpha1, alpha2 (numerator)
314 Arguments:
315 a - s-domain numerator coefficients
316 b - s-domain denominator coefficients
317 k - filter gain factor. Initially set to 1 and modified by each
318 biquad section in such a way, as to make it the
319 coefficient by which to multiply the overall filter gain
320 in order to achieve a desired overall filter gain,
321 specified in initial value of k.
322 fs - sampling rate (Hz)
323 coef - array of z-domain coefficients to be filled in.
325 Return: On return, set coef z-domain coefficients and k to the gain
326 required to maintain overall gain = 1.0;
328 static void af_filter_bilinear(const FLOAT_TYPE* a, const FLOAT_TYPE* b, FLOAT_TYPE* k,
329 FLOAT_TYPE fs, FLOAT_TYPE *coef)
331 FLOAT_TYPE ad, bd;
333 /* alpha (Numerator in s-domain) */
334 ad = 4. * a[2] * fs * fs + 2. * a[1] * fs + a[0];
335 /* beta (Denominator in s-domain) */
336 bd = 4. * b[2] * fs * fs + 2. * b[1] * fs + b[0];
338 /* Update gain constant for this section */
339 *k *= ad/bd;
341 /* Denominator */
342 *coef++ = (2. * b[0] - 8. * b[2] * fs * fs)/bd; /* beta1 */
343 *coef++ = (4. * b[2] * fs * fs - 2. * b[1] * fs + b[0])/bd; /* beta2 */
345 /* Numerator */
346 *coef++ = (2. * a[0] - 8. * a[2] * fs * fs)/ad; /* alpha1 */
347 *coef = (4. * a[2] * fs * fs - 2. * a[1] * fs + a[0])/ad; /* alpha2 */
352 /* IIR filter design using bilinear transform and prewarp. Transforms
353 2nd order s domain analog filter into a digital IIR biquad link. To
354 create a filter fill in a, b, Q and fs and make space for coef and k.
357 Example Butterworth design:
359 Below are Butterworth polynomials, arranged as a series of 2nd
360 order sections:
362 Note: n is filter order.
364 n Polynomials
365 -------------------------------------------------------------------
366 2 s^2 + 1.4142s + 1
367 4 (s^2 + 0.765367s + 1) * (s^2 + 1.847759s + 1)
368 6 (s^2 + 0.5176387s + 1) * (s^2 + 1.414214 + 1) * (s^2 + 1.931852s + 1)
370 For n=4 we have following equation for the filter transfer function:
372 T(s) = --------------------------- * ----------------------------
373 s^2 + (1/Q) * 0.765367s + 1 s^2 + (1/Q) * 1.847759s + 1
375 The filter consists of two 2nd order sections since highest s power
376 is 2. Now we can take the coefficients, or the numbers by which s
377 is multiplied and plug them into a standard formula to be used by
378 bilinear transform.
380 Our standard form for each 2nd order section is:
382 a2 * s^2 + a1 * s + a0
383 H(s) = ----------------------
384 b2 * s^2 + b1 * s + b0
386 Note that Butterworth numerator is 1 for all filter sections, which
387 means s^2 = 0 and s^1 = 0
389 Let's convert standard Butterworth polynomials into this form:
391 0 + 0 + 1 0 + 0 + 1
392 --------------------------- * --------------------------
393 1 + ((1/Q) * 0.765367) + 1 1 + ((1/Q) * 1.847759) + 1
395 Section 1:
396 a2 = 0; a1 = 0; a0 = 1;
397 b2 = 1; b1 = 0.765367; b0 = 1;
399 Section 2:
400 a2 = 0; a1 = 0; a0 = 1;
401 b2 = 1; b1 = 1.847759; b0 = 1;
403 Q is filter quality factor or resonance, in the range of 1 to
404 1000. The overall filter Q is a product of all 2nd order stages.
405 For example, the 6th order filter (3 stages, or biquads) with
406 individual Q of 2 will have filter Q = 2 * 2 * 2 = 8.
409 Arguments:
410 a - s-domain numerator coefficients, a[1] is always assumed to be 1.0
411 b - s-domain denominator coefficients
412 Q - Q value for the filter
413 k - filter gain factor. Initially set to 1 and modified by each
414 biquad section in such a way, as to make it the
415 coefficient by which to multiply the overall filter gain
416 in order to achieve a desired overall filter gain,
417 specified in initial value of k.
418 fs - sampling rate (Hz)
419 coef - array of z-domain coefficients to be filled in.
421 Note: Upon return from each call, the k argument will be set to a
422 value, by which to multiply our actual signal in order for the gain
423 to be one. On second call to szxform() we provide k that was
424 changed by the previous section. During actual audio filtering
425 k can be used for gain compensation.
427 return -1 if fail 0 if success.
429 int af_filter_szxform(const FLOAT_TYPE* a, const FLOAT_TYPE* b, FLOAT_TYPE Q, FLOAT_TYPE fc,
430 FLOAT_TYPE fs, FLOAT_TYPE *k, FLOAT_TYPE *coef)
432 FLOAT_TYPE at[3];
433 FLOAT_TYPE bt[3];
435 if(!a || !b || !k || !coef || (Q>1000.0 || Q< 1.0))
436 return -1;
438 memcpy(at,a,3*sizeof(FLOAT_TYPE));
439 memcpy(bt,b,3*sizeof(FLOAT_TYPE));
441 bt[1]/=Q;
443 /* Calculate a and b and overwrite the original values */
444 af_filter_prewarp(at, fc, fs);
445 af_filter_prewarp(bt, fc, fs);
446 /* Execute bilinear transform */
447 af_filter_bilinear(at, bt, k, fs, coef);
449 return 0;