1 /***********************************************************************
2 Copyright (c) 2006-2011, Skype Limited. All rights reserved.
3 Redistribution and use in source and binary forms, with or without
4 modification, are permitted provided that the following conditions
5 are met:
6 - Redistributions of source code must retain the above copyright notice,
7 this list of conditions and the following disclaimer.
8 - Redistributions in binary form must reproduce the above copyright
9 notice, this list of conditions and the following disclaimer in the
10 documentation and/or other materials provided with the distribution.
11 - Neither the name of Internet Society, IETF or IETF Trust, nor the
12 names of specific contributors, may be used to endorse or promote
13 products derived from this software without specific prior written
14 permission.
15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25 POSSIBILITY OF SUCH DAMAGE.
26 ***********************************************************************/
27
28 #ifdef HAVE_CONFIG_H
29 #include "config.h"
30 #endif
31
32 #include "SigProc_FLP.h"
33 #include "tuning_parameters.h"
34 #include "define.h"
35
36 #define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384*/
37
38 /* Compute reflection coefficients from input signal */
silk_burg_modified_FLP(silk_float A[],const silk_float x[],const silk_float minInvGain,const opus_int subfr_length,const opus_int nb_subfr,const opus_int D)39 silk_float silk_burg_modified_FLP( /* O returns residual energy */
40 silk_float A[], /* O prediction coefficients (length order) */
41 const silk_float x[], /* I input signal, length: nb_subfr*(D+L_sub) */
42 const silk_float minInvGain, /* I minimum inverse prediction gain */
43 const opus_int subfr_length, /* I input signal subframe length (incl. D preceding samples) */
44 const opus_int nb_subfr, /* I number of subframes stacked in x */
45 const opus_int D /* I order */
46 )
47 {
48 opus_int k, n, s, reached_max_gain;
49 double C0, invGain, num, nrg_f, nrg_b, rc, Atmp, tmp1, tmp2;
50 const silk_float *x_ptr;
51 double C_first_row[ SILK_MAX_ORDER_LPC ], C_last_row[ SILK_MAX_ORDER_LPC ];
52 double CAf[ SILK_MAX_ORDER_LPC + 1 ], CAb[ SILK_MAX_ORDER_LPC + 1 ];
53 double Af[ SILK_MAX_ORDER_LPC ];
54
55 silk_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE );
56
57 /* Compute autocorrelations, added over subframes */
58 C0 = silk_energy_FLP( x, nb_subfr * subfr_length );
59 silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( double ) );
60 for( s = 0; s < nb_subfr; s++ ) {
61 x_ptr = x + s * subfr_length;
62 for( n = 1; n < D + 1; n++ ) {
63 C_first_row[ n - 1 ] += silk_inner_product_FLP( x_ptr, x_ptr + n, subfr_length - n );
64 }
65 }
66 silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( double ) );
67
68 /* Initialize */
69 CAb[ 0 ] = CAf[ 0 ] = C0 + FIND_LPC_COND_FAC * C0 + 1e-9f;
70 invGain = 1.0f;
71 reached_max_gain = 0;
72 for( n = 0; n < D; n++ ) {
73 /* Update first row of correlation matrix (without first element) */
74 /* Update last row of correlation matrix (without last element, stored in reversed order) */
75 /* Update C * Af */
76 /* Update C * flipud(Af) (stored in reversed order) */
77 for( s = 0; s < nb_subfr; s++ ) {
78 x_ptr = x + s * subfr_length;
79 tmp1 = x_ptr[ n ];
80 tmp2 = x_ptr[ subfr_length - n - 1 ];
81 for( k = 0; k < n; k++ ) {
82 C_first_row[ k ] -= x_ptr[ n ] * x_ptr[ n - k - 1 ];
83 C_last_row[ k ] -= x_ptr[ subfr_length - n - 1 ] * x_ptr[ subfr_length - n + k ];
84 Atmp = Af[ k ];
85 tmp1 += x_ptr[ n - k - 1 ] * Atmp;
86 tmp2 += x_ptr[ subfr_length - n + k ] * Atmp;
87 }
88 for( k = 0; k <= n; k++ ) {
89 CAf[ k ] -= tmp1 * x_ptr[ n - k ];
90 CAb[ k ] -= tmp2 * x_ptr[ subfr_length - n + k - 1 ];
91 }
92 }
93 tmp1 = C_first_row[ n ];
94 tmp2 = C_last_row[ n ];
95 for( k = 0; k < n; k++ ) {
96 Atmp = Af[ k ];
97 tmp1 += C_last_row[ n - k - 1 ] * Atmp;
98 tmp2 += C_first_row[ n - k - 1 ] * Atmp;
99 }
100 CAf[ n + 1 ] = tmp1;
101 CAb[ n + 1 ] = tmp2;
102
103 /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */
104 num = CAb[ n + 1 ];
105 nrg_b = CAb[ 0 ];
106 nrg_f = CAf[ 0 ];
107 for( k = 0; k < n; k++ ) {
108 Atmp = Af[ k ];
109 num += CAb[ n - k ] * Atmp;
110 nrg_b += CAb[ k + 1 ] * Atmp;
111 nrg_f += CAf[ k + 1 ] * Atmp;
112 }
113 silk_assert( nrg_f > 0.0 );
114 silk_assert( nrg_b > 0.0 );
115
116 /* Calculate the next order reflection (parcor) coefficient */
117 rc = -2.0 * num / ( nrg_f + nrg_b );
118 silk_assert( rc > -1.0 && rc < 1.0 );
119
120 /* Update inverse prediction gain */
121 tmp1 = invGain * ( 1.0 - rc * rc );
122 if( tmp1 <= minInvGain ) {
123 /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */
124 rc = sqrt( 1.0 - minInvGain / invGain );
125 if( num > 0 ) {
126 /* Ensure adjusted reflection coefficients has the original sign */
127 rc = -rc;
128 }
129 invGain = minInvGain;
130 reached_max_gain = 1;
131 } else {
132 invGain = tmp1;
133 }
134
135 /* Update the AR coefficients */
136 for( k = 0; k < (n + 1) >> 1; k++ ) {
137 tmp1 = Af[ k ];
138 tmp2 = Af[ n - k - 1 ];
139 Af[ k ] = tmp1 + rc * tmp2;
140 Af[ n - k - 1 ] = tmp2 + rc * tmp1;
141 }
142 Af[ n ] = rc;
143
144 if( reached_max_gain ) {
145 /* Reached max prediction gain; set remaining coefficients to zero and exit loop */
146 for( k = n + 1; k < D; k++ ) {
147 Af[ k ] = 0.0;
148 }
149 break;
150 }
151
152 /* Update C * Af and C * Ab */
153 for( k = 0; k <= n + 1; k++ ) {
154 tmp1 = CAf[ k ];
155 CAf[ k ] += rc * CAb[ n - k + 1 ];
156 CAb[ n - k + 1 ] += rc * tmp1;
157 }
158 }
159
160 if( reached_max_gain ) {
161 /* Convert to silk_float */
162 for( k = 0; k < D; k++ ) {
163 A[ k ] = (silk_float)( -Af[ k ] );
164 }
165 /* Subtract energy of preceding samples from C0 */
166 for( s = 0; s < nb_subfr; s++ ) {
167 C0 -= silk_energy_FLP( x + s * subfr_length, D );
168 }
169 /* Approximate residual energy */
170 nrg_f = C0 * invGain;
171 } else {
172 /* Compute residual energy and store coefficients as silk_float */
173 nrg_f = CAf[ 0 ];
174 tmp1 = 1.0;
175 for( k = 0; k < D; k++ ) {
176 Atmp = Af[ k ];
177 nrg_f += CAf[ k + 1 ] * Atmp;
178 tmp1 += Atmp * Atmp;
179 A[ k ] = (silk_float)(-Atmp);
180 }
181 nrg_f -= FIND_LPC_COND_FAC * C0 * tmp1;
182 }
183
184 /* Return residual energy */
185 return (silk_float)nrg_f;
186 }
187