1 /*
2  ** Copyright 2003-2010, VisualOn, Inc.
3  **
4  ** Licensed under the Apache License, Version 2.0 (the "License");
5  ** you may not use this file except in compliance with the License.
6  ** You may obtain a copy of the License at
7  **
8  **     http://www.apache.org/licenses/LICENSE-2.0
9  **
10  ** Unless required by applicable law or agreed to in writing, software
11  ** distributed under the License is distributed on an "AS IS" BASIS,
12  ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  ** See the License for the specific language governing permissions and
14  ** limitations under the License.
15  */
16 
17 /***********************************************************************
18 *      File: levinson.c                                                *
19 *                                                                      *
20 *      Description:LEVINSON-DURBIN algorithm in double precision       *
21 *                                                                      *
22 ************************************************************************/
23 /*---------------------------------------------------------------------------*
24  *                         LEVINSON.C                        *
25  *---------------------------------------------------------------------------*
26  *                                                                           *
27  *      LEVINSON-DURBIN algorithm in double precision                        *
28  *                                                                           *
29  *                                                                           *
30  * Algorithm                                                                 *
31  *                                                                           *
32  *       R[i]    autocorrelations.                                           *
33  *       A[i]    filter coefficients.                                        *
34  *       K       reflection coefficients.                                    *
35  *       Alpha   prediction gain.                                            *
36  *                                                                           *
37  *       Initialization:                                                     *
38  *               A[0] = 1                                                    *
39  *               K    = -R[1]/R[0]                                           *
40  *               A[1] = K                                                    *
41  *               Alpha = R[0] * (1-K**2]                                     *
42  *                                                                           *
43  *       Do for  i = 2 to M                                                  *
44  *                                                                           *
45  *            S =  SUM ( R[j]*A[i-j] ,j=1,i-1 ) +  R[i]                      *
46  *                                                                           *
47  *            K = -S / Alpha                                                 *
48  *                                                                           *
49  *            An[j] = A[j] + K*A[i-j]   for j=1 to i-1                       *
50  *                                      where   An[i] = new A[i]             *
51  *            An[i]=K                                                        *
52  *                                                                           *
53  *            Alpha=Alpha * (1-K**2)                                         *
54  *                                                                           *
55  *       END                                                                 *
56  *                                                                           *
57  * Remarks on the dynamics of the calculations.                              *
58  *                                                                           *
59  *       The numbers used are in double precision in the following format :  *
60  *       A = AH <<16 + AL<<1.  AH and AL are 16 bit signed integers.         *
61  *       Since the LSB's also contain a sign bit, this format does not       *
62  *       correspond to standard 32 bit integers.  We use this format since   *
63  *       it allows fast execution of multiplications and divisions.          *
64  *                                                                           *
65  *       "DPF" will refer to this special format in the following text.      *
66  *       See oper_32b.c                                                      *
67  *                                                                           *
68  *       The R[i] were normalized in routine AUTO (hence, R[i] < 1.0).       *
69  *       The K[i] and Alpha are theoretically < 1.0.                         *
70  *       The A[i], for a sampling frequency of 8 kHz, are in practice        *
71  *       always inferior to 16.0.                                            *
72  *                                                                           *
73  *       These characteristics allow straigthforward fixed-point             *
74  *       implementation.  We choose to represent the parameters as           *
75  *       follows :                                                           *
76  *                                                                           *
77  *               R[i]    Q31   +- .99..                                      *
78  *               K[i]    Q31   +- .99..                                      *
79  *               Alpha   Normalized -> mantissa in Q31 plus exponent         *
80  *               A[i]    Q27   +- 15.999..                                   *
81  *                                                                           *
82  *       The additions are performed in 32 bit.  For the summation used      *
83  *       to calculate the K[i], we multiply numbers in Q31 by numbers        *
84  *       in Q27, with the result of the multiplications in Q27,              *
85  *       resulting in a dynamic of +- 16.  This is sufficient to avoid       *
86  *       overflow, since the final result of the summation is                *
87  *       necessarily < 1.0 as both the K[i] and Alpha are                    *
88  *       theoretically < 1.0.                                                *
89  *___________________________________________________________________________*/
90 #include "typedef.h"
91 #include "basic_op.h"
92 #include "oper_32b.h"
93 #include "acelp.h"
94 
95 #define M   16
96 #define NC  (M/2)
97 
Init_Levinson(Word16 * mem)98 void Init_Levinson(
99         Word16 * mem                          /* output  :static memory (18 words) */
100         )
101 {
102     Set_zero(mem, 18);                     /* old_A[0..M-1] = 0, old_rc[0..1] = 0 */
103     return;
104 }
105 
106 
Levinson(Word16 Rh[],Word16 Rl[],Word16 A[],Word16 rc[],Word16 * mem)107 void Levinson(
108         Word16 Rh[],                          /* (i)     : Rh[M+1] Vector of autocorrelations (msb) */
109         Word16 Rl[],                          /* (i)     : Rl[M+1] Vector of autocorrelations (lsb) */
110         Word16 A[],                           /* (o) Q12 : A[M]    LPC coefficients  (m = 16)       */
111         Word16 rc[],                          /* (o) Q15 : rc[M]   Reflection coefficients.         */
112         Word16 * mem                          /* (i/o)   :static memory (18 words)                  */
113          )
114 {
115     Word32 i, j;
116     Word16 hi, lo;
117     Word16 Kh, Kl;                         /* reflection coefficient; hi and lo           */
118     Word16 alp_h, alp_l, alp_exp;          /* Prediction gain; hi lo and exponent         */
119     Word16 Ah[M + 1], Al[M + 1];           /* LPC coef. in double prec.                   */
120     Word16 Anh[M + 1], Anl[M + 1];         /* LPC coef.for next iteration in double prec. */
121     Word32 t0, t1, t2;                     /* temporary variable                          */
122     Word16 *old_A, *old_rc;
123 
124     /* Last A(z) for case of unstable filter */
125     old_A = mem;
126     old_rc = mem + M;
127 
128     /* K = A[1] = -R[1] / R[0] */
129 
130     t1 = ((Rh[1] << 16) + (Rl[1] << 1));   /* R[1] in Q31 */
131     t2 = L_abs(t1);                        /* abs R[1]         */
132     t0 = Div_32(t2, Rh[0], Rl[0]);         /* R[1]/R[0] in Q31 */
133     if (t1 > 0)
134         t0 = -t0;                          /* -R[1]/R[0]       */
135 
136     Kh = t0 >> 16;
137     Kl = (t0 & 0xffff)>>1;
138     rc[0] = Kh;
139     t0 = (t0 >> 4);                        /* A[1] in Q27      */
140 
141     Ah[1] = t0 >> 16;
142     Al[1] = (t0 & 0xffff)>>1;
143 
144     /* Alpha = R[0] * (1-K**2) */
145     t0 = Mpy_32(Kh, Kl, Kh, Kl);           /* K*K      in Q31 */
146     t0 = L_abs(t0);                        /* Some case <0 !! */
147     t0 = vo_L_sub((Word32) 0x7fffffffL, t0);  /* 1 - K*K  in Q31 */
148 
149     hi = t0 >> 16;
150     lo = (t0 & 0xffff)>>1;
151 
152     t0 = Mpy_32(Rh[0], Rl[0], hi, lo);     /* Alpha in Q31    */
153 
154     /* Normalize Alpha */
155     alp_exp = norm_l(t0);
156     t0 = (t0 << alp_exp);
157 
158     alp_h = t0 >> 16;
159     alp_l = (t0 & 0xffff)>>1;
160     /*--------------------------------------*
161      * ITERATIONS  I=2 to M                 *
162      *--------------------------------------*/
163     for (i = 2; i <= M; i++)
164     {
165         /* t0 = SUM ( R[j]*A[i-j] ,j=1,i-1 ) +  R[i] */
166         t0 = 0;
167         for (j = 1; j < i; j++)
168             t0 = vo_L_add(t0, Mpy_32(Rh[j], Rl[j], Ah[i - j], Al[i - j]));
169 
170         t0 = t0 << 4;                 /* result in Q27 -> convert to Q31 */
171         /* No overflow possible            */
172         t1 = ((Rh[i] << 16) + (Rl[i] << 1));
173         t0 = vo_L_add(t0, t1);                /* add R[i] in Q31                 */
174 
175         /* K = -t0 / Alpha */
176         t1 = L_abs(t0);
177         t2 = Div_32(t1, alp_h, alp_l);     /* abs(t0)/Alpha                   */
178         if (t0 > 0)
179             t2 = -t2;                   /* K =-t0/Alpha                    */
180         t2 = (t2 << alp_exp);           /* denormalize; compare to Alpha   */
181 
182         Kh = t2 >> 16;
183         Kl = (t2 & 0xffff)>>1;
184 
185         rc[i - 1] = Kh;
186         /* Test for unstable filter. If unstable keep old A(z) */
187         if (abs_s(Kh) > 32750)
188         {
189             A[0] = 4096;                    /* Ai[0] not stored (always 1.0) */
190             for (j = 0; j < M; j++)
191             {
192                 A[j + 1] = old_A[j];
193             }
194             rc[0] = old_rc[0];             /* only two rc coefficients are needed */
195             rc[1] = old_rc[1];
196             return;
197         }
198         /*------------------------------------------*
199          *  Compute new LPC coeff. -> An[i]         *
200          *  An[j]= A[j] + K*A[i-j]     , j=1 to i-1 *
201          *  An[i]= K                                *
202          *------------------------------------------*/
203         for (j = 1; j < i; j++)
204         {
205             t0 = Mpy_32(Kh, Kl, Ah[i - j], Al[i - j]);
206             t0 = vo_L_add(t0, ((Ah[j] << 16) + (Al[j] << 1)));
207             Anh[j] = t0 >> 16;
208             Anl[j] = (t0 & 0xffff)>>1;
209         }
210         t2 = (t2 >> 4);                 /* t2 = K in Q31 ->convert to Q27  */
211 
212         VO_L_Extract(t2, &Anh[i], &Anl[i]);   /* An[i] in Q27                    */
213 
214         /* Alpha = Alpha * (1-K**2) */
215         t0 = Mpy_32(Kh, Kl, Kh, Kl);               /* K*K      in Q31 */
216         t0 = L_abs(t0);                            /* Some case <0 !! */
217         t0 = vo_L_sub((Word32) 0x7fffffffL, t0);   /* 1 - K*K  in Q31 */
218         hi = t0 >> 16;
219         lo = (t0 & 0xffff)>>1;
220         t0 = Mpy_32(alp_h, alp_l, hi, lo); /* Alpha in Q31    */
221 
222         /* Normalize Alpha */
223         j = norm_l(t0);
224         t0 = (t0 << j);
225         alp_h = t0 >> 16;
226         alp_l = (t0 & 0xffff)>>1;
227         alp_exp += j;         /* Add normalization to alp_exp */
228 
229         /* A[j] = An[j] */
230         for (j = 1; j <= i; j++)
231         {
232             Ah[j] = Anh[j];
233             Al[j] = Anl[j];
234         }
235     }
236     /* Truncate A[i] in Q27 to Q12 with rounding */
237     A[0] = 4096;
238     for (i = 1; i <= M; i++)
239     {
240         t0 = (Ah[i] << 16) + (Al[i] << 1);
241         old_A[i - 1] = A[i] = vo_round((t0 << 1));
242     }
243     old_rc[0] = rc[0];
244     old_rc[1] = rc[1];
245 
246     return;
247 }
248 
249 
250 
251