comparison libtwamr/r_fft.c @ 412:810ac4b99025

libtwamr: integrate VAD2 r_fft.c
author Mychaela Falconia <falcon@freecalypso.org>
date Tue, 07 May 2024 01:22:02 +0000
parents
children
comparison
equal deleted inserted replaced
411:bd5614fc780a 412:810ac4b99025
1 /*
2 *****************************************************************************
3 *
4 * GSM AMR-NB speech codec R98 Version 7.6.0 December 12, 2001
5 * R99 Version 3.3.0
6 * REL-4 Version 4.1.0
7 *
8 *****************************************************************************
9 *
10 * File : r_fft.c
11 * Purpose : Fast Fourier Transform (FFT) algorithm
12 *
13 *****************************************************************************
14 */
15
16 /*****************************************************************
17 *
18 * This is an implementation of decimation-in-time FFT algorithm for
19 * real sequences. The techniques used here can be found in several
20 * books, e.g., i) Proakis and Manolakis, "Digital Signal Processing",
21 * 2nd Edition, Chapter 9, and ii) W.H. Press et. al., "Numerical
22 * Recipes in C", 2nd Ediiton, Chapter 12.
23 *
24 * Input - There is one input to this function:
25 *
26 * 1) An integer pointer to the input data array
27 *
28 * Output - There is no return value.
29 * The input data are replaced with transformed data. If the
30 * input is a real time domain sequence, it is replaced with
31 * the complex FFT for positive frequencies. The FFT value
32 * for DC and the foldover frequency are combined to form the
33 * first complex number in the array. The remaining complex
34 * numbers correspond to increasing frequencies. If the input
35 * is a complex frequency domain sequence arranged as above,
36 * it is replaced with the corresponding time domain sequence.
37 *
38 * Notes:
39 *
40 * 1) This function is designed to be a part of a VAD
41 * algorithm that requires 128-point FFT of real
42 * sequences. This is achieved here through a 64-point
43 * complex FFT. Consequently, the FFT size information is
44 * not transmitted explicitly. However, some flexibility
45 * is provided in the function to change the size of the
46 * FFT by specifying the size information through "define"
47 * statements.
48 *
49 * 2) The values of the complex sinusoids used in the FFT
50 * algorithm are stored in a ROM table.
51 *
52 * 3) In the c_fft function, the FFT values are divided by
53 * 2 after each stage of computation thus dividing the
54 * final FFT values by 64. This is somewhat different
55 * from the usual definition of FFT where the factor 1/N,
56 * i.e., 1/64, used for the IFFT and not the FFT. No factor
57 * is used in the r_fft function.
58 *
59 *****************************************************************/
60
61 #include "tw_amr.h"
62 #include "namespace.h"
63 #include "typedef.h"
64 #include "cnst.h"
65 #include "basic_op.h"
66 #include "oper_32b.h"
67 #include "no_count.h"
68 #include "vad2.h"
69
70 #define SIZE 128
71 #define SIZE_BY_TWO 64
72 #define NUM_STAGE 6
73 #define TRUE 1
74 #define FALSE 0
75
76 static const Word16 phs_tbl[] =
77 {
78
79 32767, 0, 32729, -1608, 32610, -3212, 32413, -4808,
80 32138, -6393, 31786, -7962, 31357, -9512, 30853, -11039,
81 30274, -12540, 29622, -14010, 28899, -15447, 28106, -16846,
82 27246, -18205, 26320, -19520, 25330, -20788, 24279, -22006,
83 23170, -23170, 22006, -24279, 20788, -25330, 19520, -26320,
84 18205, -27246, 16846, -28106, 15447, -28899, 14010, -29622,
85 12540, -30274, 11039, -30853, 9512, -31357, 7962, -31786,
86 6393, -32138, 4808, -32413, 3212, -32610, 1608, -32729,
87 0, -32768, -1608, -32729, -3212, -32610, -4808, -32413,
88 -6393, -32138, -7962, -31786, -9512, -31357, -11039, -30853,
89 -12540, -30274, -14010, -29622, -15447, -28899, -16846, -28106,
90 -18205, -27246, -19520, -26320, -20788, -25330, -22006, -24279,
91 -23170, -23170, -24279, -22006, -25330, -20788, -26320, -19520,
92 -27246, -18205, -28106, -16846, -28899, -15447, -29622, -14010,
93 -30274, -12540, -30853, -11039, -31357, -9512, -31786, -7962,
94 -32138, -6393, -32413, -4808, -32610, -3212, -32729, -1608
95
96 };
97
98 static const Word16 ii_table[] =
99 {SIZE / 2, SIZE / 4, SIZE / 8, SIZE / 16, SIZE / 32, SIZE / 64};
100
101 /* FFT function for complex sequences */
102
103 /*
104 * The decimation-in-time complex FFT is implemented below.
105 * The input complex numbers are presented as real part followed by
106 * imaginary part for each sample. The counters are therefore
107 * incremented by two to access the complex valued samples.
108 */
109
110 static void c_fft(Word16 * farray_ptr)
111 {
112
113 Word16 i, j, k, ii, jj, kk, ji, kj, ii2;
114 Word32 ftmp, ftmp_real, ftmp_imag;
115 Word16 tmp, tmp1, tmp2;
116
117 /* Rearrange the input array in bit reversed order */
118 for (i = 0, j = 0; i < SIZE - 2; i = i + 2)
119 { test();
120 if (sub(j, i) > 0)
121 {
122 ftmp = *(farray_ptr + i); move16();
123 *(farray_ptr + i) = *(farray_ptr + j); move16();
124 *(farray_ptr + j) = ftmp; move16();
125
126 ftmp = *(farray_ptr + i + 1); move16();
127 *(farray_ptr + i + 1) = *(farray_ptr + j + 1); move16();
128 *(farray_ptr + j + 1) = ftmp; move16();
129 }
130
131 k = SIZE_BY_TWO; move16();
132 test();
133 while (sub(j, k) >= 0)
134 {
135 j = sub(j, k);
136 k = shr(k, 1);
137 }
138 j = add(j, k);
139 }
140
141 /* The FFT part */
142 for (i = 0; i < NUM_STAGE; i++)
143 { /* i is stage counter */
144 jj = shl(2, i); /* FFT size */
145 kk = shl(jj, 1); /* 2 * FFT size */
146 ii = ii_table[i]; /* 2 * number of FFT's */ move16();
147 ii2 = shl(ii, 1);
148 ji = 0; /* ji is phase table index */ move16();
149
150 for (j = 0; j < jj; j = j + 2)
151 { /* j is sample counter */
152
153 for (k = j; k < SIZE; k = k + kk)
154 { /* k is butterfly top */
155 kj = add(k, jj); /* kj is butterfly bottom */
156
157 /* Butterfly computations */
158 ftmp_real = L_mult(*(farray_ptr + kj), phs_tbl[ji]);
159 ftmp_real = L_msu(ftmp_real, *(farray_ptr + kj + 1), phs_tbl[ji + 1]);
160
161 ftmp_imag = L_mult(*(farray_ptr + kj + 1), phs_tbl[ji]);
162 ftmp_imag = L_mac(ftmp_imag, *(farray_ptr + kj), phs_tbl[ji + 1]);
163
164 tmp1 = round(ftmp_real);
165 tmp2 = round(ftmp_imag);
166
167 tmp = sub(*(farray_ptr + k), tmp1);
168 *(farray_ptr + kj) = shr(tmp, 1); move16();
169
170 tmp = sub(*(farray_ptr + k + 1), tmp2);
171 *(farray_ptr + kj + 1) = shr(tmp, 1); move16();
172
173 tmp = add(*(farray_ptr + k), tmp1);
174 *(farray_ptr + k) = shr(tmp, 1); move16();
175
176 tmp = add(*(farray_ptr + k + 1), tmp2);
177 *(farray_ptr + k + 1) = shr(tmp, 1); move16();
178 }
179
180 ji = add(ji, ii2);
181 }
182 }
183 } /* end of c_fft () */
184
185
186
187 void r_fft(Word16 * farray_ptr)
188 {
189
190 Word16 ftmp1_real, ftmp1_imag, ftmp2_real, ftmp2_imag;
191 Word32 Lftmp1_real, Lftmp1_imag;
192 Word16 i, j;
193 Word32 Ltmp1;
194
195 /* Perform the complex FFT */
196 c_fft(farray_ptr);
197
198 /* First, handle the DC and foldover frequencies */
199 ftmp1_real = *farray_ptr; move16();
200 ftmp2_real = *(farray_ptr + 1); move16();
201 *farray_ptr = add(ftmp1_real, ftmp2_real); move16();
202 *(farray_ptr + 1) = sub(ftmp1_real, ftmp2_real); move16();
203
204 /* Now, handle the remaining positive frequencies */
205 for (i = 2, j = SIZE - i; i <= SIZE_BY_TWO; i = i + 2, j = SIZE - i)
206 {
207 ftmp1_real = add(*(farray_ptr + i), *(farray_ptr + j));
208 ftmp1_imag = sub(*(farray_ptr + i + 1), *(farray_ptr + j + 1));
209 ftmp2_real = add(*(farray_ptr + i + 1), *(farray_ptr + j + 1));
210 ftmp2_imag = sub(*(farray_ptr + j), *(farray_ptr + i));
211
212 Lftmp1_real = L_deposit_h(ftmp1_real);
213 Lftmp1_imag = L_deposit_h(ftmp1_imag);
214
215 Ltmp1 = L_mac(Lftmp1_real, ftmp2_real, phs_tbl[i]);
216 Ltmp1 = L_msu(Ltmp1, ftmp2_imag, phs_tbl[i + 1]);
217 *(farray_ptr + i) = round(L_shr(Ltmp1, 1)); move16();
218
219 Ltmp1 = L_mac(Lftmp1_imag, ftmp2_imag, phs_tbl[i]);
220 Ltmp1 = L_mac(Ltmp1, ftmp2_real, phs_tbl[i + 1]);
221 *(farray_ptr + i + 1) = round(L_shr(Ltmp1, 1)); move16();
222
223 Ltmp1 = L_mac(Lftmp1_real, ftmp2_real, phs_tbl[j]);
224 Ltmp1 = L_mac(Ltmp1, ftmp2_imag, phs_tbl[j + 1]);
225 *(farray_ptr + j) = round(L_shr(Ltmp1, 1)); move16();
226
227 Ltmp1 = L_negate(Lftmp1_imag);
228 Ltmp1 = L_msu(Ltmp1, ftmp2_imag, phs_tbl[j]);
229 Ltmp1 = L_mac(Ltmp1, ftmp2_real, phs_tbl[j + 1]);
230 *(farray_ptr + j + 1) = round(L_shr(Ltmp1, 1)); move16();
231
232 }
233 } /* end r_fft () */