提交 | 用户 | age
|
bfc108
|
1 |
/* ----------------------------------------------------------------------
|
Q |
2 |
* Copyright (C) 2010-2014 ARM Limited. All rights reserved.
|
|
3 |
*
|
|
4 |
* $Date: 19. March 2015
|
|
5 |
* $Revision: V.1.4.5
|
|
6 |
*
|
|
7 |
* Project: CMSIS DSP Library
|
|
8 |
* Title: arm_cfft_f32.c
|
|
9 |
*
|
|
10 |
* Description: Combined Radix Decimation in Frequency CFFT Floating point processing function
|
|
11 |
*
|
|
12 |
* Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
|
|
13 |
*
|
|
14 |
* Redistribution and use in source and binary forms, with or without
|
|
15 |
* modification, are permitted provided that the following conditions
|
|
16 |
* are met:
|
|
17 |
* - Redistributions of source code must retain the above copyright
|
|
18 |
* notice, this list of conditions and the following disclaimer.
|
|
19 |
* - Redistributions in binary form must reproduce the above copyright
|
|
20 |
* notice, this list of conditions and the following disclaimer in
|
|
21 |
* the documentation and/or other materials provided with the
|
|
22 |
* distribution.
|
|
23 |
* - Neither the name of ARM LIMITED nor the names of its contributors
|
|
24 |
* may be used to endorse or promote products derived from this
|
|
25 |
* software without specific prior written permission.
|
|
26 |
*
|
|
27 |
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
28 |
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
29 |
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
|
|
30 |
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
|
|
31 |
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
|
|
32 |
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
|
|
33 |
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
|
|
34 |
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
|
|
35 |
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
36 |
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
|
|
37 |
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
|
38 |
* POSSIBILITY OF SUCH DAMAGE.
|
|
39 |
* -------------------------------------------------------------------- */
|
|
40 |
|
|
41 |
#include "arm_math.h"
|
|
42 |
#include "arm_common_tables.h"
|
|
43 |
|
|
44 |
extern void arm_radix8_butterfly_f32(
|
|
45 |
float32_t * pSrc,
|
|
46 |
uint16_t fftLen,
|
|
47 |
const float32_t * pCoef,
|
|
48 |
uint16_t twidCoefModifier);
|
|
49 |
|
|
50 |
extern void arm_bitreversal_32(
|
|
51 |
uint32_t * pSrc,
|
|
52 |
const uint16_t bitRevLen,
|
|
53 |
const uint16_t * pBitRevTable);
|
|
54 |
|
|
55 |
/**
|
|
56 |
* @ingroup groupTransforms
|
|
57 |
*/
|
|
58 |
|
|
59 |
/**
|
|
60 |
* @defgroup ComplexFFT Complex FFT Functions
|
|
61 |
*
|
|
62 |
* \par
|
|
63 |
* The Fast Fourier Transform (FFT) is an efficient algorithm for computing the
|
|
64 |
* Discrete Fourier Transform (DFT). The FFT can be orders of magnitude faster
|
|
65 |
* than the DFT, especially for long lengths.
|
|
66 |
* The algorithms described in this section
|
|
67 |
* operate on complex data. A separate set of functions is devoted to handling
|
|
68 |
* of real sequences.
|
|
69 |
* \par
|
|
70 |
* There are separate algorithms for handling floating-point, Q15, and Q31 data
|
|
71 |
* types. The algorithms available for each data type are described next.
|
|
72 |
* \par
|
|
73 |
* The FFT functions operate in-place. That is, the array holding the input data
|
|
74 |
* will also be used to hold the corresponding result. The input data is complex
|
|
75 |
* and contains <code>2*fftLen</code> interleaved values as shown below.
|
|
76 |
* <pre> {real[0], imag[0], real[1], imag[1],..} </pre>
|
|
77 |
* The FFT result will be contained in the same array and the frequency domain
|
|
78 |
* values will have the same interleaving.
|
|
79 |
*
|
|
80 |
* \par Floating-point
|
|
81 |
* The floating-point complex FFT uses a mixed-radix algorithm. Multiple radix-8
|
|
82 |
* stages are performed along with a single radix-2 or radix-4 stage, as needed.
|
|
83 |
* The algorithm supports lengths of [16, 32, 64, ..., 4096] and each length uses
|
|
84 |
* a different twiddle factor table.
|
|
85 |
* \par
|
|
86 |
* The function uses the standard FFT definition and output values may grow by a
|
|
87 |
* factor of <code>fftLen</code> when computing the forward transform. The
|
|
88 |
* inverse transform includes a scale of <code>1/fftLen</code> as part of the
|
|
89 |
* calculation and this matches the textbook definition of the inverse FFT.
|
|
90 |
* \par
|
|
91 |
* Pre-initialized data structures containing twiddle factors and bit reversal
|
|
92 |
* tables are provided and defined in <code>arm_const_structs.h</code>. Include
|
|
93 |
* this header in your function and then pass one of the constant structures as
|
|
94 |
* an argument to arm_cfft_f32. For example:
|
|
95 |
* \par
|
|
96 |
* <code>arm_cfft_f32(arm_cfft_sR_f32_len64, pSrc, 1, 1)</code>
|
|
97 |
* \par
|
|
98 |
* computes a 64-point inverse complex FFT including bit reversal.
|
|
99 |
* The data structures are treated as constant data and not modified during the
|
|
100 |
* calculation. The same data structure can be reused for multiple transforms
|
|
101 |
* including mixing forward and inverse transforms.
|
|
102 |
* \par
|
|
103 |
* Earlier releases of the library provided separate radix-2 and radix-4
|
|
104 |
* algorithms that operated on floating-point data. These functions are still
|
|
105 |
* provided but are deprecated. The older functions are slower and less general
|
|
106 |
* than the new functions.
|
|
107 |
* \par
|
|
108 |
* An example of initialization of the constants for the arm_cfft_f32 function follows:
|
|
109 |
* \code
|
|
110 |
* const static arm_cfft_instance_f32 *S;
|
|
111 |
* ...
|
|
112 |
* switch (length) {
|
|
113 |
* case 16:
|
|
114 |
* S = &arm_cfft_sR_f32_len16;
|
|
115 |
* break;
|
|
116 |
* case 32:
|
|
117 |
* S = &arm_cfft_sR_f32_len32;
|
|
118 |
* break;
|
|
119 |
* case 64:
|
|
120 |
* S = &arm_cfft_sR_f32_len64;
|
|
121 |
* break;
|
|
122 |
* case 128:
|
|
123 |
* S = &arm_cfft_sR_f32_len128;
|
|
124 |
* break;
|
|
125 |
* case 256:
|
|
126 |
* S = &arm_cfft_sR_f32_len256;
|
|
127 |
* break;
|
|
128 |
* case 512:
|
|
129 |
* S = &arm_cfft_sR_f32_len512;
|
|
130 |
* break;
|
|
131 |
* case 1024:
|
|
132 |
* S = &arm_cfft_sR_f32_len1024;
|
|
133 |
* break;
|
|
134 |
* case 2048:
|
|
135 |
* S = &arm_cfft_sR_f32_len2048;
|
|
136 |
* break;
|
|
137 |
* case 4096:
|
|
138 |
* S = &arm_cfft_sR_f32_len4096;
|
|
139 |
* break;
|
|
140 |
* }
|
|
141 |
* \endcode
|
|
142 |
* \par Q15 and Q31
|
|
143 |
* The floating-point complex FFT uses a mixed-radix algorithm. Multiple radix-4
|
|
144 |
* stages are performed along with a single radix-2 stage, as needed.
|
|
145 |
* The algorithm supports lengths of [16, 32, 64, ..., 4096] and each length uses
|
|
146 |
* a different twiddle factor table.
|
|
147 |
* \par
|
|
148 |
* The function uses the standard FFT definition and output values may grow by a
|
|
149 |
* factor of <code>fftLen</code> when computing the forward transform. The
|
|
150 |
* inverse transform includes a scale of <code>1/fftLen</code> as part of the
|
|
151 |
* calculation and this matches the textbook definition of the inverse FFT.
|
|
152 |
* \par
|
|
153 |
* Pre-initialized data structures containing twiddle factors and bit reversal
|
|
154 |
* tables are provided and defined in <code>arm_const_structs.h</code>. Include
|
|
155 |
* this header in your function and then pass one of the constant structures as
|
|
156 |
* an argument to arm_cfft_q31. For example:
|
|
157 |
* \par
|
|
158 |
* <code>arm_cfft_q31(arm_cfft_sR_q31_len64, pSrc, 1, 1)</code>
|
|
159 |
* \par
|
|
160 |
* computes a 64-point inverse complex FFT including bit reversal.
|
|
161 |
* The data structures are treated as constant data and not modified during the
|
|
162 |
* calculation. The same data structure can be reused for multiple transforms
|
|
163 |
* including mixing forward and inverse transforms.
|
|
164 |
* \par
|
|
165 |
* Earlier releases of the library provided separate radix-2 and radix-4
|
|
166 |
* algorithms that operated on floating-point data. These functions are still
|
|
167 |
* provided but are deprecated. The older functions are slower and less general
|
|
168 |
* than the new functions.
|
|
169 |
* \par
|
|
170 |
* An example of initialization of the constants for the arm_cfft_q31 function follows:
|
|
171 |
* \code
|
|
172 |
* const static arm_cfft_instance_q31 *S;
|
|
173 |
* ...
|
|
174 |
* switch (length) {
|
|
175 |
* case 16:
|
|
176 |
* S = &arm_cfft_sR_q31_len16;
|
|
177 |
* break;
|
|
178 |
* case 32:
|
|
179 |
* S = &arm_cfft_sR_q31_len32;
|
|
180 |
* break;
|
|
181 |
* case 64:
|
|
182 |
* S = &arm_cfft_sR_q31_len64;
|
|
183 |
* break;
|
|
184 |
* case 128:
|
|
185 |
* S = &arm_cfft_sR_q31_len128;
|
|
186 |
* break;
|
|
187 |
* case 256:
|
|
188 |
* S = &arm_cfft_sR_q31_len256;
|
|
189 |
* break;
|
|
190 |
* case 512:
|
|
191 |
* S = &arm_cfft_sR_q31_len512;
|
|
192 |
* break;
|
|
193 |
* case 1024:
|
|
194 |
* S = &arm_cfft_sR_q31_len1024;
|
|
195 |
* break;
|
|
196 |
* case 2048:
|
|
197 |
* S = &arm_cfft_sR_q31_len2048;
|
|
198 |
* break;
|
|
199 |
* case 4096:
|
|
200 |
* S = &arm_cfft_sR_q31_len4096;
|
|
201 |
* break;
|
|
202 |
* }
|
|
203 |
* \endcode
|
|
204 |
*
|
|
205 |
*/
|
|
206 |
|
|
207 |
void arm_cfft_radix8by2_f32( arm_cfft_instance_f32 * S, float32_t * p1)
|
|
208 |
{
|
|
209 |
uint32_t L = S->fftLen;
|
|
210 |
float32_t * pCol1, * pCol2, * pMid1, * pMid2;
|
|
211 |
float32_t * p2 = p1 + L;
|
|
212 |
const float32_t * tw = (float32_t *) S->pTwiddle;
|
|
213 |
float32_t t1[4], t2[4], t3[4], t4[4], twR, twI;
|
|
214 |
float32_t m0, m1, m2, m3;
|
|
215 |
uint32_t l;
|
|
216 |
|
|
217 |
pCol1 = p1;
|
|
218 |
pCol2 = p2;
|
|
219 |
|
|
220 |
// Define new length
|
|
221 |
L >>= 1;
|
|
222 |
// Initialize mid pointers
|
|
223 |
pMid1 = p1 + L;
|
|
224 |
pMid2 = p2 + L;
|
|
225 |
|
|
226 |
// do two dot Fourier transform
|
|
227 |
for ( l = L >> 2; l > 0; l-- )
|
|
228 |
{
|
|
229 |
t1[0] = p1[0];
|
|
230 |
t1[1] = p1[1];
|
|
231 |
t1[2] = p1[2];
|
|
232 |
t1[3] = p1[3];
|
|
233 |
|
|
234 |
t2[0] = p2[0];
|
|
235 |
t2[1] = p2[1];
|
|
236 |
t2[2] = p2[2];
|
|
237 |
t2[3] = p2[3];
|
|
238 |
|
|
239 |
t3[0] = pMid1[0];
|
|
240 |
t3[1] = pMid1[1];
|
|
241 |
t3[2] = pMid1[2];
|
|
242 |
t3[3] = pMid1[3];
|
|
243 |
|
|
244 |
t4[0] = pMid2[0];
|
|
245 |
t4[1] = pMid2[1];
|
|
246 |
t4[2] = pMid2[2];
|
|
247 |
t4[3] = pMid2[3];
|
|
248 |
|
|
249 |
*p1++ = t1[0] + t2[0];
|
|
250 |
*p1++ = t1[1] + t2[1];
|
|
251 |
*p1++ = t1[2] + t2[2];
|
|
252 |
*p1++ = t1[3] + t2[3]; // col 1
|
|
253 |
|
|
254 |
t2[0] = t1[0] - t2[0];
|
|
255 |
t2[1] = t1[1] - t2[1];
|
|
256 |
t2[2] = t1[2] - t2[2];
|
|
257 |
t2[3] = t1[3] - t2[3]; // for col 2
|
|
258 |
|
|
259 |
*pMid1++ = t3[0] + t4[0];
|
|
260 |
*pMid1++ = t3[1] + t4[1];
|
|
261 |
*pMid1++ = t3[2] + t4[2];
|
|
262 |
*pMid1++ = t3[3] + t4[3]; // col 1
|
|
263 |
|
|
264 |
t4[0] = t4[0] - t3[0];
|
|
265 |
t4[1] = t4[1] - t3[1];
|
|
266 |
t4[2] = t4[2] - t3[2];
|
|
267 |
t4[3] = t4[3] - t3[3]; // for col 2
|
|
268 |
|
|
269 |
twR = *tw++;
|
|
270 |
twI = *tw++;
|
|
271 |
|
|
272 |
// multiply by twiddle factors
|
|
273 |
m0 = t2[0] * twR;
|
|
274 |
m1 = t2[1] * twI;
|
|
275 |
m2 = t2[1] * twR;
|
|
276 |
m3 = t2[0] * twI;
|
|
277 |
|
|
278 |
// R = R * Tr - I * Ti
|
|
279 |
*p2++ = m0 + m1;
|
|
280 |
// I = I * Tr + R * Ti
|
|
281 |
*p2++ = m2 - m3;
|
|
282 |
|
|
283 |
// use vertical symmetry
|
|
284 |
// 0.9988 - 0.0491i <==> -0.0491 - 0.9988i
|
|
285 |
m0 = t4[0] * twI;
|
|
286 |
m1 = t4[1] * twR;
|
|
287 |
m2 = t4[1] * twI;
|
|
288 |
m3 = t4[0] * twR;
|
|
289 |
|
|
290 |
*pMid2++ = m0 - m1;
|
|
291 |
*pMid2++ = m2 + m3;
|
|
292 |
|
|
293 |
twR = *tw++;
|
|
294 |
twI = *tw++;
|
|
295 |
|
|
296 |
m0 = t2[2] * twR;
|
|
297 |
m1 = t2[3] * twI;
|
|
298 |
m2 = t2[3] * twR;
|
|
299 |
m3 = t2[2] * twI;
|
|
300 |
|
|
301 |
*p2++ = m0 + m1;
|
|
302 |
*p2++ = m2 - m3;
|
|
303 |
|
|
304 |
m0 = t4[2] * twI;
|
|
305 |
m1 = t4[3] * twR;
|
|
306 |
m2 = t4[3] * twI;
|
|
307 |
m3 = t4[2] * twR;
|
|
308 |
|
|
309 |
*pMid2++ = m0 - m1;
|
|
310 |
*pMid2++ = m2 + m3;
|
|
311 |
}
|
|
312 |
|
|
313 |
// first col
|
|
314 |
arm_radix8_butterfly_f32( pCol1, L, (float32_t *) S->pTwiddle, 2u);
|
|
315 |
// second col
|
|
316 |
arm_radix8_butterfly_f32( pCol2, L, (float32_t *) S->pTwiddle, 2u);
|
|
317 |
}
|
|
318 |
|
|
319 |
void arm_cfft_radix8by4_f32( arm_cfft_instance_f32 * S, float32_t * p1)
|
|
320 |
{
|
|
321 |
uint32_t L = S->fftLen >> 1;
|
|
322 |
float32_t * pCol1, *pCol2, *pCol3, *pCol4, *pEnd1, *pEnd2, *pEnd3, *pEnd4;
|
|
323 |
const float32_t *tw2, *tw3, *tw4;
|
|
324 |
float32_t * p2 = p1 + L;
|
|
325 |
float32_t * p3 = p2 + L;
|
|
326 |
float32_t * p4 = p3 + L;
|
|
327 |
float32_t t2[4], t3[4], t4[4], twR, twI;
|
|
328 |
float32_t p1ap3_0, p1sp3_0, p1ap3_1, p1sp3_1;
|
|
329 |
float32_t m0, m1, m2, m3;
|
|
330 |
uint32_t l, twMod2, twMod3, twMod4;
|
|
331 |
|
|
332 |
pCol1 = p1; // points to real values by default
|
|
333 |
pCol2 = p2;
|
|
334 |
pCol3 = p3;
|
|
335 |
pCol4 = p4;
|
|
336 |
pEnd1 = p2 - 1; // points to imaginary values by default
|
|
337 |
pEnd2 = p3 - 1;
|
|
338 |
pEnd3 = p4 - 1;
|
|
339 |
pEnd4 = pEnd3 + L;
|
|
340 |
|
|
341 |
tw2 = tw3 = tw4 = (float32_t *) S->pTwiddle;
|
|
342 |
|
|
343 |
L >>= 1;
|
|
344 |
|
|
345 |
// do four dot Fourier transform
|
|
346 |
|
|
347 |
twMod2 = 2;
|
|
348 |
twMod3 = 4;
|
|
349 |
twMod4 = 6;
|
|
350 |
|
|
351 |
// TOP
|
|
352 |
p1ap3_0 = p1[0] + p3[0];
|
|
353 |
p1sp3_0 = p1[0] - p3[0];
|
|
354 |
p1ap3_1 = p1[1] + p3[1];
|
|
355 |
p1sp3_1 = p1[1] - p3[1];
|
|
356 |
|
|
357 |
// col 2
|
|
358 |
t2[0] = p1sp3_0 + p2[1] - p4[1];
|
|
359 |
t2[1] = p1sp3_1 - p2[0] + p4[0];
|
|
360 |
// col 3
|
|
361 |
t3[0] = p1ap3_0 - p2[0] - p4[0];
|
|
362 |
t3[1] = p1ap3_1 - p2[1] - p4[1];
|
|
363 |
// col 4
|
|
364 |
t4[0] = p1sp3_0 - p2[1] + p4[1];
|
|
365 |
t4[1] = p1sp3_1 + p2[0] - p4[0];
|
|
366 |
// col 1
|
|
367 |
*p1++ = p1ap3_0 + p2[0] + p4[0];
|
|
368 |
*p1++ = p1ap3_1 + p2[1] + p4[1];
|
|
369 |
|
|
370 |
// Twiddle factors are ones
|
|
371 |
*p2++ = t2[0];
|
|
372 |
*p2++ = t2[1];
|
|
373 |
*p3++ = t3[0];
|
|
374 |
*p3++ = t3[1];
|
|
375 |
*p4++ = t4[0];
|
|
376 |
*p4++ = t4[1];
|
|
377 |
|
|
378 |
tw2 += twMod2;
|
|
379 |
tw3 += twMod3;
|
|
380 |
tw4 += twMod4;
|
|
381 |
|
|
382 |
for (l = (L - 2) >> 1; l > 0; l-- )
|
|
383 |
{
|
|
384 |
// TOP
|
|
385 |
p1ap3_0 = p1[0] + p3[0];
|
|
386 |
p1sp3_0 = p1[0] - p3[0];
|
|
387 |
p1ap3_1 = p1[1] + p3[1];
|
|
388 |
p1sp3_1 = p1[1] - p3[1];
|
|
389 |
// col 2
|
|
390 |
t2[0] = p1sp3_0 + p2[1] - p4[1];
|
|
391 |
t2[1] = p1sp3_1 - p2[0] + p4[0];
|
|
392 |
// col 3
|
|
393 |
t3[0] = p1ap3_0 - p2[0] - p4[0];
|
|
394 |
t3[1] = p1ap3_1 - p2[1] - p4[1];
|
|
395 |
// col 4
|
|
396 |
t4[0] = p1sp3_0 - p2[1] + p4[1];
|
|
397 |
t4[1] = p1sp3_1 + p2[0] - p4[0];
|
|
398 |
// col 1 - top
|
|
399 |
*p1++ = p1ap3_0 + p2[0] + p4[0];
|
|
400 |
*p1++ = p1ap3_1 + p2[1] + p4[1];
|
|
401 |
|
|
402 |
// BOTTOM
|
|
403 |
p1ap3_1 = pEnd1[-1] + pEnd3[-1];
|
|
404 |
p1sp3_1 = pEnd1[-1] - pEnd3[-1];
|
|
405 |
p1ap3_0 = pEnd1[0] + pEnd3[0];
|
|
406 |
p1sp3_0 = pEnd1[0] - pEnd3[0];
|
|
407 |
// col 2
|
|
408 |
t2[2] = pEnd2[0] - pEnd4[0] + p1sp3_1;
|
|
409 |
t2[3] = pEnd1[0] - pEnd3[0] - pEnd2[-1] + pEnd4[-1];
|
|
410 |
// col 3
|
|
411 |
t3[2] = p1ap3_1 - pEnd2[-1] - pEnd4[-1];
|
|
412 |
t3[3] = p1ap3_0 - pEnd2[0] - pEnd4[0];
|
|
413 |
// col 4
|
|
414 |
t4[2] = pEnd2[0] - pEnd4[0] - p1sp3_1;
|
|
415 |
t4[3] = pEnd4[-1] - pEnd2[-1] - p1sp3_0;
|
|
416 |
// col 1 - Bottom
|
|
417 |
*pEnd1-- = p1ap3_0 + pEnd2[0] + pEnd4[0];
|
|
418 |
*pEnd1-- = p1ap3_1 + pEnd2[-1] + pEnd4[-1];
|
|
419 |
|
|
420 |
// COL 2
|
|
421 |
// read twiddle factors
|
|
422 |
twR = *tw2++;
|
|
423 |
twI = *tw2++;
|
|
424 |
// multiply by twiddle factors
|
|
425 |
// let Z1 = a + i(b), Z2 = c + i(d)
|
|
426 |
// => Z1 * Z2 = (a*c - b*d) + i(b*c + a*d)
|
|
427 |
|
|
428 |
// Top
|
|
429 |
m0 = t2[0] * twR;
|
|
430 |
m1 = t2[1] * twI;
|
|
431 |
m2 = t2[1] * twR;
|
|
432 |
m3 = t2[0] * twI;
|
|
433 |
|
|
434 |
*p2++ = m0 + m1;
|
|
435 |
*p2++ = m2 - m3;
|
|
436 |
// use vertical symmetry col 2
|
|
437 |
// 0.9997 - 0.0245i <==> 0.0245 - 0.9997i
|
|
438 |
// Bottom
|
|
439 |
m0 = t2[3] * twI;
|
|
440 |
m1 = t2[2] * twR;
|
|
441 |
m2 = t2[2] * twI;
|
|
442 |
m3 = t2[3] * twR;
|
|
443 |
|
|
444 |
*pEnd2-- = m0 - m1;
|
|
445 |
*pEnd2-- = m2 + m3;
|
|
446 |
|
|
447 |
// COL 3
|
|
448 |
twR = tw3[0];
|
|
449 |
twI = tw3[1];
|
|
450 |
tw3 += twMod3;
|
|
451 |
// Top
|
|
452 |
m0 = t3[0] * twR;
|
|
453 |
m1 = t3[1] * twI;
|
|
454 |
m2 = t3[1] * twR;
|
|
455 |
m3 = t3[0] * twI;
|
|
456 |
|
|
457 |
*p3++ = m0 + m1;
|
|
458 |
*p3++ = m2 - m3;
|
|
459 |
// use vertical symmetry col 3
|
|
460 |
// 0.9988 - 0.0491i <==> -0.9988 - 0.0491i
|
|
461 |
// Bottom
|
|
462 |
m0 = -t3[3] * twR;
|
|
463 |
m1 = t3[2] * twI;
|
|
464 |
m2 = t3[2] * twR;
|
|
465 |
m3 = t3[3] * twI;
|
|
466 |
|
|
467 |
*pEnd3-- = m0 - m1;
|
|
468 |
*pEnd3-- = m3 - m2;
|
|
469 |
|
|
470 |
// COL 4
|
|
471 |
twR = tw4[0];
|
|
472 |
twI = tw4[1];
|
|
473 |
tw4 += twMod4;
|
|
474 |
// Top
|
|
475 |
m0 = t4[0] * twR;
|
|
476 |
m1 = t4[1] * twI;
|
|
477 |
m2 = t4[1] * twR;
|
|
478 |
m3 = t4[0] * twI;
|
|
479 |
|
|
480 |
*p4++ = m0 + m1;
|
|
481 |
*p4++ = m2 - m3;
|
|
482 |
// use vertical symmetry col 4
|
|
483 |
// 0.9973 - 0.0736i <==> -0.0736 + 0.9973i
|
|
484 |
// Bottom
|
|
485 |
m0 = t4[3] * twI;
|
|
486 |
m1 = t4[2] * twR;
|
|
487 |
m2 = t4[2] * twI;
|
|
488 |
m3 = t4[3] * twR;
|
|
489 |
|
|
490 |
*pEnd4-- = m0 - m1;
|
|
491 |
*pEnd4-- = m2 + m3;
|
|
492 |
}
|
|
493 |
|
|
494 |
//MIDDLE
|
|
495 |
// Twiddle factors are
|
|
496 |
// 1.0000 0.7071-0.7071i -1.0000i -0.7071-0.7071i
|
|
497 |
p1ap3_0 = p1[0] + p3[0];
|
|
498 |
p1sp3_0 = p1[0] - p3[0];
|
|
499 |
p1ap3_1 = p1[1] + p3[1];
|
|
500 |
p1sp3_1 = p1[1] - p3[1];
|
|
501 |
|
|
502 |
// col 2
|
|
503 |
t2[0] = p1sp3_0 + p2[1] - p4[1];
|
|
504 |
t2[1] = p1sp3_1 - p2[0] + p4[0];
|
|
505 |
// col 3
|
|
506 |
t3[0] = p1ap3_0 - p2[0] - p4[0];
|
|
507 |
t3[1] = p1ap3_1 - p2[1] - p4[1];
|
|
508 |
// col 4
|
|
509 |
t4[0] = p1sp3_0 - p2[1] + p4[1];
|
|
510 |
t4[1] = p1sp3_1 + p2[0] - p4[0];
|
|
511 |
// col 1 - Top
|
|
512 |
*p1++ = p1ap3_0 + p2[0] + p4[0];
|
|
513 |
*p1++ = p1ap3_1 + p2[1] + p4[1];
|
|
514 |
|
|
515 |
// COL 2
|
|
516 |
twR = tw2[0];
|
|
517 |
twI = tw2[1];
|
|
518 |
|
|
519 |
m0 = t2[0] * twR;
|
|
520 |
m1 = t2[1] * twI;
|
|
521 |
m2 = t2[1] * twR;
|
|
522 |
m3 = t2[0] * twI;
|
|
523 |
|
|
524 |
*p2++ = m0 + m1;
|
|
525 |
*p2++ = m2 - m3;
|
|
526 |
// COL 3
|
|
527 |
twR = tw3[0];
|
|
528 |
twI = tw3[1];
|
|
529 |
|
|
530 |
m0 = t3[0] * twR;
|
|
531 |
m1 = t3[1] * twI;
|
|
532 |
m2 = t3[1] * twR;
|
|
533 |
m3 = t3[0] * twI;
|
|
534 |
|
|
535 |
*p3++ = m0 + m1;
|
|
536 |
*p3++ = m2 - m3;
|
|
537 |
// COL 4
|
|
538 |
twR = tw4[0];
|
|
539 |
twI = tw4[1];
|
|
540 |
|
|
541 |
m0 = t4[0] * twR;
|
|
542 |
m1 = t4[1] * twI;
|
|
543 |
m2 = t4[1] * twR;
|
|
544 |
m3 = t4[0] * twI;
|
|
545 |
|
|
546 |
*p4++ = m0 + m1;
|
|
547 |
*p4++ = m2 - m3;
|
|
548 |
|
|
549 |
// first col
|
|
550 |
arm_radix8_butterfly_f32( pCol1, L, (float32_t *) S->pTwiddle, 4u);
|
|
551 |
// second col
|
|
552 |
arm_radix8_butterfly_f32( pCol2, L, (float32_t *) S->pTwiddle, 4u);
|
|
553 |
// third col
|
|
554 |
arm_radix8_butterfly_f32( pCol3, L, (float32_t *) S->pTwiddle, 4u);
|
|
555 |
// fourth col
|
|
556 |
arm_radix8_butterfly_f32( pCol4, L, (float32_t *) S->pTwiddle, 4u);
|
|
557 |
}
|
|
558 |
|
|
559 |
/**
|
|
560 |
* @addtogroup ComplexFFT
|
|
561 |
* @{
|
|
562 |
*/
|
|
563 |
|
|
564 |
/**
|
|
565 |
* @details
|
|
566 |
* @brief Processing function for the floating-point complex FFT.
|
|
567 |
* @param[in] *S points to an instance of the floating-point CFFT structure.
|
|
568 |
* @param[in, out] *p1 points to the complex data buffer of size <code>2*fftLen</code>. Processing occurs in-place.
|
|
569 |
* @param[in] ifftFlag flag that selects forward (ifftFlag=0) or inverse (ifftFlag=1) transform.
|
|
570 |
* @param[in] bitReverseFlag flag that enables (bitReverseFlag=1) or disables (bitReverseFlag=0) bit reversal of output.
|
|
571 |
* @return none.
|
|
572 |
*/
|
|
573 |
|
|
574 |
void arm_cfft_f32(
|
|
575 |
const arm_cfft_instance_f32 * S,
|
|
576 |
float32_t * p1,
|
|
577 |
uint8_t ifftFlag,
|
|
578 |
uint8_t bitReverseFlag)
|
|
579 |
{
|
|
580 |
uint32_t L = S->fftLen, l;
|
|
581 |
float32_t invL, * pSrc;
|
|
582 |
|
|
583 |
if(ifftFlag == 1u)
|
|
584 |
{
|
|
585 |
/* Conjugate input data */
|
|
586 |
pSrc = p1 + 1;
|
|
587 |
for(l=0; l<L; l++)
|
|
588 |
{
|
|
589 |
*pSrc = -*pSrc;
|
|
590 |
pSrc += 2;
|
|
591 |
}
|
|
592 |
}
|
|
593 |
|
|
594 |
switch (L)
|
|
595 |
{
|
|
596 |
case 16:
|
|
597 |
case 128:
|
|
598 |
case 1024:
|
|
599 |
arm_cfft_radix8by2_f32 ( (arm_cfft_instance_f32 *) S, p1);
|
|
600 |
break;
|
|
601 |
case 32:
|
|
602 |
case 256:
|
|
603 |
case 2048:
|
|
604 |
arm_cfft_radix8by4_f32 ( (arm_cfft_instance_f32 *) S, p1);
|
|
605 |
break;
|
|
606 |
case 64:
|
|
607 |
case 512:
|
|
608 |
case 4096:
|
|
609 |
arm_radix8_butterfly_f32( p1, L, (float32_t *) S->pTwiddle, 1);
|
|
610 |
break;
|
|
611 |
}
|
|
612 |
|
|
613 |
if( bitReverseFlag )
|
|
614 |
arm_bitreversal_32((uint32_t*)p1,S->bitRevLength,S->pBitRevTable);
|
|
615 |
|
|
616 |
if(ifftFlag == 1u)
|
|
617 |
{
|
|
618 |
invL = 1.0f/(float32_t)L;
|
|
619 |
/* Conjugate and scale output data */
|
|
620 |
pSrc = p1;
|
|
621 |
for(l=0; l<L; l++)
|
|
622 |
{
|
|
623 |
*pSrc++ *= invL ;
|
|
624 |
*pSrc = -(*pSrc) * invL;
|
|
625 |
pSrc++;
|
|
626 |
}
|
|
627 |
}
|
|
628 |
}
|
|
629 |
|
|
630 |
/**
|
|
631 |
* @} end of ComplexFFT group
|
|
632 |
*/
|