2 * Copyright (C) 2010 Google Inc. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of Apple Computer, Inc. ("Apple") nor the names of
14 * its contributors may be used to endorse or promote products derived
15 * from this software without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND ANY
18 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
19 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
20 * DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR ANY
21 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
22 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
23 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
24 * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 #include "platform/audio/FFTFrame.h"
39 #include "platform/Logging.h"
40 #include "wtf/Complex.h"
41 #include "wtf/MathExtras.h"
42 #include "wtf/OwnPtr.h"
46 void FFTFrame::doPaddedFFT(const float* data, size_t dataSize)
48 // Zero-pad the impulse response
49 AudioFloatArray paddedResponse(fftSize()); // zero-initialized
50 paddedResponse.copyToRange(data, 0, dataSize);
52 // Get the frequency-domain version of padded response
53 doFFT(paddedResponse.data());
56 PassOwnPtr<FFTFrame> FFTFrame::createInterpolatedFrame(const FFTFrame& frame1, const FFTFrame& frame2, double x)
58 OwnPtr<FFTFrame> newFrame = adoptPtr(new FFTFrame(frame1.fftSize()));
60 newFrame->interpolateFrequencyComponents(frame1, frame2, x);
62 // In the time-domain, the 2nd half of the response must be zero, to avoid circular convolution aliasing...
63 int fftSize = newFrame->fftSize();
64 AudioFloatArray buffer(fftSize);
65 newFrame->doInverseFFT(buffer.data());
66 buffer.zeroRange(fftSize / 2, fftSize);
68 // Put back into frequency domain.
69 newFrame->doFFT(buffer.data());
71 return newFrame.release();
75 // On Windows, the following pragmas are equivalent to compiling the code with /fp:fast. The
76 // following code does not need precise FP semantics, and speed is critical here. See
77 // crbug.com/316740 and crrev.com/116823002.
78 #pragma float_control(push)
79 #pragma float_control(except, off)
80 #pragma float_control(precise, off)
81 #pragma fp_contract(on)
82 #pragma fenv_access(off)
85 void FFTFrame::interpolateFrequencyComponents(const FFTFrame& frame1, const FFTFrame& frame2, double interp)
87 // FIXME : with some work, this method could be optimized
89 float* realP = realData();
90 float* imagP = imagData();
92 const float* realP1 = frame1.realData();
93 const float* imagP1 = frame1.imagData();
94 const float* realP2 = frame2.realData();
95 const float* imagP2 = frame2.imagData();
97 m_FFTSize = frame1.fftSize();
98 m_log2FFTSize = frame1.log2FFTSize();
100 double s1base = (1.0 - interp);
101 double s2base = interp;
103 double phaseAccum = 0.0;
104 double lastPhase1 = 0.0;
105 double lastPhase2 = 0.0;
107 realP[0] = static_cast<float>(s1base * realP1[0] + s2base * realP2[0]);
108 imagP[0] = static_cast<float>(s1base * imagP1[0] + s2base * imagP2[0]);
110 int n = m_FFTSize / 2;
112 for (int i = 1; i < n; ++i) {
113 Complex c1(realP1[i], imagP1[i]);
114 Complex c2(realP2[i], imagP2[i]);
116 double mag1 = abs(c1);
117 double mag2 = abs(c2);
119 // Interpolate magnitudes in decibels
120 double mag1db = 20.0 * log10(mag1);
121 double mag2db = 20.0 * log10(mag2);
126 double magdbdiff = mag1db - mag2db;
128 // Empirical tweak to retain higher-frequency zeroes
129 double threshold = (i > 16) ? 5.0 : 2.0;
131 if (magdbdiff < -threshold && mag1db < 0.0) {
134 } else if (magdbdiff > threshold && mag2db < 0.0) {
139 // Average magnitude by decibels instead of linearly
140 double magdb = s1 * mag1db + s2 * mag2db;
141 double mag = pow(10.0, 0.05 * magdb);
143 // Now, deal with phase
144 double phase1 = arg(c1);
145 double phase2 = arg(c2);
147 double deltaPhase1 = phase1 - lastPhase1;
148 double deltaPhase2 = phase2 - lastPhase2;
152 // Unwrap phase deltas
153 if (deltaPhase1 > piDouble)
154 deltaPhase1 -= 2.0 * piDouble;
155 if (deltaPhase1 < -piDouble)
156 deltaPhase1 += 2.0 * piDouble;
157 if (deltaPhase2 > piDouble)
158 deltaPhase2 -= 2.0 * piDouble;
159 if (deltaPhase2 < -piDouble)
160 deltaPhase2 += 2.0 * piDouble;
162 // Blend group-delays
163 double deltaPhaseBlend;
165 if (deltaPhase1 - deltaPhase2 > piDouble)
166 deltaPhaseBlend = s1 * deltaPhase1 + s2 * (2.0 * piDouble + deltaPhase2);
167 else if (deltaPhase2 - deltaPhase1 > piDouble)
168 deltaPhaseBlend = s1 * (2.0 * piDouble + deltaPhase1) + s2 * deltaPhase2;
170 deltaPhaseBlend = s1 * deltaPhase1 + s2 * deltaPhase2;
172 phaseAccum += deltaPhaseBlend;
175 if (phaseAccum > piDouble)
176 phaseAccum -= 2.0 * piDouble;
177 if (phaseAccum < -piDouble)
178 phaseAccum += 2.0 * piDouble;
180 Complex c = complexFromMagnitudePhase(mag, phaseAccum);
182 realP[i] = static_cast<float>(c.real());
183 imagP[i] = static_cast<float>(c.imag());
187 double FFTFrame::extractAverageGroupDelay()
189 float* realP = realData();
190 float* imagP = imagData();
193 double weightSum = 0.0;
194 double lastPhase = 0.0;
196 int halfSize = fftSize() / 2;
198 const double kSamplePhaseDelay = (2.0 * piDouble) / double(fftSize());
200 // Calculate weighted average group delay
201 for (int i = 0; i < halfSize; i++) {
202 Complex c(realP[i], imagP[i]);
204 double phase = arg(c);
206 double deltaPhase = phase - lastPhase;
210 if (deltaPhase < -piDouble)
211 deltaPhase += 2.0 * piDouble;
212 if (deltaPhase > piDouble)
213 deltaPhase -= 2.0 * piDouble;
215 aveSum += mag * deltaPhase;
219 // Note how we invert the phase delta wrt frequency since this is how group delay is defined
220 double ave = aveSum / weightSum;
221 double aveSampleDelay = -ave / kSamplePhaseDelay;
223 // Leave 20 sample headroom (for leading edge of impulse)
224 if (aveSampleDelay > 20.0)
225 aveSampleDelay -= 20.0;
227 // Remove average group delay (minus 20 samples for headroom)
228 addConstantGroupDelay(-aveSampleDelay);
233 return aveSampleDelay;
236 void FFTFrame::addConstantGroupDelay(double sampleFrameDelay)
238 int halfSize = fftSize() / 2;
240 float* realP = realData();
241 float* imagP = imagData();
243 const double kSamplePhaseDelay = (2.0 * piDouble) / double(fftSize());
245 double phaseAdj = -sampleFrameDelay * kSamplePhaseDelay;
247 // Add constant group delay
248 for (int i = 1; i < halfSize; i++) {
249 Complex c(realP[i], imagP[i]);
251 double phase = arg(c);
253 phase += i * phaseAdj;
255 Complex c2 = complexFromMagnitudePhase(mag, phase);
257 realP[i] = static_cast<float>(c2.real());
258 imagP[i] = static_cast<float>(c2.imag());
263 void FFTFrame::print()
265 FFTFrame& frame = *this;
266 float* realP = frame.realData();
267 float* imagP = frame.imagData();
268 WTF_LOG(WebAudio, "**** \n");
269 WTF_LOG(WebAudio, "DC = %f : nyquist = %f\n", realP[0], imagP[0]);
271 int n = m_FFTSize / 2;
273 for (int i = 1; i < n; i++) {
274 double mag = sqrt(realP[i] * realP[i] + imagP[i] * imagP[i]);
275 double phase = atan2(realP[i], imagP[i]);
277 WTF_LOG(WebAudio, "[%d] (%f %f)\n", i, mag, phase);
279 WTF_LOG(WebAudio, "****\n");
283 } // namespace WebCore
285 #endif // ENABLE(WEB_AUDIO)