Mercurial > hg > silvet
comparison constant-q-cpp/src/ConstantQ.cpp @ 366:5d0a2ebb4d17
Bring dependent libraries in to repo
author | Chris Cannam |
---|---|
date | Fri, 24 Jun 2016 14:47:45 +0100 |
parents | |
children |
comparison
equal
deleted
inserted
replaced
365:112766f4c34b | 366:5d0a2ebb4d17 |
---|---|
1 /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */ | |
2 /* | |
3 Constant-Q library | |
4 Copyright (c) 2013-2014 Queen Mary, University of London | |
5 | |
6 Permission is hereby granted, free of charge, to any person | |
7 obtaining a copy of this software and associated documentation | |
8 files (the "Software"), to deal in the Software without | |
9 restriction, including without limitation the rights to use, copy, | |
10 modify, merge, publish, distribute, sublicense, and/or sell copies | |
11 of the Software, and to permit persons to whom the Software is | |
12 furnished to do so, subject to the following conditions: | |
13 | |
14 The above copyright notice and this permission notice shall be | |
15 included in all copies or substantial portions of the Software. | |
16 | |
17 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
18 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
19 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
20 NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY | |
21 CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF | |
22 CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION | |
23 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | |
24 | |
25 Except as contained in this notice, the names of the Centre for | |
26 Digital Music; Queen Mary, University of London; and Chris Cannam | |
27 shall not be used in advertising or otherwise to promote the sale, | |
28 use or other dealings in this Software without prior written | |
29 authorization. | |
30 */ | |
31 | |
32 #include "ConstantQ.h" | |
33 | |
34 #include "CQKernel.h" | |
35 | |
36 #include "dsp/Resampler.h" | |
37 #include "dsp/MathUtilities.h" | |
38 #include "dsp/FFT.h" | |
39 | |
40 #include <algorithm> | |
41 #include <iostream> | |
42 #include <stdexcept> | |
43 | |
44 #include <cmath> | |
45 | |
46 using std::vector; | |
47 using std::cerr; | |
48 using std::endl; | |
49 | |
50 //#define DEBUG_CQ 1 | |
51 | |
52 ConstantQ::ConstantQ(CQParameters params) : | |
53 m_inparams(params), | |
54 m_sampleRate(params.sampleRate), | |
55 m_maxFrequency(params.maxFrequency), | |
56 m_minFrequency(params.minFrequency), | |
57 m_binsPerOctave(params.binsPerOctave), | |
58 m_kernel(0), | |
59 m_fft(0) | |
60 { | |
61 if (m_minFrequency <= 0.0 || m_maxFrequency <= 0.0) { | |
62 throw std::invalid_argument("Frequency extents must be positive"); | |
63 } | |
64 | |
65 initialise(); | |
66 } | |
67 | |
68 ConstantQ::~ConstantQ() | |
69 { | |
70 delete m_fft; | |
71 for (int i = 0; i < (int)m_decimators.size(); ++i) { | |
72 delete m_decimators[i]; | |
73 } | |
74 delete m_kernel; | |
75 } | |
76 | |
77 double | |
78 ConstantQ::getMinFrequency() const | |
79 { | |
80 return m_p.minFrequency / pow(2.0, m_octaves - 1); | |
81 } | |
82 | |
83 double | |
84 ConstantQ::getBinFrequency(double bin) const | |
85 { | |
86 // our bins are returned in high->low order | |
87 bin = (getBinsPerOctave() * getOctaves()) - bin - 1; | |
88 return getMinFrequency() * pow(2, (bin / getBinsPerOctave())); | |
89 } | |
90 | |
91 void | |
92 ConstantQ::initialise() | |
93 { | |
94 m_octaves = int(ceil(log(m_maxFrequency / m_minFrequency) / log(2))); | |
95 | |
96 if (m_octaves < 1) { | |
97 m_kernel = 0; // incidentally causing isValid() to return false | |
98 return; | |
99 } | |
100 | |
101 m_kernel = new CQKernel(m_inparams); | |
102 m_p = m_kernel->getProperties(); | |
103 | |
104 if (!m_kernel->isValid()) { | |
105 return; | |
106 } | |
107 | |
108 // Use exact powers of two for resampling rates. They don't have | |
109 // to be related to our actual samplerate: the resampler only | |
110 // cares about the ratio, but it only accepts integer source and | |
111 // target rates, and if we start from the actual samplerate we | |
112 // risk getting non-integer rates for lower octaves | |
113 | |
114 int sourceRate = pow(2, m_octaves); | |
115 vector<int> latencies; | |
116 | |
117 // top octave, no resampling | |
118 latencies.push_back(0); | |
119 m_decimators.push_back(0); | |
120 | |
121 for (int i = 1; i < m_octaves; ++i) { | |
122 | |
123 int factor = pow(2, i); | |
124 | |
125 Resampler *r; | |
126 | |
127 if (m_inparams.decimator == CQParameters::BetterDecimator) { | |
128 r = new Resampler | |
129 (sourceRate, sourceRate / factor, 50, 0.05); | |
130 } else { | |
131 r = new Resampler | |
132 (sourceRate, sourceRate / factor, 25, 0.3); | |
133 } | |
134 | |
135 #ifdef DEBUG_CQ | |
136 cerr << "forward: octave " << i << ": resample from " << sourceRate << " to " << sourceRate / factor << endl; | |
137 #endif | |
138 | |
139 // We need to adapt the latencies so as to get the first input | |
140 // sample to be aligned, in time, at the decimator output | |
141 // across all octaves. | |
142 // | |
143 // Our decimator uses a linear phase filter, but being causal | |
144 // it is not zero phase: it has a latency that depends on the | |
145 // decimation factor. Those latencies have been calculated | |
146 // per-octave and are available to us in the latencies | |
147 // array. Left to its own devices, the first input sample will | |
148 // appear at output sample 0 in the highest octave (where no | |
149 // decimation is needed), sample number latencies[1] in the | |
150 // next octave down, latencies[2] in the next one, etc. We get | |
151 // to apply some artificial per-octave latency after the | |
152 // decimator in the processing chain, in order to compensate | |
153 // for the differing latencies associated with different | |
154 // decimation factors. How much should we insert? | |
155 // | |
156 // The outputs of the decimators are at different rates (in | |
157 // terms of the relation between clock time and samples) and | |
158 // we want them aligned in terms of time. So, for example, a | |
159 // latency of 10 samples with a decimation factor of 2 is | |
160 // equivalent to a latency of 20 with no decimation -- they | |
161 // both result in the first output sample happening at the | |
162 // same equivalent time in milliseconds. | |
163 // | |
164 // So here we record the latency added by the decimator, in | |
165 // terms of the sample rate of the undecimated signal. Then we | |
166 // use that to compensate in a moment, when we've discovered | |
167 // what the longest latency across all octaves is. | |
168 | |
169 latencies.push_back(r->getLatency() * factor); | |
170 m_decimators.push_back(r); | |
171 } | |
172 | |
173 m_bigBlockSize = m_p.fftSize * pow(2, m_octaves - 1); | |
174 | |
175 // Now add in the extra padding and compensate for hops that must | |
176 // be dropped in order to align the atom centres across | |
177 // octaves. Again this is a bit trickier because we are doing it | |
178 // at input rather than output and so must work in per-octave | |
179 // sample rates rather than output blocks | |
180 | |
181 int emptyHops = m_p.firstCentre / m_p.atomSpacing; | |
182 | |
183 vector<int> drops; | |
184 for (int i = 0; i < m_octaves; ++i) { | |
185 int factor = pow(2, i); | |
186 int dropHops = emptyHops * pow(2, m_octaves - i - 1) - emptyHops; | |
187 int drop = ((dropHops * m_p.fftHop) * factor) / m_p.atomsPerFrame; | |
188 drops.push_back(drop); | |
189 } | |
190 | |
191 int maxLatPlusDrop = 0; | |
192 for (int i = 0; i < m_octaves; ++i) { | |
193 int latPlusDrop = latencies[i] + drops[i]; | |
194 if (latPlusDrop > maxLatPlusDrop) maxLatPlusDrop = latPlusDrop; | |
195 } | |
196 | |
197 int totalLatency = maxLatPlusDrop; | |
198 | |
199 int lat0 = totalLatency - latencies[0] - drops[0]; | |
200 totalLatency = ceil(double(lat0 / m_p.fftHop) * m_p.fftHop) | |
201 + latencies[0] + drops[0]; | |
202 | |
203 // We want (totalLatency - latencies[i]) to be a multiple of 2^i | |
204 // for each octave i, so that we do not end up with fractional | |
205 // octave latencies below. In theory this is hard, in practice if | |
206 // we ensure it for the last octave we should be OK. | |
207 double finalOctLat = latencies[m_octaves-1]; | |
208 double finalOctFact = pow(2, m_octaves-1); | |
209 totalLatency = | |
210 int(finalOctLat + | |
211 finalOctFact * | |
212 ceil((totalLatency - finalOctLat) / finalOctFact) + .5); | |
213 | |
214 #ifdef DEBUG_CQ | |
215 cerr << "total latency = " << totalLatency << endl; | |
216 #endif | |
217 | |
218 // Padding as in the reference (will be introduced with the | |
219 // latency compensation in the loop below) | |
220 m_outputLatency = totalLatency + m_bigBlockSize | |
221 - m_p.firstCentre * pow(2, m_octaves-1); | |
222 | |
223 #ifdef DEBUG_CQ | |
224 cerr << "m_bigBlockSize = " << m_bigBlockSize << ", firstCentre = " | |
225 << m_p.firstCentre << ", m_octaves = " << m_octaves | |
226 << ", so m_outputLatency = " << m_outputLatency << endl; | |
227 #endif | |
228 | |
229 for (int i = 0; i < m_octaves; ++i) { | |
230 | |
231 double factor = pow(2, i); | |
232 | |
233 // Calculate the difference between the total latency applied | |
234 // across all octaves, and the existing latency due to the | |
235 // decimator for this octave, and then convert it back into | |
236 // the sample rate appropriate for the output latency of this | |
237 // decimator -- including one additional big block of padding | |
238 // (as in the reference). | |
239 | |
240 double octaveLatency = | |
241 double(totalLatency - latencies[i] - drops[i] | |
242 + m_bigBlockSize) / factor; | |
243 | |
244 #ifdef DEBUG_CQ | |
245 cerr << "octave " << i << ": resampler latency = " << latencies[i] | |
246 << ", drop " << drops[i] << " (/factor = " << drops[i]/factor | |
247 << "), octaveLatency = " << octaveLatency << " -> " | |
248 << int(round(octaveLatency)) << " (diff * factor = " | |
249 << (octaveLatency - round(octaveLatency)) << " * " | |
250 << factor << " = " | |
251 << (octaveLatency - round(octaveLatency)) * factor << ")" << endl; | |
252 | |
253 cerr << "double(" << totalLatency << " - " | |
254 << latencies[i] << " - " << drops[i] << " + " | |
255 << m_bigBlockSize << ") / " << factor << " = " | |
256 << octaveLatency << endl; | |
257 #endif | |
258 | |
259 m_buffers.push_back | |
260 (RealSequence(int(octaveLatency + 0.5), 0.0)); | |
261 } | |
262 | |
263 m_fft = new FFTReal(m_p.fftSize); | |
264 } | |
265 | |
266 ConstantQ::ComplexBlock | |
267 ConstantQ::process(const RealSequence &td) | |
268 { | |
269 m_buffers[0].insert(m_buffers[0].end(), td.begin(), td.end()); | |
270 | |
271 for (int i = 1; i < m_octaves; ++i) { | |
272 RealSequence dec = m_decimators[i]->process(td.data(), td.size()); | |
273 m_buffers[i].insert(m_buffers[i].end(), dec.begin(), dec.end()); | |
274 } | |
275 | |
276 ComplexBlock out; | |
277 | |
278 while (true) { | |
279 | |
280 // We could have quite different remaining sample counts in | |
281 // different octaves, because (apart from the predictable | |
282 // added counts for decimator output on each block) we also | |
283 // have variable additional latency per octave | |
284 bool enough = true; | |
285 for (int i = 0; i < m_octaves; ++i) { | |
286 int required = m_p.fftSize * pow(2, m_octaves - i - 1); | |
287 if ((int)m_buffers[i].size() < required) { | |
288 enough = false; | |
289 } | |
290 } | |
291 if (!enough) break; | |
292 | |
293 int base = out.size(); | |
294 int totalColumns = pow(2, m_octaves - 1) * m_p.atomsPerFrame; | |
295 for (int i = 0; i < totalColumns; ++i) { | |
296 out.push_back(ComplexColumn()); | |
297 } | |
298 | |
299 for (int octave = 0; octave < m_octaves; ++octave) { | |
300 | |
301 int blocksThisOctave = pow(2, (m_octaves - octave - 1)); | |
302 | |
303 for (int b = 0; b < blocksThisOctave; ++b) { | |
304 ComplexBlock block = processOctaveBlock(octave); | |
305 | |
306 for (int j = 0; j < m_p.atomsPerFrame; ++j) { | |
307 | |
308 int target = base + | |
309 (b * (totalColumns / blocksThisOctave) + | |
310 (j * ((totalColumns / blocksThisOctave) / | |
311 m_p.atomsPerFrame))); | |
312 | |
313 while (int(out[target].size()) < | |
314 m_p.binsPerOctave * (octave + 1)) { | |
315 out[target].push_back(Complex()); | |
316 } | |
317 | |
318 for (int i = 0; i < m_p.binsPerOctave; ++i) { | |
319 out[target][m_p.binsPerOctave * octave + i] = | |
320 block[j][m_p.binsPerOctave - i - 1]; | |
321 } | |
322 } | |
323 } | |
324 } | |
325 } | |
326 | |
327 return out; | |
328 } | |
329 | |
330 ConstantQ::ComplexBlock | |
331 ConstantQ::getRemainingOutput() | |
332 { | |
333 // Same as padding added at start, though rounded up | |
334 int pad = ceil(double(m_outputLatency) / m_bigBlockSize) * m_bigBlockSize; | |
335 RealSequence zeros(pad, 0.0); | |
336 return process(zeros); | |
337 } | |
338 | |
339 ConstantQ::ComplexBlock | |
340 ConstantQ::processOctaveBlock(int octave) | |
341 { | |
342 RealSequence ro(m_p.fftSize, 0.0); | |
343 RealSequence io(m_p.fftSize, 0.0); | |
344 | |
345 m_fft->forward(m_buffers[octave].data(), ro.data(), io.data()); | |
346 | |
347 m_buffers[octave] = RealSequence(m_buffers[octave].begin() + m_p.fftHop, | |
348 m_buffers[octave].end()); | |
349 | |
350 ComplexSequence cv(m_p.fftSize); | |
351 for (int i = 0; i < m_p.fftSize; ++i) { | |
352 cv[i] = Complex(ro[i], io[i]); | |
353 } | |
354 | |
355 ComplexSequence cqrowvec = m_kernel->processForward(cv); | |
356 | |
357 // Reform into a column matrix | |
358 ComplexBlock cqblock; | |
359 for (int j = 0; j < m_p.atomsPerFrame; ++j) { | |
360 cqblock.push_back(ComplexColumn()); | |
361 for (int i = 0; i < m_p.binsPerOctave; ++i) { | |
362 cqblock[j].push_back(cqrowvec[i * m_p.atomsPerFrame + j]); | |
363 } | |
364 } | |
365 | |
366 return cqblock; | |
367 } | |
368 | |
369 |