• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  Copyright (c) 2018 The WebRTC project authors. All Rights Reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 #include "modules/audio_processing/agc2/compute_interpolated_gain_curve.h"
12 
13 #include <algorithm>
14 #include <cmath>
15 #include <queue>
16 #include <tuple>
17 #include <utility>
18 #include <vector>
19 
20 #include "modules/audio_processing/agc2/agc2_common.h"
21 #include "modules/audio_processing/agc2/agc2_testing_common.h"
22 #include "modules/audio_processing/agc2/limiter_db_gain_curve.h"
23 #include "rtc_base/checks.h"
24 
25 namespace webrtc {
26 namespace {
27 
ComputeLinearApproximationParams(const LimiterDbGainCurve * limiter,const double x)28 std::pair<double, double> ComputeLinearApproximationParams(
29     const LimiterDbGainCurve* limiter,
30     const double x) {
31   const double m = limiter->GetGainFirstDerivativeLinear(x);
32   const double q = limiter->GetGainLinear(x) - m * x;
33   return {m, q};
34 }
35 
ComputeAreaUnderPiecewiseLinearApproximation(const LimiterDbGainCurve * limiter,const double x0,const double x1)36 double ComputeAreaUnderPiecewiseLinearApproximation(
37     const LimiterDbGainCurve* limiter,
38     const double x0,
39     const double x1) {
40   RTC_CHECK_LT(x0, x1);
41 
42   // Linear approximation in x0 and x1.
43   double m0, q0, m1, q1;
44   std::tie(m0, q0) = ComputeLinearApproximationParams(limiter, x0);
45   std::tie(m1, q1) = ComputeLinearApproximationParams(limiter, x1);
46 
47   // Intersection point between two adjacent linear pieces.
48   RTC_CHECK_NE(m1, m0);
49   const double x_split = (q0 - q1) / (m1 - m0);
50   RTC_CHECK_LT(x0, x_split);
51   RTC_CHECK_LT(x_split, x1);
52 
53   auto area_under_linear_piece = [](double x_l, double x_r, double m,
54                                     double q) {
55     return x_r * (m * x_r / 2.0 + q) - x_l * (m * x_l / 2.0 + q);
56   };
57   return area_under_linear_piece(x0, x_split, m0, q0) +
58          area_under_linear_piece(x_split, x1, m1, q1);
59 }
60 
61 // Computes the approximation error in the limiter region for a given interval.
62 // The error is computed as the difference between the areas beneath the limiter
63 // curve to approximate and its linear under-approximation.
LimiterUnderApproximationNegativeError(const LimiterDbGainCurve * limiter,const double x0,const double x1)64 double LimiterUnderApproximationNegativeError(const LimiterDbGainCurve* limiter,
65                                               const double x0,
66                                               const double x1) {
67   const double area_limiter = limiter->GetGainIntegralLinear(x0, x1);
68   const double area_interpolated_curve =
69       ComputeAreaUnderPiecewiseLinearApproximation(limiter, x0, x1);
70   RTC_CHECK_GE(area_limiter, area_interpolated_curve);
71   return area_limiter - area_interpolated_curve;
72 }
73 
74 // Automatically finds where to sample the beyond-knee region of a limiter using
75 // a greedy optimization algorithm that iteratively decreases the approximation
76 // error.
77 // The solution is sub-optimal because the algorithm is greedy and the points
78 // are assigned by halving intervals (starting with the whole beyond-knee region
79 // as a single interval). However, even if sub-optimal, this algorithm works
80 // well in practice and it is efficiently implemented using priority queues.
SampleLimiterRegion(const LimiterDbGainCurve * limiter)81 std::vector<double> SampleLimiterRegion(const LimiterDbGainCurve* limiter) {
82   static_assert(kInterpolatedGainCurveBeyondKneePoints > 2, "");
83 
84   struct Interval {
85     Interval() = default;  // Ctor required by std::priority_queue.
86     Interval(double l, double r, double e) : x0(l), x1(r), error(e) {
87       RTC_CHECK(x0 < x1);
88     }
89     bool operator<(const Interval& other) const { return error < other.error; }
90 
91     double x0;
92     double x1;
93     double error;
94   };
95 
96   std::priority_queue<Interval, std::vector<Interval>> q;
97   q.emplace(limiter->limiter_start_linear(), limiter->max_input_level_linear(),
98             LimiterUnderApproximationNegativeError(
99                 limiter, limiter->limiter_start_linear(),
100                 limiter->max_input_level_linear()));
101 
102   // Iteratively find points by halving the interval with greatest error.
103   while (q.size() < kInterpolatedGainCurveBeyondKneePoints) {
104     // Get the interval with highest error.
105     const auto interval = q.top();
106     q.pop();
107 
108     // Split |interval| and enqueue.
109     double x_split = (interval.x0 + interval.x1) / 2.0;
110     q.emplace(interval.x0, x_split,
111               LimiterUnderApproximationNegativeError(limiter, interval.x0,
112                                                      x_split));  // Left.
113     q.emplace(x_split, interval.x1,
114               LimiterUnderApproximationNegativeError(limiter, x_split,
115                                                      interval.x1));  // Right.
116   }
117 
118   // Copy x1 values and sort them.
119   RTC_CHECK_EQ(q.size(), kInterpolatedGainCurveBeyondKneePoints);
120   std::vector<double> samples(kInterpolatedGainCurveBeyondKneePoints);
121   for (size_t i = 0; i < kInterpolatedGainCurveBeyondKneePoints; ++i) {
122     const auto interval = q.top();
123     q.pop();
124     samples[i] = interval.x1;
125   }
126   RTC_CHECK(q.empty());
127   std::sort(samples.begin(), samples.end());
128 
129   return samples;
130 }
131 
132 // Compute the parameters to over-approximate the knee region via linear
133 // interpolation. Over-approximating is saturation-safe since the knee region is
134 // convex.
PrecomputeKneeApproxParams(const LimiterDbGainCurve * limiter,test::InterpolatedParameters * parameters)135 void PrecomputeKneeApproxParams(const LimiterDbGainCurve* limiter,
136                                 test::InterpolatedParameters* parameters) {
137   static_assert(kInterpolatedGainCurveKneePoints > 2, "");
138   // Get |kInterpolatedGainCurveKneePoints| - 1 equally spaced points.
139   const std::vector<double> points = test::LinSpace(
140       limiter->knee_start_linear(), limiter->limiter_start_linear(),
141       kInterpolatedGainCurveKneePoints - 1);
142 
143   // Set the first two points. The second is computed to help with the beginning
144   // of the knee region, which has high curvature.
145   parameters->computed_approximation_params_x[0] = points[0];
146   parameters->computed_approximation_params_x[1] =
147       (points[0] + points[1]) / 2.0;
148   // Copy the remaining points.
149   std::copy(std::begin(points) + 1, std::end(points),
150             std::begin(parameters->computed_approximation_params_x) + 2);
151 
152   // Compute (m, q) pairs for each linear piece y = mx + q.
153   for (size_t i = 0; i < kInterpolatedGainCurveKneePoints - 1; ++i) {
154     const double x0 = parameters->computed_approximation_params_x[i];
155     const double x1 = parameters->computed_approximation_params_x[i + 1];
156     const double y0 = limiter->GetGainLinear(x0);
157     const double y1 = limiter->GetGainLinear(x1);
158     RTC_CHECK_NE(x1, x0);
159     parameters->computed_approximation_params_m[i] = (y1 - y0) / (x1 - x0);
160     parameters->computed_approximation_params_q[i] =
161         y0 - parameters->computed_approximation_params_m[i] * x0;
162   }
163 }
164 
165 // Compute the parameters to under-approximate the beyond-knee region via linear
166 // interpolation and greedy sampling. Under-approximating is saturation-safe
167 // since the beyond-knee region is concave.
PrecomputeBeyondKneeApproxParams(const LimiterDbGainCurve * limiter,test::InterpolatedParameters * parameters)168 void PrecomputeBeyondKneeApproxParams(
169     const LimiterDbGainCurve* limiter,
170     test::InterpolatedParameters* parameters) {
171   // Find points on which the linear pieces are tangent to the gain curve.
172   const auto samples = SampleLimiterRegion(limiter);
173 
174   // Parametrize each linear piece.
175   double m, q;
176   std::tie(m, q) = ComputeLinearApproximationParams(
177       limiter,
178       parameters
179           ->computed_approximation_params_x[kInterpolatedGainCurveKneePoints -
180                                             1]);
181   parameters
182       ->computed_approximation_params_m[kInterpolatedGainCurveKneePoints - 1] =
183       m;
184   parameters
185       ->computed_approximation_params_q[kInterpolatedGainCurveKneePoints - 1] =
186       q;
187   for (size_t i = 0; i < samples.size(); ++i) {
188     std::tie(m, q) = ComputeLinearApproximationParams(limiter, samples[i]);
189     parameters
190         ->computed_approximation_params_m[i +
191                                           kInterpolatedGainCurveKneePoints] = m;
192     parameters
193         ->computed_approximation_params_q[i +
194                                           kInterpolatedGainCurveKneePoints] = q;
195   }
196 
197   // Find the point of intersection between adjacent linear pieces. They will be
198   // used as boundaries between adjacent linear pieces.
199   for (size_t i = kInterpolatedGainCurveKneePoints;
200        i < kInterpolatedGainCurveKneePoints +
201                kInterpolatedGainCurveBeyondKneePoints;
202        ++i) {
203     RTC_CHECK_NE(parameters->computed_approximation_params_m[i],
204                  parameters->computed_approximation_params_m[i - 1]);
205     parameters->computed_approximation_params_x[i] =
206         (  // Formula: (q0 - q1) / (m1 - m0).
207             parameters->computed_approximation_params_q[i - 1] -
208             parameters->computed_approximation_params_q[i]) /
209         (parameters->computed_approximation_params_m[i] -
210          parameters->computed_approximation_params_m[i - 1]);
211   }
212 }
213 
214 }  // namespace
215 
216 namespace test {
217 
ComputeInterpolatedGainCurveApproximationParams()218 InterpolatedParameters ComputeInterpolatedGainCurveApproximationParams() {
219   InterpolatedParameters parameters;
220   LimiterDbGainCurve limiter;
221   parameters.computed_approximation_params_x.fill(0.0f);
222   parameters.computed_approximation_params_m.fill(0.0f);
223   parameters.computed_approximation_params_q.fill(0.0f);
224   PrecomputeKneeApproxParams(&limiter, &parameters);
225   PrecomputeBeyondKneeApproxParams(&limiter, &parameters);
226   return parameters;
227 }
228 }  // namespace test
229 }  // namespace webrtc
230