1 /*
2 * Copyright (c) 2013 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/remote_bitrate_estimator/overuse_estimator.h"
12
13 #include <assert.h>
14 #include <math.h>
15 #include <string.h>
16
17 #include <algorithm>
18
19 #include "modules/remote_bitrate_estimator/include/bwe_defines.h"
20 #include "modules/remote_bitrate_estimator/test/bwe_test_logging.h"
21 #include "rtc_base/logging.h"
22
23 namespace webrtc {
24
25 enum { kMinFramePeriodHistoryLength = 60 };
26 enum { kDeltaCounterMax = 1000 };
27
OveruseEstimator(const OverUseDetectorOptions & options)28 OveruseEstimator::OveruseEstimator(const OverUseDetectorOptions& options)
29 : options_(options),
30 num_of_deltas_(0),
31 slope_(options_.initial_slope),
32 offset_(options_.initial_offset),
33 prev_offset_(options_.initial_offset),
34 E_(),
35 process_noise_(),
36 avg_noise_(options_.initial_avg_noise),
37 var_noise_(options_.initial_var_noise),
38 ts_delta_hist_() {
39 memcpy(E_, options_.initial_e, sizeof(E_));
40 memcpy(process_noise_, options_.initial_process_noise,
41 sizeof(process_noise_));
42 }
43
~OveruseEstimator()44 OveruseEstimator::~OveruseEstimator() {
45 ts_delta_hist_.clear();
46 }
47
Update(int64_t t_delta,double ts_delta,int size_delta,BandwidthUsage current_hypothesis,int64_t now_ms)48 void OveruseEstimator::Update(int64_t t_delta,
49 double ts_delta,
50 int size_delta,
51 BandwidthUsage current_hypothesis,
52 int64_t now_ms) {
53 const double min_frame_period = UpdateMinFramePeriod(ts_delta);
54 const double t_ts_delta = t_delta - ts_delta;
55 BWE_TEST_LOGGING_PLOT(1, "dm_ms", now_ms, t_ts_delta);
56 double fs_delta = size_delta;
57
58 ++num_of_deltas_;
59 if (num_of_deltas_ > kDeltaCounterMax) {
60 num_of_deltas_ = kDeltaCounterMax;
61 }
62
63 // Update the Kalman filter.
64 E_[0][0] += process_noise_[0];
65 E_[1][1] += process_noise_[1];
66
67 if ((current_hypothesis == BandwidthUsage::kBwOverusing &&
68 offset_ < prev_offset_) ||
69 (current_hypothesis == BandwidthUsage::kBwUnderusing &&
70 offset_ > prev_offset_)) {
71 E_[1][1] += 10 * process_noise_[1];
72 }
73
74 const double h[2] = {fs_delta, 1.0};
75 const double Eh[2] = {E_[0][0] * h[0] + E_[0][1] * h[1],
76 E_[1][0] * h[0] + E_[1][1] * h[1]};
77
78 BWE_TEST_LOGGING_PLOT(1, "d_ms", now_ms, slope_ * h[0] - offset_);
79
80 const double residual = t_ts_delta - slope_ * h[0] - offset_;
81
82 const bool in_stable_state =
83 (current_hypothesis == BandwidthUsage::kBwNormal);
84 const double max_residual = 3.0 * sqrt(var_noise_);
85 // We try to filter out very late frames. For instance periodic key
86 // frames doesn't fit the Gaussian model well.
87 if (fabs(residual) < max_residual) {
88 UpdateNoiseEstimate(residual, min_frame_period, in_stable_state);
89 } else {
90 UpdateNoiseEstimate(residual < 0 ? -max_residual : max_residual,
91 min_frame_period, in_stable_state);
92 }
93
94 const double denom = var_noise_ + h[0] * Eh[0] + h[1] * Eh[1];
95
96 const double K[2] = {Eh[0] / denom, Eh[1] / denom};
97
98 const double IKh[2][2] = {{1.0 - K[0] * h[0], -K[0] * h[1]},
99 {-K[1] * h[0], 1.0 - K[1] * h[1]}};
100 const double e00 = E_[0][0];
101 const double e01 = E_[0][1];
102
103 // Update state.
104 E_[0][0] = e00 * IKh[0][0] + E_[1][0] * IKh[0][1];
105 E_[0][1] = e01 * IKh[0][0] + E_[1][1] * IKh[0][1];
106 E_[1][0] = e00 * IKh[1][0] + E_[1][0] * IKh[1][1];
107 E_[1][1] = e01 * IKh[1][0] + E_[1][1] * IKh[1][1];
108
109 // The covariance matrix must be positive semi-definite.
110 bool positive_semi_definite =
111 E_[0][0] + E_[1][1] >= 0 &&
112 E_[0][0] * E_[1][1] - E_[0][1] * E_[1][0] >= 0 && E_[0][0] >= 0;
113 assert(positive_semi_definite);
114 if (!positive_semi_definite) {
115 RTC_LOG(LS_ERROR)
116 << "The over-use estimator's covariance matrix is no longer "
117 "semi-definite.";
118 }
119
120 slope_ = slope_ + K[0] * residual;
121 prev_offset_ = offset_;
122 offset_ = offset_ + K[1] * residual;
123
124 BWE_TEST_LOGGING_PLOT(1, "kc", now_ms, K[0]);
125 BWE_TEST_LOGGING_PLOT(1, "km", now_ms, K[1]);
126 BWE_TEST_LOGGING_PLOT(1, "slope_1/bps", now_ms, slope_);
127 BWE_TEST_LOGGING_PLOT(1, "var_noise", now_ms, var_noise_);
128 }
129
UpdateMinFramePeriod(double ts_delta)130 double OveruseEstimator::UpdateMinFramePeriod(double ts_delta) {
131 double min_frame_period = ts_delta;
132 if (ts_delta_hist_.size() >= kMinFramePeriodHistoryLength) {
133 ts_delta_hist_.pop_front();
134 }
135 for (const double old_ts_delta : ts_delta_hist_) {
136 min_frame_period = std::min(old_ts_delta, min_frame_period);
137 }
138 ts_delta_hist_.push_back(ts_delta);
139 return min_frame_period;
140 }
141
UpdateNoiseEstimate(double residual,double ts_delta,bool stable_state)142 void OveruseEstimator::UpdateNoiseEstimate(double residual,
143 double ts_delta,
144 bool stable_state) {
145 if (!stable_state) {
146 return;
147 }
148 // Faster filter during startup to faster adapt to the jitter level
149 // of the network. |alpha| is tuned for 30 frames per second, but is scaled
150 // according to |ts_delta|.
151 double alpha = 0.01;
152 if (num_of_deltas_ > 10 * 30) {
153 alpha = 0.002;
154 }
155 // Only update the noise estimate if we're not over-using. |beta| is a
156 // function of alpha and the time delta since the previous update.
157 const double beta = pow(1 - alpha, ts_delta * 30.0 / 1000.0);
158 avg_noise_ = beta * avg_noise_ + (1 - beta) * residual;
159 var_noise_ = beta * var_noise_ +
160 (1 - beta) * (avg_noise_ - residual) * (avg_noise_ - residual);
161 if (var_noise_ < 1) {
162 var_noise_ = 1;
163 }
164 }
165 } // namespace webrtc
166