Improve adaptation to reordered packets in delay manager.

This is done by adding a reorder optimizer that estimates the probability of receiving reordered packets.

The optimal delay is decided by balancing the cost of increasing the delay against the probability of missing a reordered packet, resulting in a loss. This balance is decided using the `ms_per_loss_percent` parameter.

The usage and parameters can be controlled via field trial.

Bug: webrtc:10178
Change-Id: Ic484df0412af35610e74b3a6070f2bac7a926a63
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/231541
Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
Commit-Queue: Jakob Ivarsson <jakobi@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#34954}

modules/audio_coding/neteq/reorder_optimizer_unittest.cc

@@ -0,0 +1,70 @@
+/*
+ *  Copyright (c) 2021 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "modules/audio_coding/neteq/reorder_optimizer.h"
+
+#include "test/gtest.h"
+
+namespace webrtc {
+
+namespace {
+
+constexpr int kForgetFactor = 32745;  // 0.9993 in Q15.
+constexpr int kMsPerLossPercent = 20;
+constexpr int kStartForgetWeight = 1;
+
+}  // namespace
+
+TEST(ReorderOptimizerTest, OnlyIncreaseDelayForReorderedPackets) {
+  ReorderOptimizer reorder_optimizer(kForgetFactor, kMsPerLossPercent,
+                                     kStartForgetWeight);
+  EXPECT_FALSE(reorder_optimizer.GetOptimalDelayMs());
+
+  // Delay should not increase for in-order packets.
+  reorder_optimizer.Update(60, /*reordered=*/false, 0);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 20);
+
+  reorder_optimizer.Update(100, /*reordered=*/false, 0);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 20);
+
+  reorder_optimizer.Update(80, /*reordered=*/true, 0);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 100);
+}
+
+TEST(ReorderOptimizerTest, AvoidIncreasingDelayWhenProbabilityIsLow) {
+  ReorderOptimizer reorder_optimizer(kForgetFactor, kMsPerLossPercent,
+                                     kStartForgetWeight);
+
+  reorder_optimizer.Update(40, /*reordered=*/true, 0);
+  reorder_optimizer.Update(40, /*reordered=*/true, 0);
+  reorder_optimizer.Update(40, /*reordered=*/true, 0);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 60);
+
+  // The cost of the delay is too high relative the probability.
+  reorder_optimizer.Update(600, /*reordered=*/true, 0);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 60);
+}
+
+TEST(ReorderOptimizerTest, BaseDelayIsSubtractedFromCost) {
+  constexpr int kBaseDelayMs = 200;
+  ReorderOptimizer reorder_optimizer(kForgetFactor, kMsPerLossPercent,
+                                     kStartForgetWeight);
+
+  reorder_optimizer.Update(40, /*reordered=*/true, kBaseDelayMs);
+  reorder_optimizer.Update(40, /*reordered=*/true, kBaseDelayMs);
+  reorder_optimizer.Update(40, /*reordered=*/true, kBaseDelayMs);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 60);
+
+  // The cost of the delay is too high relative the probability.
+  reorder_optimizer.Update(600, /*reordered=*/true, kBaseDelayMs);
+  EXPECT_EQ(reorder_optimizer.GetOptimalDelayMs(), 620);
+}
+
+}  // namespace webrtc