Added configurable offsets to the per-packet overhead in ANA.

webrtc/modules/audio_coding/audio_network_adaptor/bitrate_controller.h

Index: webrtc/modules/audio_coding/audio_network_adaptor/bitrate_controller.h
diff --git a/webrtc/modules/audio_coding/audio_network_adaptor/bitrate_controller.h b/webrtc/modules/audio_coding/audio_network_adaptor/bitrate_controller.h
index 740edbd4e039fec9dae8bcb78be441eeaab23f71..e8e69fc9ba88ac0ab5e9c5564cfe13f7969d6d71 100644
--- a/webrtc/modules/audio_coding/audio_network_adaptor/bitrate_controller.h
+++ b/webrtc/modules/audio_coding/audio_network_adaptor/bitrate_controller.h
@@ -20,10 +20,15 @@ namespace audio_network_adaptor {
 class BitrateController final : public Controller {
  public:
   struct Config {
-    Config(int initial_bitrate_bps, int initial_frame_length_ms);
+    Config(int initial_bitrate_bps,
+           int initial_frame_length_ms,
+           int fl_increase_overhead_offset,

[ossu, 2017/09/14 12:16:14]

What does this mean?

[ivoc, 2017/09/14 13:03:38]

It is the offset that should be added to the packet overhead value that the
bitrate controller gets in the UpdateNetworkMetrics function (I guess it comes
from the BWE). That value is too optimistic, as it does not consider the header
compression that happens on typical networks.
We would like to apply different offsets when the frame length is increased and
decreased, to prevent bandwidth overuse on poor networks. Alex may be able to
provide a better explanation here.

[alexnarest, 2017/09/14 13:26:11]

This is quite precise explanation. There is no easy way to do precise overhead
calculation because we do not know which operators compress headers and what is
the level of the compression.

+           int fl_decrease_overhead_offset);
     ~Config();
     int initial_bitrate_bps;
     int initial_frame_length_ms;
+    int fl_increase_overhead_offset;
+    int fl_decrease_overhead_offset;
   };
 
   explicit BitrateController(const Config& config);