RR-WebRTC: Efficient Real-Time Communications With Fast Loss Recovery For WebRTC

In this work, we addresses the challenge of enhancing network utilization in WebRTC under packet loss scenarios, aiming to improve the overall quality and user experience of real-time multimedia applications. Through extensive real-world experiments, we identify critical relationships between packet loss durations and the sending rate in WebRTC, revealing a recoverable-threshold phenomenon that significantly impacts network efficiency. Based on these findings, we introduce RRWebRTC, a novel joint Reinforcement Learning (RL) and rulebased online rate control algorithm. RR-WebRTC leverages a loss-rate model to dynamically adapt to varying network conditions, demonstrating superior generalization and efficiency compared to existing methods. Our evaluations show that RRWebRTC effectively reduces recovery times by at least 50% and increases the sending rate by approximately 10% on average, with a maximum improvement of 100%.