MobStream: A Reinforcement-learning driven mobile streaming methodology based on Multipath QUIC

Currently, applying smartphone or other mobile device to stream the live content becomes increasingly popular. This paper proposes MobStream, a novel reinforcement learning-driven solution based on multipath QUIC (MP-QUIC). MobStream maximizes streaming capacity by fully utilizing the bandwidth of both WiFi and cellular networks through MP-QUIC. To address the high dynamics of mobile environments, MobStream incorporates partial reliability and a layered coding scheme, enabling adaptation to varying bandwidth conditions while minimizing retransmissions. Furthermore, to align data delivery across multiple paths and optimize streaming bitrate, we formulate the problem as a stochastic optimization task and demonstrate its solution using reinforcement learning methods. Extensive experiments demonstrate that MobStream outperforms state-of-the-art solutions in terms of bitrate, packet loss, and delay.