A Cross-Layer Congestion Control Method Leveraging Radio Access Network Information

Most studies about congestion control only predict the conditions of network congestion based on the states of the transport layer. However, it is difficult to infer the packet loss caused by the channel, or even to distinguish the packet loss cause only from the transport layer state. We propose a cross-layer congestion control method based on the information of Radio Access Network (RAN). First, the information of RAN is added to the state component of Markov Decision Process (MDP). Then, the cause of packet loss are categorized, based on which the corresponding congestion window is set and dynamically adjusted by Soft Actor-Critic (SAC) algorithm. Simulation experiments demonstrate that the proposed Soft Actor-Critic Transmission Control Protocol Wireless (SAC-TCPW) significantly improves throughput when coexisting with TCP streams using NewReno. The average throughput rate reaches 71.06Mbps, which is 230.05 % higher than the lowest throughput rate of 21.53Mbps. The proposed method achieves the highest throughput across various scenarios while maintaining low latency and demonstrating notable advantages in fairness and friendliness.