Nuwa-RL: A Reinforcement Learning based Receiver-side Congestion Control Algorithm to Meet Applications Demands over Dynamic Wireless Networks

The advent of the wireless network has greatly increased the amount of data transmitted through the networks. Depending on the location of the user and the conditions of the network, the data transmitted over the network show the different transmission characteristics. Data transmission over long periods of time is very sensitive to network vertical switching. It's highly susceptible to massive loss of throughput problems. Therefore, it is essential to design a congestion algorithm for networks that switch frequently. However, traditional congestion control is a deployment based on a rule-based heuristics and tested on a predetermined set of benchmarks. As a result, these congestion control algorithms cannot solve the problem of access point switching under real networks. To address these issues, we propose Nuwa-RL, a receiver-side driven congestion control scheme. It combines the reinforcement learning algorithm PPOLSTM with the receiver-side congestion control algorithm for controlling the data transmission process. Nuwa-RL uses reinforcement learning to adjust the heuristic algorithm parameters in order to optimise the data transfer process. We verified that the algorithm is effective in reducing throughput degradation during network switching. Experience shows that Nuwa-RL can improve the data transmission ability of the congestion control algorithm in wireless networks.