Adaptive Underwater Medium Access Control Protocol based on Reinforcement Learning

In underwater acoustic communication networks (UACNs), mobile nodes play a crucial role, including continuously tracking and detecting multiple underwater engineering situations within a certain range, transmitting sensing environment data through underwater acoustic channels. In this article, we propose a medium access control protocol for high mobility underwater acoustic communication networks. By applying deep reinforcement learning (DRL) technology, nodes can adapt faster and more efficiently when moving between multiple independent underwater acoustic communication networks, avoiding the impact of high uncertainty in node motion trajectories on nearby networks. Nodes utilize the high propagation delay of underwater acoustic networks to train models and learn the optimal roundtrip transmission delay (RTT) and window size in different network environments, thereby improving the overall throughput of the network. The simulation results show that our proposed protocol is superior to the latest existing protocol DR-DLMA and UW-ALOHA-QM, achieving a channel utilization improvement of up to 38% in the most variable scenario.