Machine Learning-Based Performance-Efficient MAC Protocol for Single Hop Underwater Acoustic Sensor Networks

Advances in the development of underwater full duplex sensors have meant they can now provide solutions to underwater acoustic sensor networks (UASN) that suffer from narrow bandwidth and long end-to-end transmission delay. In this paper, a full machine learning-based duplex medium access control (MAC) protocol for single hop underwater acoustic sensor networks is proposed to enhance the performance on throughput, delay and fairness access. The proposed protocol is designed in different access schemes for uplink and downlink transmission according to their main uses, respectively. The access scheme of the downlink is contention free to ensure that command information can be delivered to sensor nodes with a low collision rate. A hybrid scheme is utilized in the uplink to maximize the network throughput to meet transmission requirements of monitoring data collected by sensor nodes. In this way, transmission resources can be used more efficiently and fairer. In addition, as a unique characteristic of underwater acoustic communication, propagation time in the transmission is taken into consideration in the design of the MAC protocol. Simulation results and analysis exhibited that the proposed protocol performs significantly better than state-of-the-art MAC protocols for UASN on network throughput, end-to-end delay and transmission fairness. Underwater acoustic sensor networks (UASN), machine learning medium access control (MAC) protocol, full duplex transmission, network throughput, and end-to-end delay.