A Deep Reinforcement-Learning-Based Relay Selection for Underwater Sensors Network

Abstract

Abstract Due to their limited frequency range and fast fading channels, underwater sensor networks (USNs) are vulnerable to collisions of packets. In this paper, we propose a deep reinforcement learning-based relay selection scheme with shortest latency (DRL-SL) for USNs that enables to choose the relay based on the state that comprised of the bit error rate (BER) of the previous transmission, and the jamming power measured by the relay node. The DRL-SL-based relay selection scheme completed in two phases. In the first phase, a deep neural network based learning is performed and second phase is the real-time interaction with the underwater sensor network. Numerical results give the bound on how efficiently the system performs in terms of bit error rate, energy use, and node utility. According to the numerical results, the proposed DRL-SL based relay selection scheme can enhance relay performance in comparison to the benchmark underwater relay techniques.