Medium Access Control Applied to Long-Range Underwater Acoustic Networks Using an SNR-Aware Adaptive Learning Agent

Abstract

In this paper, a signal-to-noise ratio (SNR)-aware deep reinforcement adaptive learning–based medium access control (SNR-DRAL MAC) protocol is proposed for Underwater Acoustic Networks (UANs). By adapting transmission decisions to channel conditions, the proposed approach reduces packet collisions and efficiently utilizes idle time slots left unused by other MAC agents, thereby improving overall network throughput in challenging underwater environments. The novelty of this paper lies in the proposed SNR-aware Deep Reinforcement Adaptive Learning (DRAL) MAC protocol that explicitly couples physical-layer reliability with MAC-layer decision-making under the unique constraints of underwater acoustic networks. By learning directly from the Acknowledgment (ACK) outcomes driven by SNR-dependent Frame Error Rate (FER) and long propagation delays, the proposed agent adapts its transmission policy in a highly non-static channel environment characterized by collisions, burst errors, and delayed feedback. Unlike classical Stop-and-Wait Automatic Repeat reQuest (ARQ)-based ALOHA schemes and prior learning-based MAC approaches such as Actor-Critic, the proposed framework demonstrates robust throughput gains across a wide range of SNR scenarios and under both AWGN and multipath-induced channel variability, thereby validating the effectiveness of learning-based MAC adaptation in realistic underwater environments.