A Deep MARL-Based Power-Management Strategy for Improving the Fair Reuse of UWSNs

Abstract

Providing qualified and fair communications for underwater wireless sensor networks (UWSNs) has garnered considerable interest in light of scarce resources and dynamic channel conditions. Fairness is critical in various situations, including emergency communications in resource-constrained underwater networks that balance load and energy amongst nodes to optimize network performance. Existing solutions for fair communications, on the other hand, frequently come at the expense of network capacity. This paper focuses on the power management strategy that jointly optimizes the reuse, fairness, and capacity of UWSNs while also proposing a new metric for fair spatial reuse in networks: the fair reuse index. We observe that the fair reuse index for UWSNs is strongly reliant on the network density, channel conditions, and application requirements. As a result, UWSNs commonly exhibit load imbalance and struggle to meet required network lifetimes. Towards this end, we propose DMPM, a Deep Multi-agent reinforcement learning-based Power Management strategy for increasing the fair reuse of UWSNs. DMPM strives to maximize the network’s fair reuse while allowing for gentle network capacity decrease. In two representative communication scenarios, numerical results demonstrate that DMPM achieves a significantly better trade-off between network capacity and fair reuse than baseline techniques. We also construct three reward functions for DMPM and discuss how different reward functions affect node behaviors. Delivery delays of different models are discussed. We hope that the work provided in this study will prove to be invaluable in the design and optimization of UWSNs.