A Scalable and Fair Power Allocation Scheme Based on Deep Multi-Agent Reinforcement Learning in Underwater Wireless Sensor Networks

Abstract

Providing qualified communications and optimizing network performance for Underwater Wireless Sensor Networks (UWSNs) is difficult due to limited battery power and storage, unpredictable channel conditions, and significant communication interference (including ambient noise and inter-nodes interferences). Power allocation is an important technology for UWSNs. In this paper, we analyzed the constraints of UWSNs and proposed a distributed power allocation scheme based on deep multi-agent reinforcement learning, which dynamically tunes the independent transmit power according to changing environments. We improve the number of concurrent communications and optimizes network capacity by fully leveraging the spatial separation of wireless networks. We compared the proposed approach with baseline methods in network capacity and communication fairness in different communication scenarios when the number of underwater nodes increases. Experiments confirmed that our solution achieves a significantly better trade-off between network capacity and fairness, while still satisfying the lifetime criteria.