Delay-Fluctuation-Resistant Underwater Acoustic Network Access Method Based on Deep Reinforcement Learning.

Area of Science:

• Underwater acoustic sensor networks (UASNs)
• Deep Reinforcement Learning (DRL)
• Network communication protocols

Background:

• Acoustic wave propagation in water causes variable communication delays in UASNs.
• Conventional DRL methods for UASNs struggle with random delay fluctuations and have low learning efficiency.

Purpose of the Study:

• To propose a DRL-based underwater acoustic network access method resistant to delay fluctuations.
• To enhance learning efficiency and decision-making in complex underwater acoustic environments.

Main Methods:

• Integrated delay fluctuations into the DRL state model for adaptive learning.
• Optimized a double deep Q-network (DDQN) structure for improved performance.
• Simulated the proposed method under varying delay fluctuation conditions.

Main Results:

• Achieved average improvements of 29.3% and 15.5% in convergence speed compared to other DRL methods.
• Significantly enhanced normalized throughput compared to TDMA and DOTS protocols.
• Demonstrated adaptability to random delay fluctuations in underwater acoustic links.

Conclusions:

• The proposed DRL method effectively addresses delay fluctuations in UASNs.
• The optimized DDQN enhances learning and decision-making capabilities.
• The method offers superior performance in terms of convergence speed and throughput for underwater acoustic networks.