Reinforcement Learning-Based Power Control for MACA-Based Underwater MAC Protocol

Abstract

A major amount of the energy of battery-powered sensors is spent during packet transmissions. This issue has led to the development of power-control-based multiple-access collision avoidance (MACA) protocols that can reduce the packet transmission power and conserve energy. However, the reduction in transmission power renders the packets susceptible to collisions. To reduce these collisions while maintaining high energy efficiency, we propose a power control protocol that utilizes reinforcement learning to choose the optimal transmission power. The total reward is determined by the occurrence of a collision, amount of transmission power used, frequency of DATA packet retransmissions, and update of the interference range. A key feature of the proposed protocol is that it enables sensors to prevent collisions without any prior knowledge of interferences, thus eliminating the need for additional signaling. Simulation results under varying average traffic loads indicate that the proposed protocol can improve network throughput by up to 20% compared to benchmark protocols, while minimizing network energy consumption with a similar gain and reducing collisions per packet by more than 35%. These results demonstrate that the proposed protocol is effective.

Keywords

• Collisions
• interference
• medium access control
• power control
• Q-learning
• underwater acoustic sensor networks