Smart duty cycle control with reinforcement learning for machine to machine communications

Machine to machine (M2M) communications is one of the key underpinning technologies for Internet of Things (IoT) applications in 5G networks. The large scale of M2M devices imposes challenge on conventional medium access control protocols. In this paper, we propose a reinforcement learning (RL) based duty cycle control for dominant short-range technology IEEE 802.15.4 to provide high performance and reliable M2M communication. We first model a practical multi-hop M2M communication network that takes various network dynamics into consideration. Then, we mathematically derive the distributed optimal duty cycle control policy to optimise the energy efficiency, end-to-end delay and transmission reliability. Finally, a RL based practical duty cycle control is developed to learn the optimal policy directly without priori network information, which contributes to the smart duty cycle control under various network dynamics. Simulation results show that the proposed RL based duty cycle control achieves the best balance between optimality and stability, compared with the optimal and the existing IEEE 802.15.4 duty cycle controls.