An Energy-Efficient Communication Protocol for Power-Constrained IoT Networks: A Deep Reinforcement Learning Approach

Power-limited devices (or sensors) constrain the deployment of modern IoT networks, such as Next-Generation Industrial IoT (NG-IIoT). These networks are envisioned as the key enablers to facilitate connectivity to billions of devices for applications like smart industries, smart healthcare, etc. This paper presents an optimal up-link communication protocol for a power-limited unlicensed sensor operating among numerous licensed sensors communicating in a time division multiple access (TDMA)-based scheme. The transmission of the power-limited sensor is ensured by employing the non-orthogonal multiple access (NOMA) technique during the time slot of a licensed sensor. To ensure energy-efficient communication, we maximize the throughput of the power-limited sensor using a deep reinforcement learning (DRL) framework recognized as a combined experience replay deep deterministic policy gradient (CER-DDPG) algorithm. Our simulation results demonstrated that the CER-DDPG-based communication protocol outperforms the benchmark schemes, such as DDPG and stochastic algorithms, in terms of throughput.