6G wireless communication faces challenges of energy requirement to support billions of sensor devices in the Internet of  Things (IoT). Challenges can be overcome if sensor nodes harvest energy from the RF signal of primary user in the network. Duration for non overlapping time slot for energy harvesting and data transmission in a single time frame is a trade off in challenging wireless network, that can be addressed by  Reconfigurable Intelligent Surfaces (RIS). Insufficient energy at the sensor node causes failure of data transmission which leads to increased outage probability.
This work aims to minimize outage probability using reinforcement learning (RL) approaches. A Deep Deterministic Policy Gradient (DDPG) RL framework is proposed to minimize outage probability while optimizing the energy harvesting time fraction and RIS phase-shift control. Performance of DDPG algorithm is compared with a baseline Q-learning approach. Simulation results demonstrate that the proposed DDPG method significantly outperforms Q-learning, reducing outage probability by 28\% in the optimal energy harvesting region and achieving more than 56\% improvement as the number of RIS elements increases with reduced learning latency by nearly 40\%.

RIS,DDPG