Reinforcement Learning Based Resource Allocation in IRS Assisted SWIPT Systems

In this paper, an intelligent reflecting surface (IRS)-assisted a simultaneous wireless information and power transfer (SWIPT) communication system is proposed, the IRS harvests energy to supply the energy consumption of its circuit operation. The harvested energy at an energy user (EU) is maximized by jointly optimizing the transmits beamforming, IRS phase shifts and time switching factor. The formulated problem is non-convex due to multiple coupled variables. We propose the deep reinforcement learning (DRL) based algorithm, in which the joint design can be gained and continuously improved through interacting with the environment to solve the formulated non-convex problem. Simulation results further prove the effectiveness of the proposed DRL approach in optimizing multiple coupled variables.