Minimizing Power in Buffer-Aided SWIPT-WPRNs Using Deep Reinforcement Learning

In this paper, we design a network utilizing a wirelessly-powered relay (WPR) to simultaneously receive power and information from an access point (AP) and forward the information to a distant wireless device (WD) that cannot receive information directly from the AP. The WPR is equipped with a finite battery and a data buffer and using the saved energy re-transmits the AP's data to the WD in a decode-and-forward (DF) manner. The objective is to completely deliver a variable-sized data packet to the WD while minimizing the total energy consumption of the AP within a defined time frame composed of multiple slots. Using deep reinforcement learning (DRL) methods, we devise an algorithm to train a proximal policy optimization (PPO) agent that optimizes the time and power allocation as well as data transfer rates of the AP and WPR in this problem. The simulation results show that the proposed approach can efficiently optimize the network to guarantee delivering the complete data packet. Moreover, the energy consumption of the proposed technique is much lower than the optimal greedy algorithm which occasionally fails to deliver the data packet due to deep channel fades.