A Deep Reinforcement Learning Approach for Backscatter-Assisted Relay Communications

A backscatter-assisted relaying network (BRN) has been recently proposed to improve data rate, transmission range, and energy efficiency of D2D communications. In the BRN, as a D2D transmitter actively transmits data to a receiver in its time slot, other D2D transmitters can act as relays, i.e., helpers, through backscattering signal from the D2D transmitter to the receiver. This passive relay method has shown to be effective in terms of diversity gain. However, this impairs energy harvesting by the helpers and thus degrades their active data transmission performance. Therefore, the problem in the BRN is to optimize backscatter relaying policies, i.e., reflection coefficients, for the helpers to maximize the total network throughput over time slots. Finding the optimal decisions is generally challenging as energy in batteries, i.e., energy states, of the helpers and communication channels are dynamic and uncertain. In this letter, we propose to adopt the Deep Deterministic Policy Gradient (DDPG) algorithm to determine the optimal reflection coefficients of the helpers. The simulation results show that the proposed DRL scheme significantly improves the throughput performance.