DDPG-Enabled Accumulate-Then-Transmit Protocol for Interference-Assisted D2D-WPC Network Under Imperfect CSI

Abstract

This letter investigates an interference-assisted device-to-device (D2D) wireless powered communication (WPC) network that employs an accumulate-then-transmit (ATT) protocol to enable multi-slot energy buffering and power-adaptive transmission under stochastic interference dynamics for imperfect channel state information (CSI). The proposed system comprises a source node that harvests energy from an ambient interferer while accounting for the interference on the destination node. A non-convex optimisation problem is formulated to maximise the long-term average capacity under a practical non-linear energy-harvesting model, and a deep deterministic policy gradient (DDPG) algorithm is applied to solve it by learning the optimal transmit-power and time-allocation policies. To capture realistic system variations, real-time imperfect CSI are estimated using the MantisWave 5G D2D Box, serving as a hardware-based reflective evaluation platform. Simulation results show that the proposed ATT-DDPG scheme improves both average capacity and queue stability compared with HTT and other baseline scenarios, under both indoor and outdoor channel conditions. The average achieved capacities are 28.8460, Mbps and 24.9230 Mbps for indoor and outdoor channels, respectively.