EBi2ET protocol for self-sustainable IRS-assisted bidirectional communication: A transfer learning approach

Abstract

Internet of Things (IoT) networks require sensors to collect real-time data for various applications, but they have limited battery capacity for ubiquitous services. In this work, we address the critical energy limitations in intelligent reflecting surface (IRS)-assisted wireless powered communication networks (WPCNs) by introducing an energy borrowing (EB) mechanism to ensure reliable bidirectional communication. We develop a novel framework, EB-assisted integrated information and energy transfer (EBi${}^2$ET) protocol for self-sustainable IoT networks. Here, a hybrid access point (HAP) performs bidirectional communication with a source (or sensor) node via an IRS. For self-sustainability, HAP harvests energy from the ambient solar/wind. But for ubiquitous energy supply, HAP borrows energy from the power grid (PG) and returns the energy with interest, when it has sufficient energy from the ambient energy harvesting (EH). Further, HAP transfers the energy to the source via IRS using nonlinear radio frequency EH (RF-EH). Then, both HAP and source perform bidirectional communication. To enhance self-sustainable communication, we formulate a joint optimization problem aimed at maximizing the bidirectional data rate while accounting for energy transactions. Although managing these transactions and bidirectional communications in an unknown environment presents challenges, we address them effectively by transforming the problem into a Markov Decision Process (MDP). Leveraging the Deep Deterministic Policy Gradient (DDPG) algorithm, we derive the optimal online policy within a continuous action space. To further accelerate convergence and improve stability, we incorporate a transfer learning (TL)-based DDPG approach. Lastly, we demonstrate the superiority of our proposed method through performance comparisons with existing benchmark schemes, where the EBi${}^2$ET protocol achieves an impressive average performance gain of 193.58%. Furthermore, TD3 without TL, IRS-WPCN, EBi${}^2$ER, and i${}^2$ER protocols also yield notable improvements of 187.33%, 95.33%, 67.02%, and 43.18%, respectively, over the benchmark.