Enhancing Secure Energy Efficiency of SWIPT IoT Network Considering IRS and Artificial-Noise: A Deep Learning Approach
Kimchheang Chhea, Jung-Ryun Lee · IEEE Access · 2025
Internet of things (IoT) networks connect sensing devices to the Internet for tasks like tracking, monitoring, and data analysis. However, the large data generated, especially in resource-heavy services, challenges energy efficiency (EE) and security. The use of intelligent reflecting surfaces (IRS) in the physical layer security of IoT devices shows promise in improving secrecy performance and EE. This paper addresses IRS and AN-aided secure communication against eavesdroppers, where eavesdroppers are IoT devices equipped with simultaneous wireless information and power transfer (SWIPT) using power splitting policy. Our goal is to maximize secure EE, where the secure EE is defined as the secrecy rate over the power consumption of the user. We aim to jointly optimize phase-shift of IRS, transmit power of jammer and user, to maximize the objective function while ensuring that the secrecy rate is greater than the predefined threshold while allowing the eavesdroppers to harvest energy using SWIPT. We first formulate a joint optimization model of IRS and AN aided secure communication, which is a non-convex optimization problem with linear and non-linear constraints. To solve this problem, we propose a deep neural network (DNN) algorithm that utilizes a loss function derived from Lagrange duality function. The simulation results show that the proposed DNN achieves near-global optimal solutions. In addition, the results also prove the effectiveness of implementing IRS to enhance the secure EE of the user.