Teacher–student learning based low complexity relay selection in wireless powered communications

Abstract

Radio Frequency Energy Harvesting (RF-EH) networks are pivotal in enabling massive Internet-of-Things by facilitating controlled, long-distance energy transfer to energy-constrained devices. Relays, which assist in either energy or information transfer, significantly enhance the performance of such networks. However, the relay selection problem in multiple-source–multiple-relay RF-EH networks poses substantial computational challenges. To address these, this paper proposes a novel deep-learning-based relay selection framework that integrates convolutional neural networks (CNNs) and teacher–student learning. Specifically, the joint relay selection, time allocation, and power control problem are studied under non-linear EH conditions. First, the optimal solution to the time and power allocation problem for a given relay selection is derived. Then, the relay selection problem is formulated as a classification task, and two CNN-based architectures are proposed. To further improve computational efficiency without compromising accuracy, the teacher–student learning paradigm is employed, wherein a smaller student network is trained with the distilled knowledge of a larger teacher network. A novel dichotomous search-based algorithm is introduced to determine the optimal architecture of the student network. Simulation results demonstrate that the proposed solutions achieve lower complexity compared to state-of-the-art iterative approaches while maintaining optimality.