Resource Allocation in Hybrid Radio-Optical IoT Networks using GNN with Multi-task Learning

Abstract

This paper addresses the problem of dual-technology scheduling in hybrid Internet-of-Things (IoT) networks that integrate Optical Wireless Communication (OWC) with Radio Frequency (RF). We first present an optimization formulation that jointly maximizes throughput and minimizes delivery-based Age of Information (AoI) between access points and IoT nodes under energy and link availability constraints. However, solving such NP-hard problems at scale is computationally intractable and typically assumes full channel observability, which is impractical in real deployments. To address this challenge, we propose the Dual-Graph Embedding with Transformer (DGET) framework, a supervised multi-task learning architecture that combines a two-stage Graph Neural Network (GNN) with a Transformer encoder. The first stage employs a transductive GNN to encode the known graph topology together with initial node and link states, such as energy levels, available links, and queued transmissions. The second stage introduces an inductive GNN for temporal refinement, enabling the model to generalize these embeddings to evolving network states while capturing variations in energy and queue dynamics over time through a consistency loss. The resulting embeddings are then processed by a Transformer-based classifier that models cross-link dependencies using multi-head self-attention. Simulation results show that hybrid RF-OWC networks outperform standalone RF systems by supporting higher traffic loads and reducing AoI by up to 20% while maintaining comparable energy consumption. Compared with optimization-based methods, the proposed DGET framework achieves near-optimal scheduling with over 90% classification accuracy, lower computational complexity, and improved robustness under partial channel observability.