Abstract : The rapid growth of connected devices and data traffic in modern wireless networks has made efficient resource management increasingly challenging. Applications such as smart grids, industrial IoT, autonomous transportation, and smart city infrastructures require high-performance, reliable communication. Effective link scheduling is crucial for optimizing spectrum allocation, prioritizing transmissions, and minimizing interference in these dynamic environments. This paper introduces a novel Topology-aware Greedy Link Scheduling algorithm, which leverages lightweight Deep Neural Networks (DNNs) and Explainable AI (XAI) techniques. Our approach is designed for networks with known topologies and dynamically changing link priorities, typical in environments such as smart grids and industrial IoT systems. It enhances scheduling by learning a network’s topological importance offline and applying a dual-level sorting mechanism in real-time. Extensive simulations demonstrate significant improvements over benchmark algorithms, such as Local Greedy Scheduling and Greedy Maximal Scheduling, in terms of total weight and active link count, indicating better spectrum and bandwidth usage. The simulations further show that our algorithm’s scheduling performance is on par with state-of-theart Machine Learning (ML) methods. Additionally, the algorithm consistently performs real-time scheduling computations much faster than ML methods. These results highlight our algorithm’s robustness, scalability, and practical applicability in dynamic wireless environments.
Index terms : Link Scheduling, Deep Learning Optimization, Explainable AI (XAI) Techniques, Topology-aware Wireless Networks