The deployment of large-scale LoRaWAN networks necessitates the optimization of competing metrics such as Packet Delivery Ratio (PDR) and Energy Efficiency (EE). Existing solutions typically oversimplify these challenges, often focusing on a singular performance metric or lacking the capability to adapt within dynamic channels. To mitigate these shortcomings, the proposed D-LoRa framework employs a Combinatorial Multi-Armed Bandit approach for resource allocation.
D-LoRa enhances network performance through decentralized decision-making, enabling nodes to learn and adapt autonomously in non-stationary environments. Further, the CD-LoRa framework includes a centralized initialization phase to provide quasi-optimal channel assignments, significantly accelerating the learning process in stationary scenarios. Both frameworks have been validated through extensive simulations and real-world field trials, highlighting improvements in PDR and EE of up to 10.8% and 26.1%, respectively. The implications of these developments illustrate a scalable and efficient path for advancing LoRaWAN technologies in various applications.
👉 Pročitaj original: arXiv AI Papers
