A lightweight, fully-distributed AI framework for energy-efficient resource allocation in LoRa networks

As the Internet of Things (IoT) continues to grow rapidly, efficient resource utilization is crucial for the sustainability and performance of IoT networks. In this context, LoRa technology, known for its low-power, long-range communication, has become popular for IoT applications. However, the limited energy and spectrum resources in Long Range (LoRa) networks present challenges in achieving optimal network performance in dense deployments. In particular, existing centralized approaches, such as LoRa Adaptive Data Rate, may fail to scale up to the large networks typical of IoT scenarios. To address all of these issues, we propose a smart, fully-distributed resource allocation scheme based on multi-agent cooperative Q-learning approach. Simulations results prove that our approach improves the Packet Delivery Ratio (PDR) and reduces the energy consumption of up to 43% as compared to fixed SF and random non-smart strategies, respectively, while also keeping the decision process at a device-level, with no centralized entities involved.