FlexADR: a Machine Learning-Based ADR Mechanism Including New Parameters for Optimizing LoraWAN Transmission

LPWAN (Low Power Wide Area Networks) play an essential role in the Internet of Things (IoT) evolution. Among these networks, LoRaWAN (Long Range Wide Area Network) stands out for its advantageous characteristics and its popularity within the scientific community. This network incorporates a mechanism called Adaptive Data Rate (ADR), which dynamically adjusts the spreading factor (SF) and transmission power (TP). However, this mechanism converges with difficulty or slowly towards optimal configurations for better performances. This paper proposes an innovative approach, called Flexible ADR (FlexADR) to outperform ADR by integrating two new parameters, namely the coding rate and bandwidth, into the adjustment process, using machine learning. In fact, it relies on a combination of unsupervised learning (K-means) and reinforcement learning (SARSA) to dynamically adapt transmission parameters. This solution is implemented and simulated in OMNeT++ using the FLoRa and AI-Toolbox frameworks. Simulation results show that our new approach improves network performance in terms of packet reception rate and energy efficiency, compared with ADR.