LoRaDoctor: LLM-Driven Diagnosis and Adaptive Policy Optimization for Reducing Packet Error Rate in LoRaWAN Networks
Conference
Ma, D, Ma, A, Luo, S et al. (2025). LoRaDoctor: LLM-Driven Diagnosis and Adaptive Policy Optimization for Reducing Packet Error Rate in LoRaWAN Networks
. 1181-1188. 10.1109/ICMLA66185.2025.00181
Ma, D, Ma, A, Luo, S et al. (2025). LoRaDoctor: LLM-Driven Diagnosis and Adaptive Policy Optimization for Reducing Packet Error Rate in LoRaWAN Networks
. 1181-1188. 10.1109/ICMLA66185.2025.00181
LoRaWAN is widely used for large-scale Internet of Things (IoT) deployments, but real-world reliability is often affected by high packet error rate. Existing optimization methods, such as heuristics or supervised learning, cannot fully capture the effects of environment, spatial layout, and network dynamics, which limits their adaptability. In this paper, we present LORADOCTOR, the first framework that leverages Large Language Models (LLMs) to optimize LoRaWAN. LORADOCTOR performs causal analysis, generates adaptive transmission policies, and predicts network performance in an interpretable way. We perform simulation-based evaluations using a year-long dataset collected from eight sensors deployed in East London. The results show that LORADOCTOR can significantly reduce packet error rates compared to both the default LoRaWAN settings and standard machine learning methods. Our evaluation also identifies distance-based path loss, temperature effects, and human activity as the main causes of packet error rate, showing its potential to support more reliable and adaptive LoRaWAN deployments in future urban environments.
