Weather-Aware LoRaWAN Sensor localization

In the context of low-power wide-area networking (LPWAN) Internet-of-Things (IoT) communication, GNSS-free localization methods based on Network-as-a-Sensor (NasS) principles are preferable in terms of hardware costs and power consumption. However, applying these techniques to wireless communication systems operating in the sub-gigahertz frequency spectrum, e.g. LoRaWAN, imposes non negligible errors in localization estimation [5], [14], [18]. In this work, we improve the LoRaWAN localization methods from [5], which are based on a combination of RSSI fingerprinting and machine learning regression algorithms, i.e. k nearest neighbor and random forest regression, by integrating additional climatic parameters. To demonstrate the benefits of the proposed approach, we further give a comparison to traditional multilateration based techniques based on Time-of-Arrival (ToA) and range-estimation based localization methods. The performance evaluation shows that our approach can improve state-of-the-art location methods by 23 % on the same dataset.