Sensorless Air Temperature Sensing Using Lora Link Characteristics
Conference
Ma, A, Rodriguez, JT, Sha, M et al. (2025). Sensorless Air Temperature Sensing Using Lora Link Characteristics
. 1-8. 10.1109/DCOSS-IoT65416.2025.00010
Ma, A, Rodriguez, JT, Sha, M et al. (2025). Sensorless Air Temperature Sensing Using Lora Link Characteristics
. 1-8. 10.1109/DCOSS-IoT65416.2025.00010
Air temperature monitoring is essential for many Internet of Things (IoT) applications. Many existing applications rely on the readings provided by the weather stations maintained by federal, regional, or local government agencies. While the accuracy of the data provided by those weather stations is high, the ability of such data to reflect the temperature variability experienced by urban populations is generally low because the measurements are collected at the mesoscale. In reality, the air temperature varies at the microscale and local scale, and the health risks associated with extreme weather are assumed to vary with the exposure. In this paper, we present LoRATEMP, a novel solution that uses LoRa link characteristics and advanced machine learning techniques to predict air temperature. LORATEMP leverages a unique correlation map representation and a novel dual attention network to capture the complex dependency between LoRa link characteristics and air temperature, and employs adversarial domain adaptation to transfer the temperature prediction knowledge learned from one device to those without temperature sensors using a few temporal measurements. We implement LORATEMP and evaluate it in real-world environments. Experimental results show that LORATEMP significantly outperforms all baselines and reduces air temperature prediction errors by at least 30 %.
