Efficient Fingerprint Augmentation Evaluation on the Antwerp LoRaWAN Setting

The trade-off between accuracy and cost of data collection has been identified as pivotal in the Localization of IoT devices within Low Power Wide Area Networks (LPWANs). This study investigates the effectiveness of a fingerprint augmentation method, ProxyFAUG, in improving the positioning accuracy and label efficiency of such localization systems, specifically utilizing the Antwerp LoRaWAN datasets. To set the ground, this work offers a detailed analysis of the usage of the Antwerp LoRaWAN datasets in the relevant literature, discussing localization methods, evaluation methodologies, and comparing performances. Subsequently, this work (i) evaluates ProxyFAUG’s capacity to improve the performance of the voluminous training set, while it also (ii) explores its capacity to improve label efficiency in settings of severe data scarcity. Results indicate a significant reduction in the median localization error ($\mathbf{1 9 \%}$) when the full training set gets augmented by ProxyFAUG. Moreover, the results demonstrate that the performance obtained when using the full original training set can be matched by only using 40% of the training data to feed ProxyFAUG. In addition, our work replicates in a LoRaWAN setting ProxyFAUG’s initial demonstration in a Sigfox setting, obtaining consistent results. The work uses Open Data and provides all described methods and experimentation as Open Code [1], promoting transparency of its claims and facilitating reproducibility of research results.