Spreading Factor and RSSI for Localization in LoRa Networks: A Deep Reinforcement Learning Approach

Recent advancements in Internet of Things (IoT) technologies have resulted in a tightening of requirements from various applications including localization in LoRa networks. To address the growing demand for LoRaWAN-powered IoT location-based services, accurate localization solutions are more crucial than ever. As such, in this work, we develop an accurate deep neural network based localization framework over a LoRa network by proposing a novel approach that builds the network radio map with the combination of RSSI recordings and the spreading factors (SF) used by LoRa devices during the transmissions. Then, we validate our framework using a publicly available experimental dataset recorded in an urban LoRa network. The performance evaluation shows the prominence of adding the spreading factor as an additional fingerprint, since we can achieve, by our approach, an improvement in localization accuracy by up to 6.67% compared to the state-of-the-art methods which employ uniquely the RSSI fingerprints. Additionally, we provide an analysis of the impact of the SF on the localization performance which reveals that the localization accuracy relies on the SF used for position request. Finally, we propose a deep reinforcement learning based localization system to capture the ever-growing complexity of LoRa networks environment and cope with the scalability issue in LoRa enabled massive IoT, and the results show an improvement of 63.3% in terms of accuracy.