Analysis of time-weighted LoRa-based positioning using machine learning

Localization systems have gained attention owing to the paradigm shift from human-centric communication systems (1G to 4G) to the machine-to-machine architectures (5G and beyond). The commercial localization applications standardized for 5G systems have served as a precursor to the cardinality of positioning technologies in the next-generation communication systems. The stringent requirements of these use-cases have motivated researchers to propose novel architectures and techniques to develop scalable, accurate, reliable, robust, and low-power positioning and tracking systems. Low-power wide-area network (LPWAN) technologies have found their niche in the Internet-of-thing (IoT)-focused industrial and research communities, since they promise wide area coverage to many battery-operated devices. LoRaWAN, with regulatory features and high network density, has emerged as the widely adopted long-range, low-power solution for scheduled IoT applications. This paper explores the feasibility of LoRa technology for satellite navigation-independent positioning, using received signal strength indicator (RSSI) fingerprinting. We explore traditional path-loss models, machine learning and deep learning techniques to develop an accurate RSSI-to-distance mapping. We further use the analytically optimal model as the underlying ranging function for trilateration-based deterministic positioning. The results indicate that LoRa technology is a feasible alternate for fingerprinting-based positioning in line-of-sight and non-line-of-sight scenarios, with accuracies ranging from 6 to 15 m.