Propagation characterization of a low altitude platform for wireless internet of things network using hybrid machine learning

Abstract

The Internet of Things (IoT) is rapidly expanding to connect everything in a smart future world. Various new challenges have appeared, including power consumption, quality of service, localization, security, and accurate modeling and characterization of wireless channel propagation to meet IoT application requirements while maintaining optimal performance. However, previous studies on wireless IoT characterization and modeling are restricted to controlled environments and inaccurate for deployments in harsh tropical environments. Besides, real-world performance evaluations and reference guidelines for selecting the optimal wireless IoT technology are lacking for such deployments. This work describes the design and development of a novel Low Altitude Platform-based Airborne IoT Network (LAP-AIN) system architecture for water quality monitoring, combining existing wireless technologies with the aid of a LAP to relay data over long distances. It aims specifically at communication range evaluation and characterization of wireless channel propagation for the developed LAP-AIN system. The study begins by comprehensively evaluating the performance of three wireless IoT technologies, namely Narrowband-IoT (NB-IoT), Sigfox®, and LoRaWAN®, to determine the most suitable one for the development of the LAP-AIN system. NB-IoT was evaluated in the real world according to coverage parameters, path loss (PL), packet delivery rate (PDR), and latency limits. The power consumption and average battery life were then calculated and compared for all technologies, considering various factors. A measurement campaign was then conducted to assess the LAP-AIN system focusing on the communication link reliability, as well as the LAP stability and robustness. Finally, a novel three-stage stacked Machine Learning (ML)-based semi-empirical PL model, combining a free space PL model, a Stepwise ML model, and an Ensemble of Bagged Trees ML model, is proposed for Long Range (LoRa) wireless communication. An overall PDR of 91.76% was achieved for NB-IoT, indicating its capability to handle high data rates with minimal signal quality. However, latency varied greatly with signal quality (170 ms to 10 s), reducing NB-IoT battery life. LoRaWAN® technology, followed by Sigfox®, proved to be more battery efficient than NB-IoT. Hence, LoRaWAN® was selected for integration into the LAP-AIN system. The results also validated the proposed system’s effectiveness, unique characteristics, and capabilities in such a harsh environment. Several critical success factors for the LAP platform were also highlighted to achieve extensive and reliable coverage. The LAP evaluation revealed a significant difference in PDR for different spreading factor (SF) configurations. For instance, switching SF7 to SF12 increased PDR by 28.7%. Meanwhile, increasing the gateway height increased the PDR by 29.2% for similar SF configurations. The evaluation also revealed that none of the established PL models are suitable to represent harsh tropical environments. Contrary to expectations, existing models proved the lowest prediction accuracy of 38.3% (R2 = 0.383) even after modification. Finally, the results indicated that the proposed model significantly improves the PL fit, has a high correlation coefficient, and uniformly distributed residuals across all PL range. It achieved 89.2% prediction accuracy for testing samples and a remarkable 91.8% accuracy for training and overall measurement samples. Hence, it was concluded that the proposed model is more flexible and provides the highest prediction accuracy in Malaysia’s rural and suburban areas compared to the existing conventional models.