Radio Environment Compensation in a Narrowband IoT Positioning System : Using Radio Signal Metrics Between Stationary Devices

Abstract: The Internet of Things (IoT) has emerged as a powerful tool for meeting our need to collect information about and interact with our environments. One important aspect of this technology is positioning which imposes requirements on both the energy consumption and the arrangement of the systems. For devices that run on battery, low energy consumption is critical and for large deployments, there is a need to minimize the number of devices to reduce cost. Narrowband sub-GHz positioning systems allow for long-range and low-energy functionality at the cost of positioning accuracy compared to wider bandwidth systems. One significant factor that impacts the accuracy of positioning systems is the quality of the radio environment. The aim of this thesis is to investigate methods for using stationary devices in positioning systems for mitigating the effect of the radio environment. The focus lies on a specific IoT narrowband sub-GHz system that use received signal strength indicator and time-of-flight to create range estimates. To meet the objectives, a measurement analysis is performed on a real system deployment. The analysis results in two methods for compensating the range estimations between the devices to locate and the stationary devices in the system. The first method creates a compensation factor based on the measurements from a closely located stationary device. The second method implement and tests six regression models trained on measurements between one or several closely located stationary devices. The results show that both approaches improve the range estimates in the system for two different system deployments. The best method show approximately 76 % improvement on the first deployment and 66 % improvement on the second. The results also show that the training set has to include data from a similar environment for the model to improve the range estimates. Further, for the implemented positioning algorithm, the best methods show no effect on the positioning accuracy in the first deployment and approximately 15 % improvement in the second.