Simulation and prototyping of integration methods between GNSS, INS, and signals-of-opportunity sources
University of New Brunswick
Seamless positioning and navigation are a recurring academic and industrial challenge due to single sensor solutions' limitations. Modern technologies allow developers to work with multiple sensors and signals in parallel, solving systems of equations that a decade ago would not be handled in real-time. In this context, this research aims to explore the plausible integration between absolute GNSS positioning, inertial systems, and two signal-of-opportunity candidates: Wi-Fi and 5G mmWave signals. The integration of different signals and sensors is explored in a low computational power filter context, such as the extended Kalman filter (EKF), particle filter (PF), and one of the most efficient machine learning algorithms: the support vector machine (SVM). All three methods have been used for a long time in the most different applications. In this thesis, their loosely and tightly integration will be explored to design, implement, and develop novel positioning algorithms using the strengths of different sensors and filters. Currently, literature is not settled on this integration's potential, therefore, creating a need for more comprehensive developments and tests in this scope. Furthermore, this thesis presents a framework for simulation and validation of sensor integration algorithms and novel contributions for multi-sensor navigation, including integration at the post-filter constraint level and machine-learning aided filtering. The methods proposed in this thesis were implemented with real-time application considerations and shown to be beneficial in challenging urban areas scenarios. Improvements in accuracy ranging from 90.1% in horizontal accuracy compared to the GNSS-only solution and 36.4% when compared to GNSS and Inertial System (INS) solution are achieved in real-world data. Integrity metrics are also shown to be improved between 1 and 14% when utilizing signal-of-opportunity information. With simulated mmWave ranging methods, the GNSS solution is shown to be significantly more robust, improving accuracy by up to 69.4% in the horizontal axis and an improved error and uncertainty estimation with values up to 12% better. The results and analysis are presented considering the scientific and industrial standards, leading to the conclusion that cost-effective methods built with off-the-shelf equipment can improve the current standard in sensor integration for navigation.