Content area
The Global Positioning System (GPS) provides user-friendly and highly accurate navigation capabilities worldwide. Since its inception, its range of applications has expanded significantly, encompassing areas such as remote delivery, ride-sharing, and autonomous vehicles. However, the positioning accuracy provided by the GPS heavily deteriorates in urban canyons. Recent research has leveraged deep learning models trained on large GPS and environmental information datasets to enhance pseudorange and position estimates, mitigating the urban canyon effect. Alternatively, signals of opportunity (SoOP) from low Earth orbit (LEO) communication satellites offer a promising alternative for localization, particularly in GPS-denied environments.
In this dissertation, we propose algorithms to perform navigation in GPS-poor and GPS-denied environments with the use of deep learning and LEO satellites. First, we introduce an algorithm that leverages the multiple input multiple output (MIMO) capabilities of LEO satellites and vehicular users for navigation in GPS-denied environments. We estimate the direction of arrival (DoA) and the Doppler shift from the LEO satellite communication signals and fuse them with the measurements from accelerometer and gyroscope installed in the user vehicle to perform position tracking after the loss of GPS signal. In a second approach, we propose to introduce corrected GPS measurements into the fusion system. To develop this strategy, we need to create a channel data generation framework for satellite-based systems using ray tracing simulations. We develop methods to generate the user and satellites’ trajectories and also to incorporate the effects of the atmosphere on the channel generation. We also create a channel dataset for LEO and GPS satellite systems serving a user vehicle traversing in deep urban canyons. The final step is a navigation algorithm for fusing the measurements from GPS, LEO and inertial measurement unit (IMU) sensors. Our proposed method first exploits a deep network to refine the GPS position estimates and then uses a Kalman filter to fuse them with LEO and IMU measurements. Our results show significant improvements in terms of position accuracy as compared to the baseline algorithms. This work demonstrates that deep learning and LEO satellites can effectively enhance the navigation capabilities of GPS in urban canyon environments.