Machine Learning Enhanced Signal Quality Assessment Leveraged with GDOP for GNSS/INS Fusion

It is well-recognised that the observed satellite number is usually superabundant yielding extensive computation consumption of processing redundant observations or measurements from unnecessary satellites, hereby the selection of the most suitable satellite combination in Receiver Autonomous Integrity Monitoring (RAIM) becomes prominent for preventing the performance degradation resulting from pseudorange errors. This work proposes an enhanced Global Navigation Satellite System (GNSS)/Inertial Navigation System (INS) navigation method with the principle of compensating geometry-related performance degradation through ranking satellites from Machine Learning (ML) outcomes. An ML-assisted satellite signal quality assessment scheme is developed for Fault Detection and Exclusion (FDE) in RAIM with weights leveraged between satellite geometry information and signal quality to incorporate factors of pseudorange predictions and Geometric Dilution of Precision (GDOP). The pseudorange prediction relies on assessing signal nature represented by Carrier-to-noise (C/N0) ratio, satellite navigation information of satellite elevation angle, and estimated receiver position. Poorly ranked signals are excluded from the navigation solution computation after tuning of weights. With respect to performing evaluations under controlled environments, the proposition is verified by Hardware-In-the-Loop (HIL) testing using Spirent's GSS 7000 GNSS constellation simulator and U-Blox ZED F9P GNSS receiver. Through performance analysis over an urban scenario against a tightly coupled Extended Kalman Filter (EKF) incorporating the proposed GNSS signals ranking schemes, it is found that the proposed combination of GDOP and signal quality assessment aided by ML presents outstanding improvement in terms of positioning accuracy.