A statistical framework for false-fix elimination in tercom using real flight test data

Abstract

Terrain-Referenced Navigation (TRN) has been considered as a feasible alternative navigation method for aircraft operating in Global Navigation Satellite System (GNSS) denied environments. Terrain contour matching (TERCOM), a method based on the correlation between measured elevation profiles and stored digital terrain elevation data (DTED), is one of the most traditional TRN techniques. Despite its operational simplicity and proven applicability, TERCOM is inevitably susceptible to false-fix errors, where an incorrect terrain match is selected as the navigation solution. When these errors are transmitted to navigation systems, serious errors occur in situational awareness systems, and flight reliability decreases. This thesis presents a statistical framework for identifying and eliminating errors that may be encountered in TERCOM. To do this, the core working logic of TERCOM is not altered; instead, a classification is performed using features generated during matching. First, a set of information-carrying features is identified, and then the candidate navigation solution is classified according to its relationship with this information. This method ensures that if a candidate navigation solution is considered unreliable, it is not transmitted to the navigation system. This study is based on synthetic data generated from real Flight Test Instrumentation (FTI) recordings of the Hürjet, an advanced jet training aircraft. These recordings allow for the production of repeatable data while preserving realistic dynamics. A total of 780 different trajectories were generated. For each trajectory, six different metrics were examined: mean absolute difference (MAD), mean squared difference (MSD), normalized cross-correlation (NCC), dynamic time warping (DTW), mutual information (MI), and roughness-oriented comparison. The tests revealed that MAD and DTW have complementary characteristics. To assess the reliability of the TERCOM solution, a total of forty-five candidate features were first extracted in three different contexts: trajectory-based, terrain map-based, and score map-based. These contexts respectively encompassed the characteristics of the altitude measurement sequence, the characteristics of the terrain map, and the characteristics of the values generated during matching. Then, feature elimination was performed using correlation analysis, mutual information ranking, receiver operating characteristic evaluation, and minimum redundancy–maximum relevance criteria methods. As a result of the elimination process, the minima-ratio, the ratio of the two lowest scores in the score map, emerged as the most distinctive feature. Roughness and roughness difference features were also added as auxiliary features. A confidence-based elimination strategy was developed based on these indicators. The framework chooses between MAD and DTW position solutions according to their minima ratios. Then, utilizes threshold-based decisions to mitigate unreliable fixes. The experimental results indicate that the proposed framework maintains overall navigation accuracy while achieving a reduction in false-fix occurrences by 44\%. The validation with actual flight test data provides the practical evidence of the approach. This thesis illustrates that a statistical reliability assessment can significantly improve the robustness of TERCOM while causing minimal computational costs. The proposed framework offers a practical safeguard for TRN systems, enhancing operational safety in GNSS-denied environments.