CRLB Analysis for a Robust TRN Based on a Combination of RNN and PF

Abstract

Terrain-referenced navigation (TRN) is a technology that estimates the position of an aircraft by comparing the terrain elevation on a digital elevation model (DEM) and the altitude measured by an altimeter. The particle filter (PF)-based TRN has been widely used for unmanned aerial vehicles (UAVs) operating at a high altitude. Even though TRN performs well in rough and unique terrains, its performance degrades on flat and repetitive terrain. For this study, a robust PF-based TRN was designed, which uses a recurrent neural network (RNN)-based deep learning method to function on flat and repetitive terrains. Noise covariances and the measurement model of the PF were also trained. An RNN is an artificial neural network that recognizes patterns from a time-series. Due to the difficulty in theoretically verifying that the model generated by the deep learning method was designed optimally, we performed a Cramer–Rao lower bound (CRLB) analysis to evaluate how close the proposed method was to the optimal design. We also used a perturbation error model of an interferometric radar altimeter (IRA) as the measurement noise covariance needed for an accurate CRLB analysis.