Terrain Referenced Navigation with State Augmentation Using Adaptive Two Stage Point Mass Filter

Abstract

In this dissertation, a novel approach by the adaptive two stage point mass filter (ATSPMF) is proposed for improving computation efficiency and estimation performance in point mass filter (PMF) based terrain referenced navigation (TRN). The inertial navigation system (INS) is provides the position, velocity and attitude of the vehicle based on dead reckoning with an inertial measurement unit (IMU) alone. INS has advantages for small short term error, high update frequency and robustness of external disturbance, but it has a fatal disadvantage that it diverges over time. To overcome this problem, INS should be aided by another sensor that provides absolute position or other information that can be used to estimate position such as global positioning system (GPS). INS aided GPS is the most widely used for integrated navigation system because it can be configured easily and estimates position, velocity, attitude and bias error of the IMU by only GPS position information as a measurement of the Kalman filter. In recently, however, GPS can be disturbed by jamming or spoofing and it may lost reliability or become unusable.

The TRN is a navigation system suitable for alternative navigation system for INS aided GPS. It uses the difference between measured terrain elevation which received by radar altimeter (RA) and barometric altimeter (BA) and terrain elevation information provided by digital elevation map (DEM) as a measurement of the TRN. Also, the TRN uses the nonlinear filter such as the PMF by using the measurement mentioned above because the nonlinearity of the measurement is severe. For improving PMF based TRN, I have proposed the novel approaches by grid support adaptation and two stage filtering.

First, I have proposed adaptive grid support algorithm for improving estimation performance and computation efficiency in PMF based TRN. In general PMF based TRN, the size of the grid support is maintained constantly. But that simple way has some disadvantages for computation burden and estimation performance. So, I have proposed new grid support adaptation method which can consider the roughness of the terrain elevation and accuracy of the measurement by using mutual information (MI) as an adaptation index. The adaptation index determines whether to increase the size of the grid support or decrease.

Second, the two stage PMF (TSPMF) is proposed for state augmentation with computation efficiency. For improving estimation performance of PMF based TRN itself, it is advantageous to set more state variables. But, the more state variables, the more computational burden exponentially. The TSPMF can provide great efficiency with state augmentation by two stages. In first stage, the nonlinear state variables are estimated by general PMF. Next, in second stage, the linear state variables are estimated by a single Kalman filter. At this time, some information that can be obtained by PMF in first stage is used in second stage for considering the correlation nonlinear and linear state variables.

In simulation results, the estimation performance and computational efficiency is improved by grid support adaptation in two dimensional state variables PMF and TRN. Also, when the state variables are augmented by three dimensions, the computation efficiency is improved by TSPMF as the estimation performance is maintained.