Coupling between nonlinear estimation and dynamic sensor tasking applied to satellite tracking.

摘要

The tracking of Earth orbiting objects has been a topic of growing concern, due to the fact that the amount of man-made orbital debris, and the number of active and inactive space objects have been steadily increasing over the past several decades. Space Situational Awareness (SSA) is concerned with the tracking, detection, and cataloging of numerous space objects using relatively few ground- and space-based sensors known as the Space Surveillance Network (SSN). While these sensors provide observations of object characteristics (range, azimuth, elevation, etc), the large number of objects compared to the limited sensors available to track them results in measurements occurring infrequently. These potentially long periods of either inability to make observations (due to line-of-sight access) or unavailability of sensors (due to scheduling constraints) necessitates the need to intelligently determine which objects should be observed and which should be ignored at various times, a process known as sensor tasking or sensor network management.;In order to make these tasking decisions, it is necessary to create some form of utility metric to determine which sensors should observe which objects at a particular instant of time. This dissertation examines the use of utility metrics from two forms of expected information gain for each object-sensor pair as well as the approximated stability of the estimation errors in order to work towards a tasking strategy. The information theoretic approaches use the calculation of Fisher information gain (FIG), an estimate of the upper bound of information present in an unbiased estimator, and Shannon information gain (SIG), a measure of information gained about the particular state. Both of these methods are considered myopic or greedy in nature, due to the fact that they calculate only information gained over one simulation time step. FIG has been studied previously as a potential sensor tasking metric, and has even been investigated in applications to SSA, while SIG has been suggested as a possible sensor tasking metric, but has yet to be investigated when applied to sensor tasking in the SSA problem. The stability approach reflects a new type of metric referred to in these studies as largest Lyapunov exponent estimation (LLE), and has yet to be studied as a sensor tasking utility metric.;The process of evaluating these utility metrics is intrinsically tied in with state and uncertainty estimates provided by a nonlinear filter. That is, each utility metric requires estimates provided by the filters in order to be calculated, creating a coupling effect between the estimation and tasking components of the satellite tracking problem. In order to investigate this, three candidate nonlinear estimators, an extended Kalman filter (EKF), an unscented Kalman filter (UKF) and a recently introduced adaptive entropy-based Gaussian-mixture information synthesis (AEGIS) filter are tested. The primary difference in these filters is their ability to approximate system nonlinearities in their application, with previous work showing that the AEGIS filter performs the best in this regard, while the EKF performs the worst. While many studies have shown how an EKF and UKF differ in estimation performance when applied to orbit determination problems, little work has been done to investigate the AEGIS filter in these regards.;While much recent research has been conducted investigating specific methods of either sensor tasking or nonlinear estimation, there is yet to be any studies which investigate the coupling of the two, as it is related to overall tracking performance. The investigation of this coupling demonstrates that the use of more accurate filters leads to better overall estimates, not only due to the advantages within the estimation methods, but also from the improvement in tasking decisions due to selection of these estimators.