Design and analysis of an inertial calibration Kalman filter

Abstract

This thesis addresses the application of Kalman filtering to the in-flight calibration of an Inertial Navigation System (INS) using decentralized cascaded filtering. The system consists of a calibration filter receiving inertial reference data from an INS and radio measurements from an inertially aided Global Positioning System (GPS) receiver. The master GPS/INS and primary GPS filters receive inertial data from the same inertial reference system.

The master calibration filter is a 14 state Kalman filter derived from error models commonly used in existing aided inertial navigation systems. The master filter errors states are position, velocity. attitude, gyro bias and accelerometer bias. The GPS primary filter chosen for this study is a 12 state Kalman filter. The primary filter error states are position. velocity, attitude, clock bias and clock drift.

The study of this architecture focuses on the design of an embedded real-time Kalman filter application using standardized GPS and INS sensors not specifically designed for that purpose. The presentation of the design begins with a review of the system architecture and the operation of the sensor subsystems based on current literature. The design and implementation algorithms of the calibration filter that follows represents the core of the research performed.

Unique design features resulting from the research are the use of an INS alignment and error model to derive the initial filter covariance, the characterization of the discrete prefiltered measurement data and the design features added as a result of simulation and empirical data. The covariance initialization, in particular has shown to significantly improve the transient estimation performance of the calibration filter over more traditional diagonal covariance initializations. Other design features that account for sensor peculiarities are a dynamically computed measurement screening algorithm to detect and eliminate spurious data and situation dependent measurement variance adjustments. The combination of these features improves the robustness and reliability of the calibration filter estimates.