Real-Time Tuning Unscented Kalman Filter for a Redundant Attitude Estimator in Microsatellites

摘要

This paper deals with microsatellite attitude determination systems which are a combination of a main estimator with high-power, high-precision sensors for higher accuracy estimation and a redundant estimator with low-power, lower-precision sensors for backup. Measurement data from all sensors in the redundant estimator are fused by the unscented Kalman filter to provide estimated attitude and the gyro bias values. Besides the accuracy of attitude sensors, the accuracy of this estimator depends largely on the selection of the process and measurement noise covariance matrices. In this paper, a novel real-time tuning unscented Kalman filter for redundant attitude estimator is introduced to tune these matrices efficiently in each filter step. The tuning process uses the estimated attitude of the main estimator as an independent truth reference data to calculate the cost function which is minimized by a downhill simplex algorithm. In the scheme developed in this paper a fine-tuning process is used, which results in faster convergence speed and higher estimated accuracy of the redundant estimator. Another important feature of the developed filter is that a flexibly estimated accuracy and system power consumption can be archived by choosing the duration and repeat frequency of turn-on time of the main estimator.