Satellite Attitude Determination Using Magnetometer Data Only

Abstract

On March 8, 2007, the U.S. Air Force Academy launched FalconSAT 3, a complex small satellite with three plasma-researching payloads supporting Department of Defense funded science missions. This is the first three-axis stabilized satellite designed, tested, and built by cadets. FalconSAT 3 is equipped with two types of satellite attitude sensors: magnetometers and sun sensors. However, a software issue inhibits the use of sun sensors, so development of a new method of attitude determination was required. The three-axis magnetometers would need to provide the means to accurately depict the spacecraft’s orientation at any given time. Because of the ambiguity involved with using a single form of spacecraft attitude information, Extended Kalman Filtering became a viable solution that would combine insufficient data with a mathematical model of satellite motion to achieve accurate attitude estimation. First, a single-axis Kalman Filter was created for determining the yaw axis, assuming roll and pitch remain zero due to gravity-gradient stabilization. It was found that a constant gain filter can sufficiently provide pointing knowledge, depending on the user’s needs. This yaw-estimator also initializes a starting point for a three-axis estimator. Generating a six-state, three-axis control Extended Kalman Filter for attitude determination became the primary focus of research because the FalconSAT 3 payloads require a level of attitude determination accuracy greater then the magnetometers can provide alone. Looking to the future, modification of this three-axis estimator accounts for a pitch wheel considered for FalconSAT 5, a satellite that the U.S. Air Force Academy proposes to launch in December 2009. These three Kalman Filters were developed in England at Surrey Space Centre and were tested and tuned with magnetometer data of a Surrey Satellite Technology Limited (SSTL) satellite called UoSAT. Comparing the attitude estimation results to the truth model of UoSAT shows that the Kalman Filters converge to within one degree of accuracy. This paper details the theory behind these Kalman Filters in addition to their tuning and results.