Abstract. This paper presents a brief synthesis and useful performance analysis of different attitude filtering algorithms (attitude determination algorithms, attitude estimation algorithms, and nonlinear observers) applied to Low Earth Orbit Satellite in terms of accuracy, convergence time, amount of memory, and computation time. This latter is calculated in two ways, using a personal computer and also using On-board computer 750 (OBC 750) that is being used in many SSTL Earth observation missions. The use of this comparative study could be an aided design tool to the designer to choose from an attitude determination or attitude estimation or attitude observer algorithms. The simulation results clearly indicate that the nonlinear Observer is the more logical choice.
IntroductionThe objective of the attitude filtering algorithms is to calculate the satellite orientation, using measurement data of satellite sensors such as star sensor, sun sensor, infrared horizon sensor, gyroscope and the magnetometer, etc. Among these, Sun sensors or star trackers are very used in small satellites because of the limitations of small satellite missions in terms of power and structure [1]. There are many algorithms for determining the attitude of the satellite. These algorithms are divided into three families: attitude determination algorithms (such as TRIAD, Q-Method, Quest, Least Squares, etc.), attitude estimation algorithms (Kalman Filter, H-infinity Filter, Particle Filter, etc.), and nonlinear observers (Sliding Mode Observer, Luenberger observer, Backstepping observer, etc.). Each of these algorithms is able to determinate the attitude, and it has an advantages and disadvantages. The attitude determination algorithms have an advantage of low execution time. This is due to its independence of dynamic models, but they are able to estimate only the attitude but not other system states such as angular rate, biases and noise measurement (not presented here). So, what do we do if the angular rates are also calculated from angular sensor measurements? Because we can't use the attitude determination algorithms to estimate this system state, to solve this problem we use the attitude estimation algorithms. The attitude estimation algorithms are characterized by good performances in term of accuracy, but they have a high time consuming because its structure uses the dynamic models. This point presents a disadvantage in comparison with attitude determination algorithms. In addition, the attitude estimation algorithms are able to estimate the attitude and the angular rate using the sensors installed in the
