Particle filter using a new resampling approach applied to LEO satellite autonomous orbit determination with a magnetometer

Abstract

Particle filter (PF) is widely used in nonlinear and non-Gaussian systems. Resampling is one of the significant steps in PF. However, PF using conventional resampling approaches may lead to divergent solutions because of the degeneracy phenomenon or sample impoverishment associated with a multidimensional system. In this article, an efficient alternative to conventional resampling approaches, called adaptive partial systematic resampling (APSR) with Markov chain Monte Carlo move and intelligent roughening is proposed for satellite orbit determination using a magnetometer. The results of the new resampling approach are compared with conventional resampling approaches and with unscented Kalman filter (UKF) for various initial errors in position and velocity, measurement sampling periods, and measurement noises to evaluate and verify the performance of the new resampling approach. The results of the new resampling approach in all cases are significantly better than the results of conventional resampling approaches. The velocity accuracy of the orbit determination of APSR is slightly poorer than UKF for relatively small initial errors, and small Gaussian measurement noise. However, the proposed approach yields more robust and stable convergence than UKF under large initial errors, long measurement sampling period, large Gaussian measurement noise, or non-Gaussian noise.