Identification of Signal and Noise Components in Spacecraft Neutral Particle Data Using a Bi-Level Mixture Model

Abstract

We propose a method to classify the data from the Mars Express spacecraft using a bi-level mixture model. The spacecraft data contain various kind of signals and noises, and the causes of such signals and noises are unknown. In order to identify various kind of signals and noises and to clarify the features of them, we classify the data into multiple component using a mixture model. The mixture model used here is a bi-level model which is suitable for the features of the data used in this study. In this bi-level model, the parameters for the upper level model and the lower level models can be estimated separately. This allows us to improve the computational efficiency. The estimation for both of the upper and lower level models is carried out using the EM algorithm. In order to determine the number of component, we use a stepwise approach; that is, we start with a few components, and increase the number of components one by one. The increase is terminated if the number of components becomes sufficient in terms of the Bayesian information criterion (BIC). Finally, we map the result of the classification to the space which represents the measurement conditions. This mapping helps us resolve the cause of each component.