Multiple neural network filtering for in-flight calibration of satellite measurements

A set of nonlinear backpropagation neural nets were used to make an in-flight calibration of the three-dimensional ion composition spectrometer (TICS), on the FREJA satellite. TICS measures the ion composition, (the ion mass spectra) in the three dimensional velocity space. The ion detection unit on TICS uses a circular micro channel plate (MCP) divided into 32 angular sectors to amplify each ion count. However all the MCP sectors have different measuring sensitivity. The sector sensitivity was estimated in calibrations before launch of FREJA, but once in space, various measurements indicated that this calibration needed to be renewed. The in-flight calibration was solved using a set of nonlinear neural network filters to normalize all sectors compared to a selected reference sector and then determine a new set of (both relative and absolute) sector sensitivity coefficients. The result shows that the backpropagation neural nets used as nonlinear filters could normalize each sector with an accuracy better than 10%