Real-time optimum calibration of large sensor systems by Kalman filtering

Abstract

It is noted that a large hybrid position location or navigation system would contain far too many state parameters if all its calibration drifts were to be optimally estimated, because the Kalman recursion formulae would require an inversion of an excessively large matrix. This problem has been overcome by using an analytical sparse-matrix inversion method. The new statistical technique is a combination of Kalman filtering and an optimum calibration algorithm. The initialization of the Kalman calibration filtering was implemented on the Cray XMP-48 supercomputer system of the European Centre for Medium Range Weather Forecasts (ECMWF) in Reading, UK. Radiance measurements from the polar-orbiting NOAA satellites and weather reports from the worldwide radiosonde network were used for limited experimentation with encouraging results.<>