ESA GNC Conference Papers Repository

The classical approach to sensor data fusion is via Kalman filtering. The filter is optimal in the sense that the estimation error variance is minimized, thus the integral over the power spectral density of the estimation error. It does not matter in which frequency range the power is located. On the other hand, performance and knowledge error indices as defined in the ESA Pointing Error Engineering Handbook are based on weighting the estimation error in the frequency domain. In that sense, a good estimator has to minimize the error in a certain frequency range, e.g. high frequencies for the Relative Knowledge Error. The paper presents an H-infinity based systematic approach for sensor data fusion that can deal with both coloured noise in measurements and frequency-based specifications on the estimation error. A number of test cases confirm the feasibility of this approach compared to the conventional "hand-tuning" of a Kalman filter.