ENHANCED KINEMATIC POSITIONING METHODS BY SHAPING FILTER AUGMENTATION
Katrin Ramm · 2006
Kalman filtering is an important tool for positioning for vehicle navigation and for location based services. This paper, therefore, deals with methods to enhance a standard Kalman filter approach for kinematic positioning. Any modelling of Kalman filter approaches requires white measurement and process noise. GPS positions as input quantities do not meet these requirements due to existence of autocorrelation. Red noise has to be assumed due to slowly changing measurement deviations caused by tropospheric propagation delay, for example. In a first step, the filter respectively the state vector is augmented by a shaping filter to consider red noise of kinematic GPS positions. The appropriate autocorrelation function is a bell-shaped curve and its characteristic parameter - the attenuation factor - is determined empirically. So data of some test runs (approx. 650 km, driven with the measuring vehicle of the institute (MOdular Positioning SYstem (MOPSY), see (21))) are evaluated by time series analysis. The resulting autocorrelation functions are approximated by regression analysis to derive a functional description. In a next step, the attenuation factor is supposed to be unknown because of uncertainties in determining it accurately. Therefore a second approach is proposed to estimate the attenuation factor by an adaptive estimation for the augmented Kalman filter. These approaches are evaluated with respect to the standard one. Both simulated data to conduct variance-based sensitivity analysis and real data are used. 1. Motivation One research focus of the institute for applications of geodesy to engineering (IAGB) is modelling of vehicle motion by Kalman filtering. Several modifications for improvements of position estimation are developed. This paper focuses on correct stochastic modelling especially of the DGPS kinematic measurements. Improvements are expected for the quality of the different filter tests, significantly influencing the filter performance, and for more realistic accuracy estimations. Sensitivity analysis enables the investigations of dependencies between output and input uncertainties, thus giving hints for further improvements.