Kalman filtering state noise adaptive estimation

In the presence of estimates divergence due to errors in system dynamics modelling and to algorithm numerical ill-conditioning, Kalman filter usefulness is seriously impaired. A procedure to estimate the state noise level adaptively and in real time in order to represent the errors and to prevent divergence is proposed. The adequate noise level is adjusted by using a criterion of statistical consistency between estimated errors and observation residues. Results of preliminary tests, in the case of artificial satellite orbit determination in real time, are presented and show the procedure capability of avoiding divergence and of giving results of acceptable accuracy.