Self-tuning distributed measurement fusion Kalman filter

Abstract

Based on weighted least squares method(WLS), a equivalent fusion measurement equation is obtained for the multisensor linear discrete stochastic time-invariant system with unknown noise statistics and the measurement matrices having the same right factor. Using the modern time series analysis method, based on on-line identification of the moving average(MA) innovation model parameters, unknown noise variances can on-line be estimated, and a self-tuning weighted measurement fusion Kalman filter is presented. Under the assumptions that the parameter estimation of the MA innovation model is consistent and the measurement data is bounded. It is proved that self-tuning Kalman filter converges to globally optimal fusion Kalman filter with known noise statistics, so that it has asymptotic global optimality. A simulation example for a tracking system with 4-sensor shows its effectiveness.