Self-tuning weighted measurement fusion Kalman filtering algorithm

Abstract

For the multisensor linear discrete system with correlated noises and same measurement matrix, the self-tuning weighted measurement fusion Kalman filtering algorithm is presented when the model parameters and noise variances are all unknown. It can handle the self-tuning fused Kalman filtering, smoothing, and prediction problem and the input white noise deconvolution estimation problem. By the dynamic variance error system analysis (DVESA) method, it is proved that the solution of the self-tuning Riccati equation converges to the solution of the steady-state Riccati equation. Based on the convergence of the self-tuning Riccati equation, the convergence of the proposed self-tuning weighted measurement fusion Kalman estimator is proved. So it has asymptotic global optimality. Applying to the multi-channel autoregressive moving average (ARMA) signal with sensor bias, the corresponding self-tuning weighted measurement fusion Kalman estimator of the signal is also presented, where the estimates of unknown model parameters and noise variances are obtained by the multi-dimension recursive extended least squares (RELS) algorithm, the correlation method and the Gevers–Wouters algorithm with a dead band. One simulation example shows the effectiveness.