Research, The Identification Method Of Multi-sensor System In A Noisy Environment

Kalman filtering of the signals often occurs in fields of communication and signal processing etc, but the precondition of its application is that the model parameters and the noise statistics are assumed to be completely known. However, we often encounter the problems of multisensor system information fusion filtering with the unknown model parameters and the noise statistics in the practical applications. In fields of adaptive control and self-tuning control, we often encounter system control problems with the unknown model parameters and noise statistics. In order to solve these problems, we must solve the identification problem of the unknown model parameters and noise statistics. Particularly, the multisensor system identification problem in noisy environment has important theoretical and application significance.For the one-dimensional autoregressive (AR) model with the ARMA coloured observation noise, the recursive instrumental variable (RIV) algorithm of AR parameter estimation is presented. For the multidimensional autoregressive (AR) model with the moving converage (MA) coloured observation noise, the multidimensional and multiple recursive instrumental variable (MRIV) algorithms of AR parameter estimation are presented. The noise variance estimators can be obtained based on the correlation function method.For the multivariable autoregressive (AR) model with white or coloured observation noise, the multivariable bias compensated recursive least-squares (MBCRLS) algorithm is presented, and its strong convergence is rigorously proved by the dynamic error system analysis (DESA) method.For the multisensor single channel or multi-channel autoregressive moving average(ARMA) signals with white or coloured measurement noises, common disturbance noise, and with unknown model parameters and noise variances, a multi-stage identification method is presented. In the first stage, using the RIV algorithm or recursive extended least-squares (RELS) method, the on-line information fusion estimators of the unknown AR parameters are obtained by taking the average of local model parameter estimators. In the second stage, by using the correlation function method, the on-line information fusion estimators of the unknown noise variances are obtained by taking the average of local noise variance estimators. In the third stage, by using the correlation function method, Gevers-Wouters algorithm with dead zone and least-squares (LS) method, the information fusion parameter estimators of MA parameter are obtained.Many simulation examples show the effectiveness of the proposed algorithms and methods.