The digital filtering technology has been widely applied in a majority of signal processing applications. For the linear systems with state-space model, Kalman filter provides optimal state estimates in the sense of minimum-meansquared errors and maximum-likelihood estimation. However, only with accurate system parameters and noise statistical properties, the estimation obtained by standard Kalman filter is the optimal state estimate. Most of time, the exact noise statistical properties could not be obtained as a priori information or even wrong statistical properties may be captured by the offline method. This may lead to a poor performance (even divergence) of Kalman filtering algorithm. In this study, a novel real-time filter, named as fast minimum norm filtering algorithm, has been proposed to deal with the case when the covariance matrices of the process and measurement noises were unknown in the linear time-invariant systems with state-space model. Tests have been performed on numerical examples to illustrate that the fast minimum norm filtering algorithm could be used to obtain acceptable precision state estimation in comparison with the standard Kalman filter for the discrete-time linear time-invariant systems.
IntroductionState estimation is a key part in most applications of model-based control techniques since feedback control is usually designed based on accurate state information in most existing control approaches, such as model predictive control and linear quadratic regulator. For example, in [1] a high-gain observer is adopted to cope with the problem of estimating partial state information for the purpose of synchronised tracking control, and it is well known that linear state observers can be designed to generate asymptotically accurate state estimates for a deterministic linear system. For stochastic linear systems with state-space model, Kalman filtering algorithm is known as a common optimal state estimating method and has been widely used in many applications, such as navigation [2], communication [3] and fault diagnosis [4]. In Kalman's pioneering work [5,6], state estimate and prediction problems were described in a different versions; he also proposed a method for the optimal solution of this general problem when system model is linear, precisely known and the statistical properties of process and measurement noises are obtained precisely in advance. However, the requirements of Kalman filtering algorithm can be seldom completely satisfied in the practical engineering systems for a variety of reasons. In order to use Kalman filtering algorithm to deal with different state estimating problems, a series of modified Kalman filter algorithms were proposed to resolve certain practical issues. Stanley F. Schmidt proposed extend Kalman filter (EKF) [7,8] algorithm to solve filtering and prediction problems of non-linear systems encountered in Apollo program in the 1960s. On the basis of the idea of Taylor expansion, EKF is designed to approximate non-linear system by linear model and then use s...