Non-gaussian Arma Model Identification Method Based On Higher-order Statistics

With the development of modern control theory and technology,system identifications are expanding further and abroad. Complicated systems,complicated measurement noise,and restricted conditions within wide limits are increasing interested in system identifications. Non-Gaussian,non-stationary,nonlinear and time-variant system problems have been increasing interested in system identifications. Compared with second-order statistics,higher-order statistics not only reveal information about a process,but also are blind to any kind of a Gaussian process. Higher-order statistics can deal with either colored noise,non-Gaussian,nonlinearities or nonminimum phase,whereas second-order statistics can not. So these methods based higher-order statistics are wide applicable in signal process and system identifications.This paper describes the basic theory and methods based higher-order statistics in applicability of system identification and signal process,introduces and analyses the theoretical and algorithmic results of identifying non-Gaussian ARMA process during recent years. Some linear approaches reported recently without postprocessing to FIR system identification are discussed. This paper improves the question of the promulgation of error due to estimating middle parameter frequently in other related methods,and presents a direct algorithm of estimating parameter without estimating middle coefficient. Through proper mathematic means,for example,singular value decomposition (SVD) or total-least squares solution (TLS),this algorithm smoothes the noise and improves effectively the estimation performance. Simulations show that our algorithm performs nonminimum phase FIR systems identification under colored ARMA Gaussian noises with higher efficiency and better accuracy as compared with the related algorithms in the literature.