Identification of noncausal nonminimum phase AR models using higher-order statistics

In this paper, we address the problem of estimating the parameters of a noncausal autoregressive (AR) signal from estimates of the higher-order cumulants of noisy observations. The proposed family of techniques uses both 3rd-order and 4th-order cumulants of the observed output data. Consequently, at low SNR, they provide superior performance to methods based on autocorrelations. The measurement noise is assumed to be Gaussian and may be colored. The AR model parameters here are directly related to the solution of a generalized eigenproblem. The performance is illustrated by means of simulation examples.