Blind Identification Using Higher Order Statistics and Eigendecomposition

抄録

In this paper we develop a new approach to identify noncausal AR models driven by non-Gaussian i. i. d input. Under a few moderate assumptions, we derive the necessary and sufficient conditions of rebuilding the parameters of the AR models from 2nd and higher order statistics. We show that the AR model parameters are directly related to the solution of an eigenproblem. Based on the result, we present a method of AR model identification, applying eigenvector computation. The unique solution of AR parameters is guaranteed up to sign and linear phase ambiguity. Model order determination is not crucial in the method. If the order is overestimated, several equivalent AR models with different linear phases will be obtained.