Order-recursive blind identification of linear models using mixed cumulants

The problem of determining the AR order and parameters of a nonminimum phas e ARMA model from observations of the system output is considered. The mode l is driven by a sequence of random variables which is assumed unobservable . A novel identification algorithm based on the second- and third-order cum ulants of the output sequences is introduced. It performs order-recursively by minimising a well defined cost function. Strong convergence and consist ency of the algorithm are proved and the weight of the cost function is bal anced between the second-order and the third-order cumulants of output sequ ences. The influence of the weight on the estimation accuracy is also evalu ated. Theoretical analyses and numerical simulations show that the proposed algorithm is satisfactory for both order and parameter identification of a n AR model which is subordinate to a nonminimum phase ARMA model.