Parameter-free structural modeling: a contribution to the solution of the separation of highly correlated AR-signals

This paper develops the concepts and properties of composite parameter structural (CPS) modeling, and shows how such properties can be exploited for the separation of very highly correlated autoregressive signals. A CPS model recently developed and used to represent a signal of a given structure (given order of an AR model) but of unknown, or partially unknown, parameters, is investigated. The main feature of the described CPS model is the utilization in its design of almost ideal null filters, resulting in low noise sensitivity. The performance of the proposed algorithms is analyzed using computer simulations