Improved methods for the blind system identification using higher order statistics

It is demonstrated by a detailed analysis of blind system identification that under specific system configurations, a recently published least-squares algorithm shows a poor convergence behavior, especially if the system order is overdetermined. To overcome these problems, a supplementary condition is introduced that guarantees proper convergence in most cases. An alternative approach for the blind identification of mixed-phase systems, the so-called cumulant zero-matching method, is presented. In this approach, the solution of a set of nonlinear equations, which is necessary in the least-squares method, is replaced by the calculation of zeros of polynomials. The main advantage over the least-squares solution is that overdetermination of the system order is rather harmless, since it only results in additional zeros in the origin of the z-plane. The different methods for system identification presented are illustrated by simulation results