IV Subspace System Identification of Noisy Input-Output Systems

Abstract

In this paper we present a novel instrumental variable subspace method which estimates the system orders as well as the system parameters of a system with a rational transfer function and colored noise on both input and output signals. The method is an intuitive and efficient geometrical method that only uses linear projections, no singular value or eigen-decompositions (SVD’s and ED’s) are needed. The method estimates the autoregressive (AR) and the moving average (MA) parts (order + parameters) seperately using the same type of projections. This allows the method to be efficiently implemented on a parallel computer. The method handles more general noise situations than most of the newly developed Subspace based State Space System Identification algorithms, is more efficient and estimates the transfer function parameters rather than the state space system matrices. The method is shown to perform well even at signal-to-noise ratios (SNR’s) as low as 10 dB.