Performance analysis of N4SID state-space system identification

This paper describes a statistical performance analysis method for multivariate subspace system identification algorithms. The analysis is based on a recently developed perturbation expansion of the singular value decomposition. We demonstrate the technique by analyzing the numerical state-space subspace system identification algorithm N4SID. An expression for the variance of the identified eigenvalues based on a finite record of input/output data is derived and verified using MATLAB simulations.