A New Subspace Identification Method for Open and Closed Loop Data

Subspace methods have emerged as useful toolsfor the identification oflinear time invariant discrete time systems. Most of the methods havebeen developed for the open loop case to avoid difficulties with datacorrelations due to the feedback.This paper extends some recent ideas for developing subspace methodsthat can perform well on data collected both in open and closed loopconditions. Here, a method that aims atminimizing the prediction errors in several approximate steps isproposed. The steps involve using constrained least squares estimationon models with different degrees of structure such as block-toeplitz,and reduced rank matrices.The statistical estimation performance of the method is shown to becompetitive to existing subspace methods ina simulation example.