A comparison of three correlation techniques for system identification

This paper deals with the problem of identifying the system matrix (i.e. the matrix characterizing the system evolution) for time-invariant discrete linear dynamical models operating in a stochastic environment. Three approaches for solving such an identification problem are surveyed in some detail, and a comparison of effectiveness is presented including simulated results. The system identification methods discussed here present the following common aspects : they are correlation techniques which work under the quite real assumption of accessibility only over the noisy observation sequence, opposite to the practice of using both input and output data.