A subspace algorithm for the identification of discrete time frequency domain power spectra (Q1388108)

Subspace identification algorithms for the identification of frequency domain data have already been extensively described in the literature. These techniques are however not directly usable for the identification of frequency domain power spectra. The problem addressed in this paper is the one of fitting a linear discrete time power spectrum through given measured frequency domain power spectrum samples. The major contributions of this paper are the following: 1) The authors show how the inverse discrete Fourier transform of the given power spectrum can be expressed as a function of the system matrices. 2) A given set of system matrices does not necessarily describe a valid power spectrum. Indeed, a power spectrum has to be positive real to be physically meaningful. This problem is solved through a slight modification of the basic subspace power spectrum identification algorithm. The identified power spectrum is always positive real. 3) The practical relevance of the problem treated in this paper is illustrated with two examples which solve an acceleration sensitivity and an acoustic power spectrum modeling problem.