On a Method for Estimating Power Spectra

In many engineering problems, the power spectra of random signals have to be measured. In this paper, a method for estimating the unknown parameters in the power spectral density functions of discrete random processes is presented. The spectral density functions are represented by a ratio of two real polynomials in z. It is assumed that only a finite record of the random process is observed and no knowledge about its probability distribution is required. An estimation procedure is developed. It is shown that the denominator parameter estimates can be first obtained. Then the numerator parameters are estimated by filtering the poles from the spectrum. A four parameter example is given to demonstrate the method.