The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical and Computer Engineering University of Southern California

Technical Report USC-SIPI-192

“Moving-Average-System Identification Using High-Order Spectra: A Simulation Comparison of Four Methods”

by Wendy Huang and Jerry M. Mendel

October 1991

Four methods of identification of nonminimum phase systems with finite impulse response are compared in this project.

Via Monte Carlo simulation, the following four methods are compared:

(1) GM+T method - Giannakis and Mendel (2) Bicepstrum method - Pan and Nikias (3) MU method - Matsuoka and Ulrych (4) RG method - Rangoussi and Giannakis

Different zero locations, noise types, signal-to-noise ratios and data lengths, resulted in 48 cases. Zeros move-close to the unit circle with one located inside it and the other one located outside it. Additive noise is Gaussian, white or colored. Signals are non-Gaussian distributed.

The simulation study suggests that estimation of the IT coefficients, the bicepstrum method is the preferred method. GM+T method shows very good stability at estimating the magnitude of IT coefficients. Both MU and RG methods are stable at estimating the phase of IR coefficients, but poor at estimating the magnitude.

To download the report in PDF format click here: USC-SIPI-192.pdf (1.8Mb)