Author: Danilo Korze
Mentor: Asst. Prof. Dr. Damjan Zazula
Co mentor: Asst. Prof. Dr. Gorazd Lesnjak
Date: Dec. 14, 1992

Higher-Order Cumulants and Multiple Cepstra in One-Dimensional Signal Prosessing

Keywords: digital signal processing, system identification, polyspectra, higher-order statistics, higher-order cumulants, differential cepstrum, complex cepstrum, differential bicepstrum, complex bicepstrum, ARMA models, impulse response, transfer function, Fourier transform, white noise, deconvolution.

Abstract: An overview of possibilities for signal processing with the help of higher-order statistics ( polispectra, higher-order cumulants and multiple cepstra) is given. The higher-order cumulants are described in more detail. The methods for system identification from only output measurements based on higher-order cumulants and diferential cepstrum are implied. The possibilities for variance reduction are described. Results of two diferent sytem identification procedures are presented. With the C (q, k) method, the effect of variance reduction is studied. The time complexity of methods mentioned is unsuitable for the real-time applications with general-purpose computer architectures. With the help of cumulant-based methods, better results are achieved as it was the case with the conventional techniques.