On identifiability, maximum-likelihood, and novel HOS based criteria

Author

Giannakis, Georgios

Author_Institution

Virginia Univ., Charlottesville, VA, USA

fYear

1990

fDate

10-12 Oct. 1990

Firstpage

217

Lastpage

221

Abstract

Considers estimation and classification problems for a stretch of stationary data containing a non-Gaussian linear process and additive Gaussian noise of unknown covariance (AGN/UC). To allow general noncausal and nonminimum phase (NC/NMP) ARMA models, and develop estimation and classification schemes which are immune to AGN/UC higher-order statistics (HOS) are resorted to. Time-domain optimality criteria are discussed which employ a finite set of sample cumulant lags, while the frequency-domain criteria involve sample polyspectral lags.<>

Keywords

frequency-domain analysis; parameter estimation; random noise; spectral analysis; statistical analysis; time-domain analysis; additive Gaussian noise; classification problems; frequency-domain criteria; higher-order statistics; identifiability; maximum-likelihood approach; nonGaussian linear process; noncausal nonminimum phase ARMA models; parameter estimation; sample cumulant lags; sample polyspectral lags; signal processing; stationary data; time domain optimality criteria; unknown covariance; Additive noise; Gaussian noise; Higher order statistics; Image processing; Maximum likelihood estimation; Pattern recognition; Signal processing; Testing; Time domain analysis; Vectors;

fLanguage

English

Publisher

ieee

Conference_Titel

Spectrum Estimation and Modeling, 1990., Fifth ASSP Workshop on

Conference_Location

Rochester, NY, USA

Type

conf

DOI

10.1109/SPECT.1990.205578

Filename

205578