Identification of AR parameters at a very low SNR using estimated spectral distribution in DCT domain

Author

Connie, A.T. ; Ferdousi, F. ; Sharmin, M. ; Khan, M.R.

Author_Institution

Dept. of Electr. & Electron. Eng., Bangladesh Univ. of Eng. & Technol., Dhaka, Bangladesh

Volume

153

Issue

2

fYear

2006

fDate

4/6/2006 12:00:00 AM

Firstpage

95

Lastpage

100

Abstract

A simple and efficient method for system identification even at a very low signal-to-noise ratio (SNR) is presented. At an SNR as low as -7.5 dB, noise dominates the spectrum and system poles are almost lost in the profound noise. In the proposed method, an enhanced spectrum is estimated in the discrete cosine transform (DCT) domain using the least squares curve-fitting technique. The system modes that were previously indistinguishable become prominent in the enhanced spectrum. The system order is then overestimated using least squares higher order Yule-Walker (LSHOYW) equations to obtain better accuracy. The poles having higher strength in the autocorrelation domain are then identified as system poles.

Keywords

autoregressive processes; correlation theory; curve fitting; discrete cosine transforms; least squares approximations; parameter estimation; DCT; SNR; autocorrelation domain; autoregressive parameters identification; discrete cosine transform; least squares curve-fitting technique; least squares higher order Yule-Walker equations; signal-to-noise ratio; spectral distribution estimation;

fLanguage

English

Journal_Title

Vision, Image and Signal Processing, IEE Proceedings -

Publisher

iet

ISSN

1350-245X

Type

jour

DOI

10.1049/ip-vis:20045254

Filename

1610524