Optimal Cepstrum Smoothing
Author
Summary, in English
Abstract in Undetermined
The cepstrum of a random process has proven to be a useful tool in a wide range of applications. The common cepstrum estimator based on the periodogram suffers from large variance, and, to a smaller degree, from bias. The variance can be reduced by smoothing. However, the smoothing may be performed in four different domains: the covariance, the spectral, the log-spectral, and the cepstral domain. We present the mean square error (MSE) optimal smoothing kernels in each domain for estimation of the cepstrum. The lower MSE bound of each of the four families of estimators are compared. We also demonstrate how the four MSE optimal estimators differ in robustness.
The cepstrum of a random process has proven to be a useful tool in a wide range of applications. The common cepstrum estimator based on the periodogram suffers from large variance, and, to a smaller degree, from bias. The variance can be reduced by smoothing. However, the smoothing may be performed in four different domains: the covariance, the spectral, the log-spectral, and the cepstral domain. We present the mean square error (MSE) optimal smoothing kernels in each domain for estimation of the cepstrum. The lower MSE bound of each of the four families of estimators are compared. We also demonstrate how the four MSE optimal estimators differ in robustness.
Department/s
- Mathematical Statistics
- Statistical Signal Processing Group
Publishing year
2012
Language
English
Pages
1290-1301
Publication/Series
Signal Processing
Volume
92
Issue
5
Document type
Journal article
Publisher
Elsevier
Topic
- Probability Theory and Statistics
Keywords
- Cepstrum
- Smoothing
Status
Published
Research group
- Statistical Signal Processing
- Stochastics in Medicine
- Statistical Signal Processing Group
ISBN/ISSN/Other
- ISSN: 0165-1684