TY - GEN
T1 - Frequency-domain characterization of Singular Spectrum Analysis eigenvectors
AU - Leles, Michel C.R.
AU - Cardoso, Adriano S.V.
AU - Moreira, Mariana G.
AU - Guimaraes, Homero N.
AU - Silva, Cristiano M.
AU - Pitsillides, Andreas
N1 - Publisher Copyright:
© 2016 IEEE.
PY - 2017/3/23
Y1 - 2017/3/23
N2 - Singular Spectrum Analysis (SSA) is a nonparametric approach used to decompose a time series into meaningful components, related to trends, oscillations and noise. SSA can be seen as a spectral decomposition, where each term is related to an eigenvector derived from the trajectory matrix. In this context the eigenvectors can be viewed as eigenfilters. The frequency domain interpretation of SSA is a relatively recent subject. Although the analytic solution for the frequency-response of eigenfilters is already known, the periodogram is often applied for their frequency characterization. This paper presents a comparison of these methods, applied to eigenfilters' frequency characterization for time series components identification. To perform this evaluation, several tests were carried out, in both a synthetic and real data time series. In every situations the eigenfilters analytic frequency response method provided better results compared to the periodogram in terms of frequency estimates as well as their dispersion and sensitivity to variations in the SSA algorithm parameter.
AB - Singular Spectrum Analysis (SSA) is a nonparametric approach used to decompose a time series into meaningful components, related to trends, oscillations and noise. SSA can be seen as a spectral decomposition, where each term is related to an eigenvector derived from the trajectory matrix. In this context the eigenvectors can be viewed as eigenfilters. The frequency domain interpretation of SSA is a relatively recent subject. Although the analytic solution for the frequency-response of eigenfilters is already known, the periodogram is often applied for their frequency characterization. This paper presents a comparison of these methods, applied to eigenfilters' frequency characterization for time series components identification. To perform this evaluation, several tests were carried out, in both a synthetic and real data time series. In every situations the eigenfilters analytic frequency response method provided better results compared to the periodogram in terms of frequency estimates as well as their dispersion and sensitivity to variations in the SSA algorithm parameter.
UR - http://www.scopus.com/inward/record.url?scp=85017581369&partnerID=8YFLogxK
U2 - 10.1109/ISSPIT.2016.7886003
DO - 10.1109/ISSPIT.2016.7886003
M3 - Conference contribution
AN - SCOPUS:85017581369
T3 - 2016 IEEE International Symposium on Signal Processing and Information Technology, ISSPIT 2016
SP - 22
EP - 27
BT - 2016 IEEE International Symposium on Signal Processing and Information Technology, ISSPIT 2016
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2016 IEEE International Symposium on Signal Processing and Information Technology, ISSPIT 2016
Y2 - 12 December 2016 through 14 December 2016
ER -