Propose a new definition for the time-dependent spectrum of a non-stationary process. The authors show that this spectrum can be considered as the dual of Priestley's (1981) evolutionary spectrum. They demonstrate how this dual spectrum successfully characterizes signals whose components are highly localized in time. They give an estimator of this dual spectrum. Finally, they present examples which demonstrate that the new spectral definition is more suitable than current approaches for many signals. They show that the value of the method depends on the signal under consideration. They also illustrate that if the signal content is not known, then combining estimates of the evolutionary spectrum and its dual yields good results.<<ETX>>
[1]
D. B. Preston.
Spectral Analysis and Time Series
,
1983
.
[2]
A. El-Jaroudi,et al.
Evolutionary periodogram for nonstationary signals
,
1994,
IEEE Trans. Signal Process..
[3]
J. Bertrand,et al.
Time-frequency representations of broad-band signals
,
1988
.
[4]
Les E. Atlas,et al.
Bilinear time-frequency representations: new insights and properties
,
1993,
IEEE Trans. Signal Process..
[5]
Amro El-Jaroudi,et al.
Relating the bilinear distributions and the evolutionary spectrum
,
1993,
1993 IEEE International Conference on Acoustics, Speech, and Signal Processing.