Information Technology Reference
In-Depth Information
Comparison and Application of Time-Frequency
Analysis Methods for Nonstationary Signal Processing
Qiang Zhu, Yansong Wang, and Gongqi Shen
College of Automotive Engineering, Shanghai University of Engineering Science,
Shanghai, 201620, China
Abstract. Most of signals in engineering are nonstationary and time-varying.
The Fourier transform as a traditional approach can only provide the feature
information in frequency domain. The time-frequency techniques may give a
comprehensive description of signals in time-frequency planes. Based on some
typical nonstationary signals, five time-frequency analysis methods, i.e., the
short-time Fourier transform (STFT), wavelet transform (WT), Wigner- Ville
distribution (WVD), pseudo-WVD (PWVD) and the Hilbert-Huang transform
(HHT), were performed and compared in this paper. The characteristics of each
method were obtained and discussed. Compared with the other methods, the
HHT with a high time-frequency resolution can clearly describe the rules of the
frequency compositions changing with time, is a good approach for feature
extraction in nonstationary signal processing.
Keywords: Nonstationary signal, Short-time Fourier transform, Wavelet
transform, Wigner-Ville distribution, Hilbert-Huang transform.
1 Introduction
Many signals in engineering are time-varying. And the frequency features of the signals
are very important and can usually be used to distinguish one signal from the others.
The Fourier transform and its inversion play important roles in establishment of the
relationship between time and frequency domains. They are defined as follows:
()
j
2
π
ft
X
f
=
x
(
t
)
e
dt
(1)
.
()
j
2
π
ft
x
(
t
)
=
X
f
e
df
(2)
.
Based on the Fourier transform, the description and energy distribution of a signal in
frequency domain can only reflect its frequency features. As the Fourier transform and
its inversion are global transform, the signals can only be described entirely in time
domain or frequency domain. In practical applications, the Fourier transform is not
the best tool, due to most of signals encountered in engineering are nonstationary and
time varying, such as the signals of engine noises and vibrations. In order to study the
 
Search WWH ::




Custom Search