| Summary: | 碩士 === 大同工學院 === 電機工程研究所 === 81 === Higherorder spectrum are defined in terms of cumulants
therefore are also called cumulant spectrum. In recent years,
bispectral analysis has gained its popularity in many
application fields because of affordable fast computing
facilities and better interpretation and understanting of
higher-order statistics. Unlike the power spectrum which
contains the Fourier-magnitude information only, the bispectrum
contains all the information concerning the Fourier phase and
Fourier-magnitude of the signal. To use this property of
bispectrum, this research first attempts to investigate methods
of bispectrum estimation for finite data. We then use these
properties to reconstruct the signal. In addition we discuss
methods for signal reconstruction solely from the bispectral
phase of a sequence to achieve data reduction.Furthermore, it
is known that the Gaussian noise was zero at the higher-order
(>2) statistics. We attempt to suppress the Gaussian noise by
the characteristic that the third-order moment of the Gaussian
noise is zero. From our experimental results, it is found that
bispectrum can be used to suppress the Gaussian noise and
enhance speech signals.
|
|---|
