没有合适的资源?快使用搜索试试~ 我知道了~
Matlab语音信号分析使用STFT论文-Robust STFT with Adapative Window Length an...
需积分: 9 0 下载量 24 浏览量
2019-08-13
06:33:54
上传
评论
收藏 237KB PDF 举报
温馨提示
试读
3页
Matlab语音信号分析使用STFT论文-Robust STFT with Adapative Window Length and Rotation.pdf 推荐以下三篇论文,是成功解决了STFT的分辨率问题,即可调的STFT,并没弃之选择小波等分析技术。 美国某大学:Improved Instantaneous Frequency Estimation Using an Adaptive Short-Time Fourier Transform 字数偏多 新加坡南洋理工大学:Robust STFT with Adaptive Window Length and Rotation Direction 绝对赞!可惜一直没有画出里面的图 拉脱维亚大学:Adaptive STFT-like Time-Frequency analysis from arbitrary distributed signal samples 多种分析技术的比较
资源推荐
资源详情
资源评论
ICICS-F'CM
2003
15-18
-ta
2003
Sinsaporc
2A4.4
Robust
STFT
with Adaptive Window Length
and
Rotation
Direction
Yongmei
Wei;
Guoan
Bi
School
of
Electrical
and
Eledronic
Enginee~ing
Nanyang
Technological Unive&ty
Abstract
This
paper
presents
a
short
time Fourier transform that
uses
windows
with adaptive length
and
rotation
direction
for
signals
that contain
both
linear
and
nonlinear
cop
nents.
Experiments
show that the
proposed
method
is
1-
bud
against impulse noise
and
provides a good
molution.
1.
Introduction
The
short
time Fourier transform (STFT)
is
defined
as:
m
STFT(t,
f)
=
1,
z(T)k*(T
-
t)
e-jZwfTd7
(1)
where
z(t)
is
the
signal
tobe
analyd
and
k(t)
is
alow
pass
filter whose
spectrum
is
parallel
with the time
ais
in the
timefrequency domain.
Due
to
its
simpliaiy,
STFT
has
been
the most
popular
method
of
time
frequency represen-
tation although
it
generally provides
a
low
rdution
for
nonstationary
signals.
Several
STFTs with adaptive
wh
dows were developed
in
the literature
by
seldi
a
win-
dow
that
matches
the
direction
of
the
signal
in
the
time-
frequency plane from a
set
of
available
windows.
For
exam
ple, without
conside
the rotation
-0%
an STFT
with adaptive window-length
wa
reported
in
[l]
to
obtain
some
limited
improvement
on
the resolution,
and
a larger
set
of
windows
was
reported
in
[Z]
for
further
improvement.
A
better resolution
can
be
achieved
because the sdected
window
has
the same direction
as
that
of
the
signal
cop
nent
in
the timefrequency
plane.
However,
it
is
not
&-
tive
to
we
thse
methods
for
signals
having
multiple
wm-
pond
and
is
difficult
to
deal
with
sigaals
in
impulse noise.
Based
on
the
observations
from the reported work,
it
can
be
conduded
that
a
good
resolution
can
be
achieved
if
the
following
issue
are properly
cornidered.
The
window
direction
in
the timsfrequency
plane
should be the same
a
that
of
the
signal
ComponenQ
Thewidow lengthshouldbedeterminedby thedegree
of
linearity
of
the
signal
components.
In
general, a
long window
is
used
for
linear
component
and
a short
window
for
nonlinear
component;
Resolution
can
be
improved for
signals
that
have mul-
tiple
compo-
if
the
window
is
achieved
by
a
nu-
metical
average
of
windows
that
are
opthid
for
in-
dividual
components.
Previody
reported work dealt with
only
some
of
these
is
sues.
This
paper
attempts
to develq
an
STFT for
signals
of
multiple
components
with adaptive
window
length
and
direction.
2.
Chirp
rate estimation
The rotated
window
is
ischiwed
by
multiplying the
convex-
tional
low-pass filter with a
hear
chirp
that
has
the same
dirpction
(or
chirp
rate)
as
that
of
the
signal
component.
Therefore,
chirp
rate
estimation
becomes
a
main
problem.
With Gausian
&e,
a good
gtimation
can
be
achieved
from the
methods
bed
on
produd high-order
ity
funrton
or
fractional
Fourier transform
[3,4].
However,
bothmethods
havedifficutiesindealingwithimpulsenoise.
Robust
discrete Foder transform method reported
in
[5]
&ectively deals with
stationary
signals
with a wide range
of
impulse
noises.
We me a
similar
colMept
to design a
robust
chirp
Fourier transform (RCFT) to
find
the
chirp
rates
of
non-stationary
signals.
Let
us
divide an N-point
signal
that
contains
multiple
linear
and/or
non-linear
wm-
pone&
into
M
small
segments,
each
is
mumed
to contain
hear
chirplike
components
that
may
not
be
parallel
with
the time
=&S.
With the
a?.snmption
that
the
signal
seg-
ment
zj(n)
is
corrqted
by
noise
U(%),
the
RCFT
can
be
achievedbythe
' ' '
''
on
problem:
N-1
RCFT(Z,
k)
=
argmin,
F(e(Z,
k,n))
(2)
4
-jzr
,..~+h.
where
e(Z,k,n)
=
zj(n)e-
-
m,
F(e)
is
a
lass
function
and
1
indicates
the
chirp
rate. To deal with the
impulse
noise (i.e. F(e)
=
lei),
the
marginal
median filter
approach
[6]
is
employed
to obtain
RCFT(~,
k)
=
median{RE[z;.]}
+
j
median{lm[*j]}
(3)
where
5j
=
zj(n)e
'N
'.
The
chirp
rate
candidate
for
each
segment
can
be found
by
arching
the
+
OF
-j2n
C.,Z+*"
0-7803-8185-8/03/$17.00
0
2003
IEZE
827
资源评论
weixin_39841882
- 粉丝: 443
- 资源: 1万+
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功