科恩系列分布 (Cohen's class distribution)于1966年由L. Cohen首次提出,且其使用双线性转换亦是此种转换形式中最通用的一种。在几种常见的时频分布 中,Cohen's class分布是最强大的转换之一。随著近几年来时频分析 发展,应用也越来越多元。Cohen's class分布和短时距傅立叶变换 比较起来有较高的清晰度,但也相对的有交叉项(cross-term)的问题,不过可选择适当的遮罩函数(mask function)来将交叉项的问题降到最低。
数学定义
C
x
(
t
,
f
)
=
∫
−
∞
∞
∫
−
∞
∞
A
x
(
η
,
τ
)
Φ
(
η
,
τ
)
exp
(
j
2
π
(
η
t
−
τ
f
)
)
d
η
d
τ
,
{\displaystyle C_{x}(t,f)=\int _{-\infty }^{\infty }\int _{-\infty }^{\infty }A_{x}(\eta ,\tau )\Phi (\eta ,\tau )\exp(j2\pi (\eta t-\tau f))\,d\eta \,d\tau ,}
,
其中
A
x
(
η
,
τ
)
=
∫
−
∞
∞
x
(
t
+
τ
/
2
)
x
∗
(
t
−
τ
/
2
)
e
−
j
2
π
t
η
d
t
.
{\displaystyle A_{x}\left(\eta ,\tau \right)=\int _{-\infty }^{\infty }x(t+\tau /2)x^{*}(t-\tau /2)e^{-j2\pi t\eta }\,dt.}
为模糊函数(Ambiguity Function) ,且
Φ
(
η
,
τ
)
{\displaystyle \Phi \left(\eta ,\tau \right)}
为一遮罩函数,通常是低通函数用来滤除杂讯。
科恩系列分布函数
韦格纳分布(Wigner Distribution Function)
当Cohen's class分布中的
Φ
(
η
,
τ
)
=
1
{\displaystyle \Phi \left(\eta ,\tau \right)=1}
时,Cohen's class分布会成韦格纳分布(Wigner distribution function)
W
x
(
t
,
f
)
=
∫
−
∞
∞
x
(
t
+
τ
/
2
)
x
∗
(
t
−
τ
/
2
)
e
−
j
2
π
f
τ
d
τ
{\displaystyle W_{x}(t,f)=\int _{-\infty }^{\infty }x(t+\tau /2)x^{*}(t-\tau /2)e^{-j2\pi f\tau }\,d\tau }
。
利用韦格纳分布对函数
x
(
t
)
=
e
x
p
(
j
0.015
t
4
+
j
0.06
t
3
−
j
0.3
t
2
+
j
t
)
{\displaystyle x(t)=exp(j0.015t^{4}+j0.06t^{3}-j0.3t^{2}+jt)}
作时频分析的结果可见右图。
锥状分布(Cone-Shape Distribution)
当Cohen's class分布中的
ϕ
(
t
,
τ
)
=
1
|
τ
|
e
x
p
(
−
2
π
α
τ
2
)
Π
(
t
τ
)
{\displaystyle \phi (t,\tau )={\frac {1}{\left|\tau \right|}}exp(-2\pi \alpha \tau ^{2})\Pi ({\frac {t}{\tau }})}
,且
Φ
(
η
,
τ
)
=
s
i
n
c
(
η
τ
)
e
x
p
(
(
−
2
π
α
τ
2
)
{\displaystyle \Phi \left(\eta ,\tau \right)=sinc(\eta \tau )exp((-2\pi \alpha \tau ^{2})}
时,
其中
ϕ
(
t
,
τ
)
=
∫
−
∞
∞
Φ
(
η
,
τ
)
e
x
p
(
j
2
π
η
t
)
d
η
{\displaystyle \phi (t,\tau )=\int _{-\infty }^{\infty }\Phi (\eta ,\tau )exp(j2\pi \eta t)d\eta }
,Cohen's class分布会成锥状分布。
右图为不同的
α
{\displaystyle \alpha }
值下的锥状分布时频分析图。
乔伊-威廉斯(Choi-Williams)
当Cohen's class分布中的
Φ
(
η
,
τ
)
=
e
x
p
[
−
α
(
η
τ
)
2
]
{\displaystyle \Phi \left(\eta ,\tau \right)=exp[-\alpha (\eta \tau )^{2}]}
时,Cohen's class分布会成乔伊-威廉斯分布。
右图为不同的
α
{\displaystyle \alpha }
值下的锥状分布时频分析图。
科恩系列分布优缺点
优点:
1.可选择适当的遮罩函数来避免掉交叉项问题 。
2.具有高清晰度。
缺点
1. 需要较高的计算量与时间。
2. 缺乏良好的数学特性。
科恩系列分布的实现
C
x
(
t
,
f
)
=
∫
−
∞
∞
∫
−
∞
∞
A
x
(
η
,
τ
)
Φ
(
η
,
τ
)
exp
(
j
2
π
(
η
t
−
τ
f
)
)
d
η
d
τ
,
{\displaystyle C_{x}(t,f)=\int _{-\infty }^{\infty }\int _{-\infty }^{\infty }A_{x}(\eta ,\tau )\Phi (\eta ,\tau )\exp(j2\pi (\eta t-\tau f))\,d\eta \,d\tau ,}
=
∫
−
∞
∞
∫
−
∞
∞
∫
−
∞
∞
x
(
u
+
τ
2
)
x
∗
(
u
−
τ
2
)
Φ
(
η
,
τ
)
e
−
j
2
π
u
η
+
j
2
π
(
η
t
−
τ
f
)
d
u
d
τ
d
η
{\displaystyle =\int _{-\infty }^{\infty }\int _{-\infty }^{\infty }\int _{-\infty }^{\infty }x(u+{\frac {\tau }{2}})x^{*}(u-{\frac {\tau }{2}})\Phi (\eta ,\tau )e^{-j2\pi u\eta +j2\pi (\eta t-\tau f)}dud\tau d\eta }
简化方法一:不是所有的
A
x
(
η
,
τ
)
{\displaystyle A_{x}(\eta ,\tau )}
的值都要计算出
对
|
η
|
>
B
{\displaystyle \ \left|\eta \right|>B\ }
或
|
τ
|
>
C
{\displaystyle \ \left|\tau \right|>C}
,若
Φ
(
η
,
τ
)
=
0
{\displaystyle \Phi (\eta ,\tau )=0}
,则
C
x
(
t
,
f
)
=
∫
−
C
C
∫
−
B
B
∫
−
∞
∞
x
(
u
+
τ
2
)
x
∗
(
u
−
τ
2
)
Φ
(
η
,
τ
)
e
−
j
2
π
u
η
+
j
2
π
(
η
t
−
τ
f
)
d
u
d
τ
d
η
{\displaystyle C_{x}(t,f)=\int _{-C}^{C}\int _{-B}^{B}\int _{-\infty }^{\infty }x(u+{\frac {\tau }{2}})x^{*}(u-{\frac {\tau }{2}})\Phi (\eta ,\tau )e^{-j2\pi u\eta +j2\pi (\eta t-\tau f)}dud\tau d\eta }
简化方法二:注意,
η
{\displaystyle \eta }
这个参数和输入及输出都无关
C
x
(
t
,
f
)
=
∫
−
C
C
∫
−
∞
∞
x
(
u
+
τ
2
)
x
∗
(
u
−
τ
2
)
[
∫
−
B
B
Φ
(
η
,
τ
)
e
−
j
2
π
,
η
(
t
−
u
)
d
η
]
e
−
j
2
π
τ
,
f
d
u
d
τ
{\displaystyle C_{x}(t,f)=\int _{-C}^{C}\int _{-\infty }^{\infty }x(u+{\frac {\tau }{2}})x^{*}(u-{\frac {\tau }{2}})[\int _{-B}^{B}\Phi (\eta ,\tau )e^{-j2\pi ,\eta (t-u)}d\eta ]e^{-j2\pi \tau ,f}dud\tau }
=
∫
−
C
C
∫
−
∞
∞
x
(
u
+
τ
2
)
x
∗
(
u
−
τ
2
)
Φ
(
τ
,
t
−
u
)
e
−
j
2
π
τ
,
f
d
u
d
τ
{\displaystyle =\int _{-C}^{C}\int _{-\infty }^{\infty }x(u+{\frac {\tau }{2}})x^{*}(u-{\frac {\tau }{2}})\Phi (\tau ,t-u)e^{-j2\pi \tau ,f}dud\tau }
,其中
Φ
(
τ
,
t
−
u
)
=
∫
−
B
B
Φ
(
η
,
τ
)
e
−
j
2
π
,
η
(
t
−
u
)
d
η
{\displaystyle \Phi (\tau ,t-u)=\int _{-B}^{B}\Phi (\eta ,\tau )e^{-j2\pi ,\eta (t-u)}d\eta }
,由于
Φ
(
τ
,
t
−
u
)
{\displaystyle \Phi (\tau ,t-u)}
和输入无关,可事先算出,因此可简化成两个积分式。
简化方法三:使用折积方法(convolution)
C
x
(
t
,
f
)
=
∫
−
∞
∞
∫
−
∞
∞
x
(
u
+
τ
2
)
x
∗
(
u
−
τ
2
)
ϕ
(
t
−
u
,
τ
)
d
u
e
−
j
2
π
f
τ
d
τ
{\displaystyle C_{x}(t,f)=\int _{-\infty }^{\infty }\int _{-\infty }^{\infty }x(u+{\frac {\tau }{2}})x^{*}(u-{\frac {\tau }{2}})\phi (t-u,\tau )due^{-j2\pi f\tau }d\tau }
,其中
ϕ
(
t
,
τ
)
=
∫
−
∞
∞
Φ
(
η
,
τ
)
e
x
p
(
j
2
π
η
t
)
d
η
{\displaystyle \phi (t,\tau )=\int _{-\infty }^{\infty }\Phi (\eta ,\tau )exp(j2\pi \eta t)d\eta }
。对
|
t
|
>
b
{\displaystyle \left|t\right|>b}
或是
|
τ
|
>
c
{\displaystyle \left|\tau \right|>c}
,则
C
x
(
t
,
f
)
=
∫
−
c
c
∫
t
−
b
t
+
b
x
(
u
+
τ
2
)
x
∗
(
u
−
τ
2
)
ϕ
(
t
−
u
,
τ
)
d
u
e
−
j
2
π
f
τ
d
τ
{\displaystyle C_{x}(t,f)=\int _{-c}^{c}\int _{t-b}^{t+b}x(u+{\frac {\tau }{2}})x^{*}(u-{\frac {\tau }{2}})\phi (t-u,\tau )due^{-j2\pi f\tau }d\tau }
,上式为一折积式。
模糊函数 (Ambiguity Function)
模糊函数的定义为:
A
x
(
η
,
τ
)
=
∫
−
∞
∞
x
(
t
+
τ
2
)
x
∗
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
{\displaystyle A_{x}\left(\eta ,\tau \right)=\int _{-\infty }^{\infty }x(t+{\tfrac {\tau }{2}})x^{*}(t-{\tfrac {\tau }{2}})e^{-j2\pi t\eta }\,dt}
Modulation 和 Time Shifting 对模糊函数的影响
我们来看一下
x
(
t
)
{\displaystyle x(t)}
对于模糊函数的影响
(1) 假设
x
1
(
t
)
{\displaystyle x_{1}(t)}
是一个高斯函数:
a
e
−
(
t
−
b
)
2
/
2
c
2
{\displaystyle ae^{-(t-b)^{2}/2c^{2}}}
, 其中
a
=
1
,
b
=
0
,
c
=
1
2
α
{\displaystyle a=1,b=0,c={\sqrt {\tfrac {1}{2\alpha }}}}
那么我们可以得到
x
1
(
t
)
=
e
−
α
π
t
2
{\displaystyle x_{1}(t)=e^{-\alpha \pi t^{2}}}
, 代入模糊函数
A
x
(
η
,
τ
)
{\displaystyle A_{x}\left(\eta ,\tau \right)}
中:
A
x
1
(
η
,
τ
)
=
∫
−
∞
∞
e
−
α
π
(
t
+
τ
2
)
2
e
−
α
π
(
t
−
τ
2
)
2
e
−
j
2
π
t
η
d
t
{\displaystyle A_{x_{1}}\left(\eta ,\tau \right)=\textstyle \int \limits _{-\infty }^{\infty }\displaystyle e^{-\alpha \pi {(t+{\tfrac {\tau }{2}})}^{2}}\ e^{-\alpha \pi {(t-{\tfrac {\tau }{2}})}^{2}}\ e^{-j2\pi t\eta }\ dt}
=
∫
−
∞
∞
e
−
α
π
(
2
t
2
+
τ
2
2
)
e
−
j
2
π
t
η
d
t
{\displaystyle =\textstyle \int \limits _{-\infty }^{\infty }\displaystyle e^{-\alpha \pi (2t^{2}+{\tfrac {\tau }{2}}^{2})}\ e^{-j2\pi t\eta }\ dt}
(2) 假设
x
2
(
t
)
{\displaystyle x_{2}(t)}
是一个经过 shifting 和 modulation 的高斯函数:
那么我们可以得到
x
2
(
t
)
=
e
−
α
π
(
t
−
t
0
)
2
+
j
2
π
f
0
t
{\displaystyle x_{2}(t)=e^{-\alpha \pi (t-t_{0})^{2}+j2\pi f_{0}t}}
, 代入模糊函数
A
x
(
η
,
τ
)
{\displaystyle A_{x}\left(\eta ,\tau \right)}
中:
A
x
2
(
η
,
τ
)
=
∫
−
∞
∞
e
−
α
π
(
t
+
τ
2
−
t
0
)
2
+
j
2
π
f
0
(
t
+
τ
2
)
e
−
α
π
(
t
−
τ
2
−
t
0
)
2
−
j
2
π
f
0
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
{\textstyle A_{x_{2}}\left(\eta ,\tau \right)=\textstyle \int \limits _{-\infty }^{\infty }\displaystyle e^{-\alpha \pi {(t+{\tfrac {\tau }{2}}-t_{0})}^{2}+j2\pi f_{0}(t+{\tfrac {\tau }{2}})}\ e^{-\alpha \pi {(t-{\tfrac {\tau }{2}}-t_{0})}^{2}-j2\pi f_{0}(t-{\tfrac {\tau }{2}})}\ e^{-j2\pi t\eta }\ dt}
=
∫
−
∞
∞
e
−
α
π
(
2
(
t
−
t
0
)
2
+
τ
/
2
2
)
+
j
2
π
f
0
τ
e
−
j
2
π
t
η
d
t
{\displaystyle =\textstyle \int \limits _{-\infty }^{\infty }\displaystyle e^{-\alpha \pi (2(t-t_{0})^{2}+{\tau /2}^{2})+j2\pi f_{0}\tau }\ e^{-j2\pi t\eta }\ dt}
=
∫
−
∞
∞
e
−
α
π
(
2
t
2
+
τ
2
2
)
e
j
2
π
f
0
τ
e
−
j
2
π
t
η
e
−
j
2
π
t
0
η
d
t
{\displaystyle =\textstyle \int \limits _{-\infty }^{\infty }\displaystyle e^{-\alpha \pi (2t^{2}+{\tfrac {\tau ^{2}}{2}})}\ e^{j2\pi f_{0}\tau }\ e^{-j2\pi t\eta }\ e^{-j2\pi t_{0}\eta }\ dt}
我们可以看到
|
A
x
1
(
τ
,
η
)
|
=
|
A
x
2
(
τ
,
η
)
|
{\displaystyle |A_{x_{1}}\left(\tau ,\eta \right)|=|A_{x_{2}}\left(\tau ,\eta \right)|}
,
因此我们可以得出 time shifting
t
0
{\displaystyle t_{0}}
和 modulation
f
0
{\displaystyle f_{0}}
并不会影响
|
A
x
(
τ
,
η
)
|
{\displaystyle |A_{x}\left(\tau ,\eta \right)|}
积分后,
A
x
(
τ
,
η
)
=
1
2
α
e
−
π
(
α
τ
2
2
+
η
2
2
α
)
e
j
2
π
(
f
0
τ
−
t
0
η
)
{\displaystyle A_{x}\left(\tau ,\eta \right)={\sqrt {\tfrac {1}{2\alpha }}}e^{-\pi ({\tfrac {\alpha \tau ^{2}}{2}}+{\tfrac {\eta ^{2}}{2\alpha }})}e^{j2\pi (f_{0}\tau -t_{0}\eta )}}
所以
A
x
(
τ
,
η
)
{\displaystyle A_{x}\left(\tau ,\eta \right)}
在
τ
=
0
,
η
=
0
{\displaystyle \tau =0,\eta =0}
的地方会有最大的
|
A
x
(
τ
,
η
)
|
{\displaystyle |A_{x}\left(\tau ,\eta \right)|}
交叉项 Cross-term 问题
上述所列出来的是当
x
(
t
)
{\displaystyle x(t)}
只有一项而已 (one term only),如果
x
(
t
)
{\displaystyle x(t)}
有两项以上的元素构成 (more than two terms),
x
(
t
)
=
x
1
(
t
)
+
x
2
(
t
)
+
⋅
⋅
⋅
+
x
n
(
t
)
{\displaystyle x(t)=x_{1}(t)+x_{2}(t)+\cdot \cdot \cdot +x_{n}(t)}
,依然会有交叉项 (cross-term) 的问题存在。
假设
x
(
t
)
=
x
1
(
t
)
+
x
2
(
t
)
{\displaystyle x(t)=x_{1}(t)+x_{2}(t)}
, 其中
{
x
1
(
t
)
=
e
−
α
1
π
(
t
−
t
1
)
2
+
j
2
π
f
1
t
x
2
(
t
)
=
e
−
α
2
π
(
t
−
t
2
)
2
+
j
2
π
f
2
t
{\displaystyle {\begin{cases}x_{1}(t)=e^{-\alpha _{1}\pi (t-t_{1})^{2}+j2\pi f_{1}t}\\x_{2}(t)=e^{-\alpha _{2}\pi (t-t_{2})^{2}+j2\pi f_{2}t}\end{cases}}}
将
x
(
t
)
{\displaystyle x(t)}
代入模糊函数
A
x
(
η
,
τ
)
=
∫
−
∞
∞
x
(
t
+
τ
2
)
x
∗
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
{\displaystyle A_{x}\left(\eta ,\tau \right)=\int _{-\infty }^{\infty }x(t+{\tfrac {\tau }{2}})x^{*}(t-{\tfrac {\tau }{2}})e^{-j2\pi t\eta }\,dt}
中:
A
x
(
η
,
τ
)
=
∫
−
∞
∞
x
1
(
t
+
τ
2
)
x
1
∗
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
⏟
A
x
1
(
τ
,
η
)
+
∫
−
∞
∞
x
2
(
t
+
τ
2
)
x
2
∗
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
⏟
A
x
2
(
τ
,
η
)
{\displaystyle A_{x}\left(\eta ,\tau \right)=\underbrace {\int _{-\infty }^{\infty }x_{1}(t+{\tfrac {\tau }{2}})x_{1}^{*}(t-{\tfrac {\tau }{2}})e^{-j2\pi t\eta }\,dt} _{A_{x_{1}}(\tau ,\eta )}+\underbrace {\int _{-\infty }^{\infty }x_{2}(t+{\tfrac {\tau }{2}})x_{2}^{*}(t-{\tfrac {\tau }{2}})e^{-j2\pi t\eta }\,dt} _{A_{x_{2}}(\tau ,\eta )}}
+
∫
−
∞
∞
x
1
(
t
+
τ
2
)
x
2
∗
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
⏟
A
x
1
x
2
(
τ
,
η
)
+
∫
−
∞
∞
x
2
(
t
+
τ
2
)
x
1
∗
(
t
−
τ
2
)
e
−
j
2
π
t
η
d
t
⏟
A
x
2
x
1
(
τ
,
η
)
{\displaystyle +\underbrace {\int _{-\infty }^{\infty }x_{1}(t+{\tfrac {\tau }{2}})x_{2}^{*}(t-{\tfrac {\tau }{2}})e^{-j2\pi t\eta }\,dt} _{A_{{x_{1}}{x_{2}}}(\tau ,\eta )}+\underbrace {\int _{-\infty }^{\infty }x_{2}(t+{\tfrac {\tau }{2}})x_{1}^{*}(t-{\tfrac {\tau }{2}})e^{-j2\pi t\eta }\,dt} _{A_{{x_{2}}{x_{1}}}(\tau ,\eta )}}
其中
{
A
u
t
o
−
t
e
r
m
s
:
A
x
1
(
τ
,
η
)
,
A
x
2
(
τ
,
η
)
C
r
o
s
s
−
t
e
r
m
s
:
A
x
1
x
2
(
τ
,
η
)
,
A
x
2
x
1
(
τ
,
η
)
{\displaystyle {\begin{cases}Auto-terms:\quad A_{x_{1}}(\tau ,\eta ),\ A_{x_{2}}(\tau ,\eta )\\Cross-terms:\ A_{{x_{1}}{x_{2}}}(\tau ,\eta ),\ A_{{x_{2}}{x_{1}}}(\tau ,\eta )\end{cases}}}
Auto - terms
A
x
1
(
τ
,
η
)
=
1
2
α
1
e
−
π
(
α
1
τ
2
2
+
η
2
2
α
1
)
e
j
2
π
(
f
1
τ
−
t
1
η
)
{\displaystyle A_{x_{1}}\left(\tau ,\eta \right)={\sqrt {\tfrac {1}{2\alpha _{1}}}}\ e^{-\pi ({\tfrac {\alpha _{1}\tau ^{2}}{2}}+{\tfrac {\eta ^{2}}{2\alpha _{1}}})}\ e^{j2\pi (f_{1}\tau -t_{1}\eta )}}
A
x
2
(
τ
,
η
)
=
1
2
α
2
e
−
π
(
α
2
τ
2
2
+
η
2
2
α
2
)
e
j
2
π
(
f
2
τ
−
t
2
η
)
{\displaystyle A_{x_{2}}\left(\tau ,\eta \right)={\sqrt {\tfrac {1}{2\alpha _{2}}}}\ e^{-\pi ({\tfrac {\alpha _{2}\tau ^{2}}{2}}+{\tfrac {\eta ^{2}}{2\alpha _{2}}})}\ e^{j2\pi (f_{2}\tau -t_{2}\eta )}}
Cross - terms
(1)
α
1
≠
α
2
{\displaystyle \alpha _{1}\neq \alpha _{2}}
A
x
1
x
2
(
τ
,
η
)
=
1
(
α
1
+
α
2
)
e
−
π
(
(
α
1
+
α
2
)
(
τ
−
t
1
+
t
2
)
2
4
+
[
(
α
1
−
α
2
)
(
τ
−
t
1
+
t
2
)
−
j
2
(
η
−
f
1
+
f
2
)
]
2
4
(
α
1
+
α
2
)
)
e
j
2
π
[
(
f
1
+
f
2
2
)
τ
−
t
1
+
t
2
2
η
+
(
f
1
−
f
2
)
(
t
1
+
t
2
)
2
]
{\displaystyle A_{{x_{1}}{x_{2}}}(\tau ,\eta )={\sqrt {\tfrac {1}{(\alpha _{1}+\alpha _{2})}}}\ e^{-\pi ({\tfrac {(\alpha _{1}+\alpha _{2})(\tau -t_{1}+t_{2})^{2}}{4}}\ +\ {\tfrac {[(\alpha _{1}-\alpha _{2})(\tau -t_{1}+t_{2})-j2(\eta -f_{1}+f_{2})]^{2}}{4(\alpha _{1}+\alpha _{2})}})}\ e^{j2\pi [({\tfrac {f_{1}+f_{2}}{2}})\tau -{\tfrac {t_{1}+t_{2}}{2}}\eta +{\tfrac {(f_{1}-f_{2})(t_{1}+t_{2})}{2}}]}}
=
1
2
α
u
e
−
π
(
α
u
(
τ
−
t
d
)
2
2
+
[
α
d
(
τ
−
t
d
)
−
j
2
(
η
−
f
d
)
]
2
8
α
u
)
e
j
2
π
(
f
u
τ
−
t
n
η
+
f
d
t
u
)
{\displaystyle ={\sqrt {\tfrac {1}{2\alpha _{u}}}}\ e^{-\pi ({\tfrac {\alpha _{u}(\tau -t_{d})^{2}}{2}}\ +\ {\tfrac {[\alpha _{d}(\tau -t_{d})-j2(\eta -f_{d})]^{2}}{8\alpha _{u}}})}\ e^{j2\pi (f_{u}\tau -t_{n}\eta +f_{d}t_{u})}}
A
x
2
x
1
(
τ
,
η
)
=
A
x
1
x
2
∗
(
−
τ
,
−
η
)
{\displaystyle A_{{x_{2}}{x_{1}}}(\tau ,\eta )=A_{{x_{1}}{x_{2}}}^{*}(-\tau ,-\eta )}
{
t
u
=
t
1
+
t
2
2
,
f
u
=
f
1
+
f
2
2
,
α
u
=
α
1
+
α
2
2
t
d
=
t
1
−
t
2
,
f
d
=
f
1
−
f
2
,
α
d
=
α
1
−
α
2
{\displaystyle {\begin{cases}t_{u}={\tfrac {t_{1}+t_{2}}{2}},\ f_{u}={\tfrac {f_{1}+f_{2}}{2}},\ \alpha _{u}={\tfrac {\alpha _{1}+\alpha _{2}}{2}}\\t_{d}=t_{1}-t_{2},\ f_{d}=f_{1}-f_{2},\ \alpha _{d}=\alpha _{1}-\alpha _{2}\end{cases}}}
(2)
α
1
=
α
2
{\displaystyle \alpha _{1}=\alpha _{2}}
A
x
1
x
2
(
τ
,
η
)
=
1
2
α
u
e
−
π
(
α
u
(
τ
−
t
d
)
2
2
+
(
η
−
f
d
)
2
2
α
u
)
e
j
2
π
(
f
u
τ
−
t
n
η
+
f
d
t
u
)
{\displaystyle A_{{x_{1}}{x_{2}}}(\tau ,\eta )={\sqrt {\tfrac {1}{2\alpha _{u}}}}\ e^{-\pi ({\tfrac {\alpha _{u}(\tau -t_{d})^{2}}{2}}\ +\ {\tfrac {(\eta -f_{d})^{2}}{2\alpha _{u}}})}\ e^{j2\pi (f_{u}\tau -t_{n}\eta +f_{d}t_{u})}}
A
x
2
x
1
(
τ
,
η
)
=
A
x
1
x
2
∗
(
−
τ
,
−
η
)
{\displaystyle A_{{x_{2}}{x_{1}}}(\tau ,\eta )=A_{{x_{1}}{x_{2}}}^{*}(-\tau ,-\eta )}
因此,我们目前得到
A
x
1
(
τ
,
η
)
,
A
x
2
(
τ
,
η
)
{\displaystyle A_{x_{1}}\left(\tau ,\eta \right),A_{x_{2}}\left(\tau ,\eta \right)}
(auto-terms) 和
A
x
1
x
2
(
τ
,
η
)
,
A
x
2
x
1
(
τ
,
η
)
{\displaystyle A_{{x_{1}}{x_{2}}}(\tau ,\eta ),A_{{x_{2}}{x_{1}}}(\tau ,\eta )}
(cross-terms) 的公式,我们再仔细的分析 auto-terms 和 cross-terms 分别发生最大值的位置。
Ambiguity Function 分析图
首先,先看 Auto-terms:
|
A
x
1
(
τ
,
η
)
|
{\displaystyle |A_{x_{1}}\left(\tau ,\eta \right)|}
最大值发生在
τ
=
0
,
η
=
0
{\displaystyle \tau =0,\eta =0}
的地方
|
A
x
2
(
τ
,
η
)
|
{\displaystyle |A_{x_{2}}\left(\tau ,\eta \right)|}
最大值发生在
τ
=
0
,
η
=
0
{\displaystyle \tau =0,\eta =0}
的地方
而 Cross-terms:
|
A
x
1
x
2
(
τ
,
η
)
|
{\displaystyle |A_{{x_{1}}{x_{2}}}(\tau ,\eta )|}
最大值发生在
τ
=
t
d
,
η
=
f
d
{\displaystyle \tau =t_{d},\eta =f_{d}}
的地方
|
A
x
2
x
1
(
τ
,
η
)
|
{\displaystyle |A_{{x_{2}}{x_{1}}}(\tau ,\eta )|}
最大值发生在
τ
=
−
t
d
,
η
=
−
f
d
{\displaystyle \tau =-t_{d},\eta =-f_{d}}
的地方
换句话说,如果我们绘制一个 x轴为
τ
{\displaystyle \tau }
, y轴为
η
{\displaystyle \eta }
的座标图,Auto-terms发生在原点
(
0
,
0
)
{\displaystyle (0,0)}
的位置,而 Cross-terms 则是以原点为对称中心,在第一象限和第三象限的位置,
这也是为什么可以透过一个低通函数来滤除杂讯,把主成分 Auto-terms 分离出来,避免交叉项的问题。
与 维格纳分布 Wigner Distribution Function 的不同
维格纳分布是由尤金·维格纳于 1932 年提出的新的时频分析方法,对于非稳态的讯号有不错的表现。
相较于傅立叶转换或是短时距傅立叶转换,维格纳分布能有比较好的解析能力。
维格纳分布的定义为:
W
x
(
t
,
f
)
=
∫
−
∞
∞
x
(
t
+
τ
2
)
x
∗
(
t
−
τ
2
)
e
−
j
2
π
τ
f
d
τ
{\displaystyle W_{x}(t,f)=\int _{-\infty }^{\infty }x(t+{\frac {\tau }{2}})x^{*}(t-{\frac {\tau }{2}})e^{-j2\pi \tau f}\,d\tau }
如果我们假设
x
(
t
)
{\displaystyle x(t)}
是一个具有弦波特性的讯号,
x
(
t
)
=
e
j
2
π
f
0
t
{\displaystyle x(t)=e^{j2\pi f_{0}t}}
那么将此
x
(
t
)
{\displaystyle x(t)}
代入维格纳分布中,
Wigner Distribution Function 分析图
W
x
(
t
,
f
)
=
∫
−
∞
∞
e
j
2
π
f
0
(
t
+
τ
2
)
e
−
j
2
π
f
0
(
t
−
τ
2
)
e
−
j
2
π
τ
f
d
τ
{\displaystyle W_{x}(t,f)=\int _{-\infty }^{\infty }e^{j2\pi f_{0}(t+{\tfrac {\tau }{2}})}e^{-j2\pi f_{0}(t-{\tfrac {\tau }{2}})}\ e^{-j2\pi \tau f}\ d\tau }
=
∫
−
∞
∞
e
j
2
π
f
0
τ
e
−
j
2
π
τ
f
d
τ
{\displaystyle =\int _{-\infty }^{\infty }e^{j2\pi f_{0}\tau }\ e^{-j2\pi \tau f}\ d\tau }
=
∫
−
∞
∞
e
−
j
2
π
τ
(
f
−
f
0
)
d
τ
{\displaystyle =\int _{-\infty }^{\infty }e^{-j2\pi \tau (f-f_{0})}d\tau }
=
δ
(
f
−
f
0
)
{\displaystyle =\delta (f-f_{0})}
所以当
x
(
t
)
=
e
j
2
π
f
0
t
{\displaystyle x(t)=e^{j2\pi f_{0}t}}
时,
W
x
(
t
,
f
)
{\displaystyle W_{x}(t,f)}
在
f
=
f
0
{\displaystyle f=f_{0}}
的地方会有最大值。
换句话说,当
x
(
t
)
{\displaystyle x(t)}
有 modulation
f
0
{\displaystyle f_{0}}
或是有 time shifting
t
0
{\displaystyle t_{0}}
的情况发生时,会影响维格纳分布 (Wigner Distribution Function) 最大值
|
W
x
(
t
,
f
)
|
{\displaystyle |W_{x}(t,f)|}
的位置
然而,对于科恩系列分布 (Cohen's class distribution)而言,time shifting
t
0
{\displaystyle t_{0}}
和 modulation
f
0
{\displaystyle f_{0}}
并不会影响
|
A
x
(
τ
,
η
)
|
{\displaystyle |A_{x}\left(\tau ,\eta \right)|}
参考
Jian-Jiun Ding, Time frequency analysis and wavelet transform class note, the Department of Electrical Engineering, National Taiwan University (NTU), Taipei, Taiwan, 2007.
Jian-Jiun Ding, Time frequency analysis and wavelet transform class note, the Department of Electrical Engineering, National Taiwan University (NTU), Taipei, Taiwan, 2018.