1. 傅里叶变换、相关函数与功率谱密度
输入信号:
x
(
t
)
=
∫
−
∞
+
∞
S
x
(
f
)
e
j
2
π
f
t
d
f
S
x
(
f
)
=
∫
−
∞
+
∞
x
(
t
)
e
−
j
2
π
f
t
d
t
\begin{array}{l} x(t)=\int_{-\infty}^{+\infty} S_{x}(f) e^{j 2 \pi f t} d f \\ S_{x}(f)=\int_{-\infty}^{+\infty} x(t) e^{-j 2 \pi f t} d t \end{array}
x(t)=∫−∞+∞Sx(f)ej2πftdfSx(f)=∫−∞+∞x(t)e−j2πftdt
输出信号:
y
(
t
)
=
∫
−
∞
+
∞
S
y
(
f
)
e
j
2
π
f
t
d
f
S
y
(
f
)
=
∫
−
∞
+
∞
y
(
t
)
e
−
j
2
π
f
t
d
t
\begin{array}{l} y(t)=\int_{-\infty}^{+\infty} S_{y}(f) e^{j 2 \pi f t} d f \\ S_{y}(f)=\int_{-\infty}^{+\infty} y(t) e^{-j 2 \pi f t} d t \end{array}
y(t)=∫−∞+∞Sy(f)ej2πftdfSy(f)=∫−∞+∞y(t)e−j2πftdt
冲击响应:
h
(
t
)
=
∫
−
∞
+
∞
H
(
f
)
e
j
2
π
f
t
d
f
H
(
f
)
=
∫
−
∞
+
∞
h
(
t
)
e
−
j
2
π
f
t
d
t
\begin{array}{l} h(t)=\int_{-\infty}^{+\infty} H(f) e^{j 2 \pi f t} d f \\ H(f)=\int_{-\infty}^{+\infty} h(t) e^{-j 2 \pi f t} d t \end{array}
h(t)=∫−∞+∞H(f)ej2πftdfH(f)=∫−∞+∞h(t)e−j2πftdt
输入信号自相关函数与自功率谱:
R
x
x
(
τ
)
=
lim
T
→
∞
1
T
∫
T
x
(
t
)
x
(
t
+
τ
)
d
t
G
x
x
(
f
)
=
∫
−
∞
+
∞
R
x
x
(
τ
)
e
−
j
2
π
f
d
τ
=
S
x
(
f
)
⋅
S
x
∗
(
f
)
\begin{array}{c} R_{x x}(\tau)=\lim _{T \rightarrow \infty} \frac{1}{T} \int_{T} x(t) x(t+\tau) d t \\ G_{x x}(f)=\int_{-\infty}^{+\infty} R_{x x}(\tau) e^{-j 2 \pi f} d \tau=S_{x}(f) \cdot S_{x}^{*}(f) \end{array}
Rxx(τ)=limT→∞T1∫Tx(t)x(t+τ)dtGxx(f)=∫−∞+∞Rxx(τ)e−j2πfdτ=Sx(f)⋅Sx∗(f)
输出信号自相关函数与自功率谱:
R
y
y
(
τ
)
=
lim
T
→
∞
1
T
∫
T
y
(
t
)
y
(
t
+
τ
)
d
t
G
y
y
(
f
)
=
∫
−
∞
+
∞
R
y
y
(
τ
)
e
−
j
2
π
f
t
d
τ
=
S
y
(
f
)
⋅
S
y
∗
(
f
)
\begin{array}{c} R_{y y}(\tau)=\lim _{T \rightarrow \infty} \frac{1}{T} \int_{T} y(t) y(t+\tau) d t \\ G_{y y}(f)=\int_{-\infty}^{+\infty} R_{y y}(\tau) e^{-j 2 \pi f t} d \tau=S_{y}(f) \cdot S_{y}^{*}(f) \end{array}
Ryy(τ)=limT→∞T1∫Ty(t)y(t+τ)dtGyy(f)=∫−∞+∞Ryy(τ)e−j2πftdτ=Sy(f)⋅Sy∗(f)
互相关函数与互功率谱:
R
y
x
(
τ
)
=
lim
T
→
∞
1
T
∫
T
y
(
t
)
x
(
t
+
τ
)
d
t
G
y
x
(
f
)
=
∫
−
∞
+
∞
R
y
x
(
τ
)
e
−
j
2
π
f
d
τ
=
S
y
(
f
)
⋅
S
x
∗
(
f
)
\begin{array}{c} R_{y x}(\tau)=\lim _{T \rightarrow \infty} \frac{1}{T} \int_{T} y(t) x(t+\tau) d t \\ G_{y x}(f)=\int_{-\infty}^{+\infty} R_{y x}(\tau) e^{-j 2 \pi f} d \tau=S_{y}(f) \cdot S_{x}^{*}(f) \end{array}
Ryx(τ)=limT→∞T1∫Ty(t)x(t+τ)dtGyx(f)=∫−∞+∞Ryx(τ)e−j2πfdτ=Sy(f)⋅Sx∗(f)
2. 频响估计方法
由于信号采集过程中不可避免会受到噪声干扰,因此频响函数估计问题是在实际工程中必须要面对的问题。为了方便起见,信号统一采用频域方式来表达,令系统的输入信号为 F ( f ) F(f) F(f),输出信号为 X ( f ) X(f) X(f),输入信号噪声为 M ( f ) M(f) M(f),输出信号噪声为 N ( f ) N(f) N(f),实际输入信号为 Z ( f ) = F ( f ) + M ( f ) Z(f)=F(f)+M(f) Z(f)=F(f)+M(f),实际输出信号为 Y ( f ) + X ( f ) + N ( f ) Y(f)+X(f)+N(f) Y(f)+X(f)+N(f)。
2.1 H 1 H_1 H1估计(输出含有噪声,输入没有)
H
1
=
G
y
f
G
f
f
=
Y
(
f
)
F
∗
(
f
)
F
(
f
)
F
∗
(
f
)
=
(
X
(
f
)
+
N
(
f
)
)
F
∗
(
f
)
F
(
f
)
F
∗
(
f
)
=
X
(
f
)
F
∗
(
f
)
+
N
(
f
)
F
∗
(
f
)
F
(
f
)
F
∗
(
f
)
H_{1}=\frac{G_{\mathrm{yf}}}{G_{f f}}=\frac{Y(f) F^{*}(f)}{F(f) F^{*}(f)}=\frac{(X(f)+N(f)) F^{*}(f)}{F(f) F^{*}(f)}=\frac{X(f) F^{*}(f)+N(f) F^{*}(f)}{F(f) F^{*}(f)}
H1=GffGyf=F(f)F∗(f)Y(f)F∗(f)=F(f)F∗(f)(X(f)+N(f))F∗(f)=F(f)F∗(f)X(f)F∗(f)+N(f)F∗(f)
因为输出噪声与输入无关,所以在经过多次平均后,
H
1
=
X
(
f
)
F
∗
(
f
)
F
(
f
)
F
∗
(
f
)
=
X
(
f
)
F
(
f
)
=
H
0
H_{1}=\frac{X(f) F^{*}(f)}{F(f) F^{*}(f)}=\frac{X(f)}{F(f)}=H_{0}
H1=F(f)F∗(f)X(f)F∗(f)=F(f)X(f)=H0
2.2 H 2 H_2 H2估计(输入,输出含有噪声)
H
1
=
G
y
z
G
z
z
=
Y
(
f
)
Z
∗
(
f
)
Z
(
f
)
Z
∗
(
f
)
=
(
X
(
f
)
+
N
(
f
)
)
(
M
∗
(
f
)
+
F
∗
(
f
)
)
(
M
(
f
)
+
F
(
f
)
)
(
M
∗
(
f
)
+
F
∗
(
f
)
)
=
X
(
f
)
M
∗
(
f
)
+
X
(
f
)
F
∗
(
f
)
+
N
(
f
)
M
∗
(
f
)
+
N
(
f
)
F
∗
(
f
)
M
(
f
)
M
∗
(
f
)
+
F
(
f
)
M
∗
(
f
)
+
M
(
f
)
F
∗
(
f
)
+
F
(
f
)
F
∗
(
f
)
=
G
x
m
+
G
x
f
+
G
n
m
+
G
n
f
G
m
m
+
G
f
n
n
+
G
m
f
+
G
f
f
\begin{aligned} H_{1} & =\frac{G_{\mathrm{yz}}}{G_{z z}}=\frac{Y(f) Z^{*}(f)}{Z(f) Z^{*}(f)}=\frac{(X(f)+N(f))\left(M^{*}(f)+F^{*}(f)\right)}{(M(f)+F(f))\left(M^{*}(f)+F^{*}(f)\right)} \\ & =\frac{X(f) M^{*}(f)+X(f) F^{*}(f)+N(f) M^{*}(f)+N(f) F^{*}(f)}{M(f) M^{*}(f)+F(f) M^{*}(f)+M(f) F^{*}(f)+F(f) F^{*}(f)} \\ & =\frac{G_{x m}+G_{x f}+G_{n m}+G_{n f}}{G_{m m}+G_{f n n}+G_{m f}+G_{f f}} \end{aligned}
H1=GzzGyz=Z(f)Z∗(f)Y(f)Z∗(f)=(M(f)+F(f))(M∗(f)+F∗(f))(X(f)+N(f))(M∗(f)+F∗(f))=M(f)M∗(f)+F(f)M∗(f)+M(f)F∗(f)+F(f)F∗(f)X(f)M∗(f)+X(f)F∗(f)+N(f)M∗(f)+N(f)F∗(f)=Gmm+Gfnn+Gmf+GffGxm+Gxf+Gnm+Gnf
因为输入与输出噪声无关,输出与输入噪声无关,所以在经过多次平均后,
H
1
=
G
x
f
G
m
m
+
G
f
f
=
X
(
f
)
F
∗
(
f
)
M
(
f
)
M
∗
(
f
)
+
F
(
f
)
F
∗
(
f
)
=
H
0
1
+
G
m
m
/
G
f
f
<
H
0
H_{1}=\frac{G_{x f}}{G_{m m}+G_{f f}}=\frac{X(f) F^{*}(f)}{M(f) M^{*}(f)+F(f) F^{*}(f)}=\frac{H_{0}}{1+G_{m m} / G_{f f}}<H_{0}
H1=Gmm+GffGxf=M(f)M∗(f)+F(f)F∗(f)X(f)F∗(f)=1+Gmm/GffH0<H0
由此可知,在这种情况下,无论经过多少次平均,频响函数的估计总是偏小的,除非激励远大于输入噪声,才可认为测量的频响为真实频响。
当
H
2
=
G
y
y
G
z
y
H_{2}=\frac{G_{y y}}{G_{z y}}
H2=GzyGyy,使用该频响函数有助于改善频响的近似性,并且
H
2
=
G
y
y
G
z
y
=
H
0
(
1
+
G
m
n
G
x
x
)
>
H
0
H_{2}=\frac{G_{y y}}{G_{z y}}=H_{0}\left(1+\frac{G_{m n}}{G_{x x}}\right)>H_{0}
H2=GzyGyy=H0(1+GxxGmn)>H0
2.3 相干
γ
f
y
2
=
H
1
H
2
=
1
(
1
+
G
m
m
/
G
f
f
)
(
1
+
G
n
n
/
G
x
x
)
\gamma_{f y}^{2}=\frac{H_{1}}{H_{2}}=\frac{1}{\left(1+G_{m m} / G_{f f}\right)\left(1+G_{n n} / G_{x x}\right)}
γfy2=H2H1=(1+Gmm/Gff)(1+Gnn/Gxx)1
当
G
m
m
≪
0
,
G
n
n
≪
0
G_{mm} \ll 0 , G_{n n} \ll 0
Gmm≪0,Gnn≪0时,相干近似为1,因此相干是衡量一次测试质量是否好(噪声的影响较低)的重要指标。