整波滤波器是一类能够整合具有任意频谱密度的静定随机信号的滤波器。其输入信号往往是白噪声。
由统计动力学笔记(二)频谱密度与线性随机系统的动态准确性(自留用)一文可以知道系统输出 x x x和输入 u u u之间的互频谱密度:
S x ( ω ) = W ( j ω ) W ( − j ω ) S u ( ω ) (1) S_x (\omega) = W(j \omega) W(-j \omega) S_u (\omega) \tag{1} Sx(ω)=W(jω)W(−jω)Su(ω)(1)当输入为白噪声时, S u ( ω ) = S n ( ω ) = 1 S_u (\omega) = S_n (\omega) = 1 Su(ω)=Sn(ω)=1,则
S x ( ω ) = W ( j ω ) W ( − j ω ) (2) S_x (\omega) = W(j \omega) W(-j \omega) \tag{2} Sx(ω)=W(jω)W(−jω)(2)这样一来,只要将输出端 x x x的频谱密度分解为2个共轭的部分,就可以得到系统的传递函数。这一步也称为频谱密度的分解。
例:输出端的频谱密度为
S x ( ω ) = 4 4 ω 2 + 1 = 2 2 j ω + 1 ⋅ 2 2 ( − j ω ) + 1 S_x (\omega) = \frac{4}{4\omega^2 + 1} = \frac{2}{2 j \omega +1} \cdot \frac{2}{2 (- j\omega) + 1 } Sx(ω)=4ω2+14=2jω+12⋅2(−jω)+12则系统的传函为
W ( j ω ) = 2 2 j ω + 1 W(j \omega) = \frac{2}{2 j \omega +1} W(jω)=2jω+12即
W ( s ) = 2 2 s + 1 W({\rm s}) = \frac{2}{2 {\rm s} +1} W(s)=2s+12
方差的定义式在统计动力学笔记(二)频谱密度与线性随机系统的动态准确性(自留用)一文的式(5)已给出:
D x = R x ( 0 ) = 1 2 π ∫ − ∞ ∞ S x ( ω ) d ω D_x = R_x (0) = \frac{1}{2\pi} \int_{-\infty} ^\infty S_x (\omega) {\rm d} \omega Dx=Rx(0)=2π1∫−∞∞Sx(ω)dω代入式(1)
D x = 1 2 π ∫ − ∞ ∞ W ( j ω ) W ( − j ω ) S u ( ω ) d ω = 1 2 π ∫ − ∞ ∞ ∣ W ( j ω ) ∣ 2 S u ( ω ) d ω (3) D_x = \frac{1}{2\pi} \int_{-\infty} ^\infty W(j \omega) W(-j \omega) S_u (\omega) {\rm d} \omega = \frac{1}{2\pi} \int_{-\infty} ^\infty \big\lvert W(j \omega) \big\rvert^2 S_u (\omega) {\rm d} \omega \tag{3} Dx=2π1∫−∞∞W(jω)W(−jω)Su(ω)dω=2π1∫−∞∞ W(jω) 2Su(ω)dω(3)式(3)的计算方式,有如下一套固定的方法,称为“ I n I_n In – 积分法”:
I n = 1 2 π ∫ − ∞ ∞ ∣ G ( j ω ) ∣ 2 ∣ H n ( j ω ) ∣ 2 d ω = 1 2 π ∫ − ∞ ∞ G n ( j ω ) H n ( j ω ) H n ( − j ω ) d ω (4) I_n = \frac{1}{2\pi} \int_{-\infty} ^\infty \frac{ \big\lvert G(j \omega) \big\rvert^2 }{ \big\lvert H_n(j \omega) \big\rvert^2 } {\rm d} \omega = \frac{1}{2\pi} \int_{-\infty} ^\infty \frac{ G_n (j \omega) }{ H_n(j \omega) H_n(-j \omega) } {\rm d} \omega \tag{4} In=2π1∫−∞∞ Hn(jω) 2 G(jω) 2dω=2π1∫−∞∞Hn(jω)Hn(−jω)Gn(jω)dω(4)其中
G n ( j ω ) = b 0 ( j ω ) 2 n − 2 + b 1 ( j ω ) 2 n − 4 + ⋯ + b n − 1 , H n ( j ω ) = a 0 ( j ω ) n + a 1 ( j ω ) n − 1 + ⋯ + a n (5) G_n (j \omega) = b_0 (j \omega)^{2n-2} + b_1 (j \omega)^{2n-4} + \cdots + b_{n-1}, \\ H_n (j \omega) = a_0 (j \omega)^{n} + a_1 (j \omega)^{n-1} + \cdots + a_n \tag{5} Gn(jω)=b0(jω)2n−2+b1(jω)2n−4+⋯+bn−1,Hn(jω)=a0(jω)n+a1(jω)n−1+⋯+an(5)关于式(4)(5)有如下几点:
(1)若积分式的分母阶数为 n n n,则实际系统中,分子的阶数不会超过 2 n − 2 2n-2 2n−2。
(2)积分式分母 H n ( j ω ) H n ( − j ω ) H_n(j \omega) H_n(-j \omega) Hn(jω)Hn(−jω)为 ω \omega ω的偶函数。
(3)积分式分子 G n ( j ω ) G_n(j \omega) Gn(jω)只含有 j ω j\omega jω的偶次幂。若出现了奇次幂,则可以直接忽视掉,因为积分后奇次幂将等于零。
(4)积分式分母中的 H n ( j ω ) H_n(j \omega) Hn(jω)应当是稳定的。
则对于 I n I_n In – 积分,其计算方法如下:
I n = ( − 1 ) n + 1 N n 2 a 0 D n (6) I_n = (-1) ^{n+1} \frac{N_n}{2a_0 D_n} \tag{6} In=(−1)n+12a0DnNn(6)其中
D n = ∣ a 1 a 0 0 ⋯ 0 a 3 a 2 a 1 ⋯ 0 a 5 a 4 a 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ 0 0 0 ⋯ a n ∣ , (7) D_n = \begin{vmatrix} a_1 & a_0 & 0 & \cdots & 0 \\ a_3 & a_2 & a_1 & \cdots & 0 \\ a_5 & a_4 & a_3 & \cdots & 0 \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ 0 & 0 & 0 & \cdots & a_n \end{vmatrix} \tag{7}, Dn= a1a3a5⋮0a0a2a4⋮00a1a3⋮0⋯⋯⋯⋱⋯000⋮an ,(7) N n = ∣ b 0 a 0 0 ⋯ 0 b 1 a 2 a 1 ⋯ 0 b 2 a 4 a 3 ⋯ 0 ⋮ ⋮ ⋮ ⋱ ⋮ b n − 1 0 0 ⋯ a n ∣ (8) N_n = \begin{vmatrix} b_0 & a_0 & 0 & \cdots & 0 \\ b_1 & a_2 & a_1 & \cdots & 0 \\ b_2 & a_4 & a_3 & \cdots & 0 \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ b_{n-1} & 0 & 0 & \cdots & a_n \end{vmatrix} \tag{8} Nn= b0b1b2⋮bn−1a0a2a4⋮00a1a3⋮0⋯⋯⋯⋱⋯000⋮an (8) N n N_n Nn只是把 D n D_n Dn中的第一列替换成了 b i b_i bi。
例:设系统的传递函数为
W ( s ) = K T s + 1 W({\rm s}) = \frac{K}{T {\rm s} +1} W(s)=Ts+1K输入信号的频谱密度为
S u ( ω ) = D u α 2 + ω 2 S_u (\omega) = \frac{D_u}{\alpha^2 + \omega^2} Su(ω)=α2+ω2Du计算该系统的均方差。
首先得到系统误差的传函:
Φ e ( s ) = 1 1 + W ( s ) = T s + 1 T s + 1 + K \Phi_e ({\rm s}) = \frac{1}{1 + W( {\rm s})} = \frac{T{\rm s} +1}{T{\rm s} + 1 + K} Φe(s)=1+W(s)1=Ts+1+KTs+1代入式(1)计算误差的频谱密度
S e ( ω ) = ∣ Φ e ( j ω ) ∣ 2 S u ( ω ) = ∣ T ( j ω ) + 1 T ( j ω ) + 1 + K ∣ 2 D u α 2 + ω 2 = D u ( T 2 ω 2 + 1 ) ∣ ( T ( j ω ) + 1 + K ) ( j ω + α ) ∣ 2 = D u ( T 2 ω 2 + 1 ) ∣ T ( j ω ) 2 + ( α T + 1 + K ) j ω + ( 1 + K ) α ∣ 2 \begin{aligned} S_e (\omega) &= \left| \Phi_e (j \omega) \right|^2 S_u (\omega) = \left| \frac{T(j \omega) +1}{T(j \omega) + 1 + K} \right|^2 \frac{D_u}{\alpha^2 + \omega^2} \\ &= \frac{D_u \left( T^2 \omega^2 + 1\right) }{ \left| \left( T( j\omega) + 1 + K \right) \left( j\omega + \alpha \right) \right|^2 } \\ &= \frac{D_u \left( T^2 \omega^2 + 1 \right) }{ \left| T(j\omega)^2 + \left( \alpha T + 1 +K \right) j\omega + (1 + K) \alpha \right|^2 } \end{aligned} Se(ω)=∣Φe(jω)∣2Su(ω)= T(jω)+1+KT(jω)+1 2α2+ω2Du=∣(T(jω)+1+K)(jω+α)∣2Du(T2ω2+1)=∣T(jω)2+(αT+1+K)jω+(1+K)α∣2Du(T2ω2+1)均方差为(类比统计动力学笔记(二)频谱密度与线性随机系统的动态准确性(自留用)一文式(5)):
e 2 ‾ = 1 2 π ∫ − ∞ ∞ S e ( ω ) d ω = D u I 2 \overline{e^2} = \frac{1}{2\pi} \int_{-\infty} ^\infty S_e (\omega) {\rm d} \omega = D_u I_2 e2=2π1∫−∞∞Se(ω)dω=DuI2其中
I 2 = 1 2 π ∫ − ∞ ∞ ( T 2 ω 2 + 1 ) d ω ∣ T ( j ω ) 2 + ( α T + 1 + K ) j ω + ( 1 + K ) α ∣ 2 I_2 = \frac{1}{2\pi} \int_{-\infty} ^\infty \frac{ \left( T^2 \omega^2 + 1 \right) {\rm d} \omega }{ \left| T(j\omega)^2 + \left( \alpha T + 1 +K \right) j\omega + (1 + K) \alpha \right|^2 } I2=2π1∫−∞∞∣T(jω)2+(αT+1+K)jω+(1+K)α∣2(T2ω2+1)dω可见
G 2 ( j ω ) = T 2 ⏟ b 0 ω 2 + 1 ⏟ b 1 , G_2 (j\omega) = \underbrace{T^2}_{b_0} \omega^2 + \underbrace{1}_{b_1}, G2(jω)=b0 T2ω2+b1 1, H 2 ( j ω ) = T ⏟ a 0 ( j ω ) 2 + ( α T + 1 + K ) ⏟ a 1 j ω + ( 1 + K ) α ⏟ a 2 H_2 (j\omega) = \underbrace{T}_{a_0} (j\omega)^2 + \underbrace{\left( \alpha T + 1 +K \right)}_{a_1} j\omega + \underbrace{(1 + K) \alpha}_{a_2} H2(jω)=a0 T(jω)2+a1 (αT+1+K)jω+a2 (1+K)α计算两个行列式
D 2 = ∣ a 1 a 0 a 3 a 2 ∣ = ∣ α T + 1 + K T 0 ( 1 + K ) α ∣ = α ( α T + 1 + K ) ( 1 + K ) , D_2 = \begin{vmatrix} a_1 & a_0 \\ a_3 & a_2 \end{vmatrix} = \begin{vmatrix} \alpha T + 1 +K & T \\ 0 & (1 + K) \alpha \end{vmatrix} = \alpha \left( \alpha T + 1 +K \right) (1 + K), D2= a1a3a0a2 = αT+1+K0T(1+K)α =α(αT+1+K)(1+K), N 2 = ∣ b 0 a 0 b 1 a 2 ∣ = ∣ T 2 T 1 ( 1 + K ) α ∣ = α T 2 ( 1 + K ) − T N_2 = \begin{vmatrix} b_0 & a_0 \\ b_1 & a_2 \end{vmatrix} = \begin{vmatrix} T^2 & T \\ 1 & (1 + K) \alpha \end{vmatrix} = \alpha T^2 (1 + K) - T N2= b0b1a0a2 = T21T(1+K)α =αT2(1+K)−T故
I 2 = ( − 1 ) 2 + 1 N 2 2 a 0 D 2 = − α T 2 ( 1 + K ) − T 2 T α ( α T + 1 + K ) ( 1 + K ) I_2 = (-1) ^{2+1} \frac{N_2}{2a_0 D_2} = - \frac{ \alpha T^2 (1 + K) - T }{2 T \alpha \left( \alpha T + 1 +K \right) (1 + K) } I2=(−1)2+12a0D2N2=−2Tα(αT+1+K)(1+K)αT2(1+K)−T则
e 2 ‾ = D u I 2 = D u [ T − α T 2 ( 1 + K ) ] 2 T α ( α T + 1 + K ) ( 1 + K ) = D u [ 1 − α T ( 1 + K ) ] 2 α ( α T + 1 + K ) ( 1 + K ) \overline{e^2} = D_u I_2 = \frac{ D_u \left[ T - \alpha T^2 (1 + K) \right] }{2 T \alpha \left( \alpha T + 1 +K \right) (1 + K) } = \frac{ D_u \left[ 1 - \alpha T (1 + K) \right] }{2 \alpha \left( \alpha T + 1 +K \right) (1 + K) } e2=DuI2=2Tα(αT+1+K)(1+K)Du[T−αT2(1+K)]=2α(αT+1+K)(1+K)Du[1−αT(1+K)]故均方差为
e 2 ‾ = D u [ 1 − α T ( 1 + K ) ] 2 α ( α T + 1 + K ) ( 1 + K ) \sqrt{\overline{e^2}} = \sqrt{ \frac{ D_u \left[ 1 - \alpha T (1 + K) \right] }{2 \alpha \left( \alpha T + 1 +K \right) (1 + K) } } e2=2α(αT+1+K)(1+K)Du[1−αT(1+K)]