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）

1. 频谱密度

频谱密度是对相关函数$R(t)$的傅里叶变换：
$$S(\omega )={\int }_{-\infty }^{\infty }R(\tau )e^{-j\omega \tau }\mathrm{d}\tau \text{\hspace{1em}(1)}$$显然，相关函数即为频谱密度的傅里叶逆变换：
$$R(\tau )=\frac{1}{2\pi }{\int }_{-\infty }^{\infty }S(\omega )e^{j\omega \tau }\mathrm{d}\omega \text{\hspace{1em}(2)}$$替换$e^{-j\omega \tau }=\cos \omega \tau -j\sin \omega \tau$，并注意到奇函数$\sin \omega \tau$从$-\infty$到$+\infty$的积分为0，而偶函数$\cos \omega \tau$的积分对称，则式(1)可以化为：
$$S(\omega )={\int }_{-\infty }^{\infty }R(\tau )\cos \omega \tau \mathrm{d}\tau -j{\int }_{-\infty }^{\infty }R(\tau )\sin \omega \tau \mathrm{d}\tau =2{\int }_0^{\infty }R(\tau )\cos \omega \tau \mathrm{d}\tau \text{\hspace{1em}(3)}$$由此可见，频谱密度是偶函数：
$$\begin{gathered} S(-\omega )=2{\int }_0^{\infty }R(\tau )\cos (-\omega \tau )\mathrm{d}\tau =S(\omega ), \\ R(\tau )=\frac{1}{\pi }{\int }_0^{\infty }S(\omega )\cos \omega \tau \mathrm{d}\omega \text{\hspace{1em}(4)} \end{gathered}$$进而均方差为：
$$\overline{x^2}=D_x=R_x(0)=\frac{1}{2\pi }{\int }_{-\infty }^{\infty }S(\omega )e^{j\omega \tau }\mathrm{d}\omega {|}_{\tau =0}=\frac{1}{\pi }{\int }_0^{\infty }S(\omega )\mathrm{d}\omega \text{\hspace{1em}(5)}$$在物理意义上，$S(\omega )$表示了不同频率$\omega$下的功率分布，而其积分表示了平均功率。

复习一下$R(\tau )$的公式：
$$R(\tau )=\underset{T\to \infty }{\lim }\frac{1}{2T}{\int }_{-T}^{T}x(t)x(t+\tau )\mathrm{d}t$$代入定义式(1)：
$$S(\omega )={\int }_{-\infty }^{\infty }R(\tau )e^{-j\omega \tau }\mathrm{d}\tau =S(\omega )={\int }_{-\infty }^{\infty }{e}^{-j\omega \tau }\mathrm{d}\tau \underset{T\to \infty }{\lim }\frac{1}{2T}{\int }_{-T}^{T}x(t)x(t+\tau )\mathrm{d}t$$设$\theta =t+\tau$，则$\tau =\theta -t,\mathrm{d}\tau =\mathrm{d}\theta$，于是
$$\begin{array}{rl}S(\omega )& ={\int }_{-\infty }^{\infty }{e}^{-j\omega \tau }\mathrm{d}\tau \underset{T\to \infty }{\lim }\frac{1}{2T}{\int }_{-T}^{T}x(t)x(t+\tau )\mathrm{d}t\\ & ={\int }_{-\infty }^{\infty }{e}^{-j\omega \theta }{e}^{j\omega t}\mathrm{d}\theta \underset{T\to \infty }{\lim }\frac{1}{2T}{\int }_{-T}^{T}x(t)x(\theta )\mathrm{d}t\\ & ={\int }_{-\infty }^{\infty }x(\theta )e^{-j\omega \theta }\mathrm{d}\theta \underset{T\to \infty }{\lim }\frac{1}{2T}{\int }_{-T}^{T}x(t)e^{j\omega t}\mathrm{d}t\\ & =\underset{T\to \infty }{\lim }\frac{1}{2T}\left[{\int }_{-\infty }^{\infty }x(\theta )e^{-j\omega \theta }\mathrm{d}\theta {\int }_{-T}^{T}x(t)e^{j\omega t}\mathrm{d}t\right]\\ & =\underset{T\to \infty }{\lim }\frac{1}{2T}X(j\omega )X\left(-j\omega \right)\end{array}$$即：
$$S(\omega )=\underset{T\to \infty }{\lim }\frac{1}{2T}{\left|X(j\omega )\right|}^2\text{\hspace{1em}(6)}$$

例：设相关函数$R(\tau )=e^{-\alpha |\tau |}$，其中$\alpha >0$。则
$$\begin{array}{rl}{S}_x(\omega )& ={\int }_{-\infty }^{\infty }{e}^{-\alpha |\tau |}\mathrm{d}\tau \\ & ={\int }_{-\infty }^0{e}^{\alpha \tau }{e}^{-j\omega \tau }\mathrm{d}\tau +{\int }_0^{\infty }{e}^{-\alpha \tau }{e}^{-j\omega \tau }\mathrm{d}\tau \\ & ={\int }_{-\infty }^0{e}^{\left(\alpha -j\omega \right)\tau }\mathrm{d}\tau +{\int }_0^{\infty }{e}^{-\left(\alpha +j\omega \right)\tau }\mathrm{d}\tau \\ & =\frac{1}{\alpha -j\omega }{e}^{\alpha -j\omega }{|}_{-\infty }^0+\frac{-1}{\alpha +j\omega }{e}^{\alpha -j\omega }{|}_0^{\infty }\\ & =\frac{1}{\alpha -j\omega }+\frac{1}{\alpha +j\omega }\\ & =\frac{2\alpha }{{\alpha }^2+{\omega }^2}\end{array}$$

根据定义式(1)，很显然可以得到互频谱密度的表达式：
$$S_{xu}(\omega )={\int }_{-\infty }^{\infty }{R}_{xu}(\tau )e^{-j\omega \tau }\mathrm{d}\tau \text{\hspace{1em}(7)}$$及
$$R_{xu}(\tau )=\frac{1}{2\pi }{\int }_{-\infty }^{\infty }{S}_{xu}(\omega )e^{j\omega \tau }\mathrm{d}\omega \text{\hspace{1em}(8)}$$

2. 线性系统输出端的随机信号的频谱密度

在统计动力学笔记（一）动态系统随机信号在时域中的变换（自留用）一文的公式(3)中给出了输出端$x$的相关函数计算公式：
$$R_x(\tau )={\int }_{-\infty }^{\infty }K(\lambda )\mathrm{d}\lambda {\int }_{-\infty }^{\infty }{R}_u(\tau +\lambda -\eta )K(\eta )\mathrm{d}\eta$$则根据本文定义式(1)，输出端$x$的频谱密度为
$$\begin{array}{rl}{S}_x(\omega )& ={\int }_{-\infty }^{\infty }{e}^{-j\omega \tau }{R}_x(\tau )\mathrm{d}\tau \\ & ={\int }_{-\infty }^{\infty }{e}^{-j\omega \tau }\mathrm{d}\tau {\int }_{-\infty }^{\infty }K(\lambda )\mathrm{d}\lambda {\int }_{-\infty }^{\infty }{R}_u(\tau +\lambda -\eta )K(\eta )\mathrm{d}\eta \end{array}$$令$\xi =\tau +\lambda -\eta$，则$\mathrm{d}\tau =\mathrm{d}\xi$，上式为
$$\begin{array}{rl}{S}_x(\omega )& ={\int }_{-\infty }^{\infty }{e}^{-j\omega \tau }\mathrm{d}\tau {\int }_{-\infty }^{\infty }K(\lambda )\mathrm{d}\lambda {\int }_{-\infty }^{\infty }{R}_u(\tau +\lambda -\eta )K(\eta )\mathrm{d}\eta \\ & ={\int }_{-\infty }^{\infty }{e}^{-j\omega \xi }{e}^{-j\omega \eta }{e}^{j\omega \lambda }\mathrm{d}\xi {\int }_{-\infty }^{\infty }K(\lambda )\mathrm{d}\lambda {\int }_{-\infty }^{\infty }{R}_u(\xi )K(\eta )\mathrm{d}\eta \\ & =\underset{W(j\omega )}{\underbrace{{\int }_{-\infty }^{\infty }{e}^{-j\omega \eta }K(\eta )\mathrm{d}\eta }}\underset{W(-j\omega )}{\underbrace{{\int }_{-\infty }^{\infty }{e}^{j\omega \lambda }K(\lambda )\mathrm{d}\lambda }}\underset{S_u(\omega )}{\underbrace{{\int }_{-\infty }^{\infty }{e}^{-j\omega \xi }{R}_u(\xi )\mathrm{d}\xi }}\\ & =W(j\omega )W(-j\omega )S_u(\omega )\\ & =|W(j\omega ){|}^2{S}_u(\omega )\end{array}$$即
$$S_x(\omega )=|W(j\omega ){|}^2{S}_u(\omega )\text{\hspace{1em}(9)}$$同理可以推出互频谱密度
$$S_{xu}(\omega )=W(j\omega )S_u(\omega )\text{\hspace{1em}(10)}$$
显然，对于强度$N^2=1$的白噪声来说，其$S_u(\omega )=1$，则$S_{xu}(\omega )=W(j\omega ),S_x(\omega )=|W(j\omega ){|}^2$，由此可见，可以在输入端加载强度为1的白噪声，通过测量输出端的频谱密度$S_x(\omega )$，即可获得系统的传递函数$W(j\omega )$，具有很强的应用价值。