正态分布
标准正态分布:
\[x \backsim N(0,1):f(x)=\frac{1}{\sqrt{2\pi}}e^{\frac{-x^2}{2}}
\]
一般正态分布:
\[x \backsim N(\mu,\sigma^2):f(x)=\frac{1}{\sqrt{2\pi\sigma^2}}e^{\frac{-(x-\mu)^2}{2\sigma^2}}
\]
多维正态分布:设\(\boldsymbol{x}=[x_1,x_2,...,x_N]^T\)是一个\(n\)维向量,第\(n\)的维度上的随机变量\(x_n\)服从均值为\(\mu_n\),方差为\(\sigma^2_n\)的高斯分布,即\(x_n \backsim N(\mu_n,\sigma^2_n)\),且各个维度相互独立,则其联合概率密度等于各个维度上概率密度的乘积,所以
\[f(\boldsymbol{x})=\frac{1}{\sqrt{2\pi\sigma^2_1}}e^{\frac{-(x_1-\mu_1)^2}{2\sigma^2_1}} \frac{1}{\sqrt{2\pi\sigma^2_2}}e^{\frac{-(x_2-\mu_2)^2}{2\sigma^2_2}}...\frac{1}{\sqrt{2\pi\sigma^2_N}}e^{\frac{-(x_N-\mu_N)^2}{2\sigma^2_N}} \tag{1}
\]
令\(\boldsymbol{\Sigma}={\rm diag}(\sigma^2_n)\),\(\boldsymbol{\mu}=[\mu_1,...,\mu_N]^T\),则\(\boldsymbol{x}-\boldsymbol{\mu}=[x_1-\mu_1,...,x_N-\mu_N]^T\),则有
\[\sqrt{2\pi\sigma^2_1}\sqrt{2\pi\sigma^2_1}...\sqrt{2\pi\sigma^2_N}=(2\pi)^{N/2}[{\rm det}(\boldsymbol{\Sigma})]^{1/2} \tag{2}
\]
\[\frac{-(x_1-\mu_1)^2}{2\sigma^2_1}+...+\frac{-(x_N-\mu_N)^2}{2\sigma^2_N}=-(\boldsymbol{x}-\boldsymbol{\mu})^T \boldsymbol{\Sigma}^{-1}(\boldsymbol{x}-\boldsymbol{\mu}) \tag{3}
\]
将\((2)(3)\)代入\((1)\)可以得到
\[f(\boldsymbol{x})=\frac{1}{(2\pi)^{N/2}[{\rm det}(\boldsymbol{\Sigma})]^{1/2}}e^{-(\boldsymbol{x}-\boldsymbol{\mu})^T \boldsymbol{\Sigma}^{-1}(\boldsymbol{x}-\boldsymbol{\mu})} \tag{4}
\]
复高斯分布:设\(X \backsim N(\mu_x,\sigma^2_x)\),\(Y \backsim N(\mu_y,\sigma^2_y)\),若\(X\)与\(Y\)统计独立,则其联合概率密度函数可表示为:
\[f_{XY}(x,y)=\frac{1}{\sqrt{2\pi\sigma^2_x}}e^{\frac{-(x-\mu_x)^2}{2\sigma^2_x}} \frac{1}{\sqrt{2\pi\sigma^2_y}}e^{\frac{-(y-\mu_y)^2}{2\sigma^2_y}} \tag{5}
\]
令\(\mu_x=\mu_y\),\(\sigma=\sigma^2_x=\sigma^2_y\),则上式就可以化简为
\[f_{XY}(x,y)=\frac{1}{2\pi\sigma^2}e^{-\frac{(x-\mu)^2+(y-\mu)^2}{2\sigma^2}} \tag{6}
\]
对应的复随机变量\(Z=X+iY\)称为复高斯随机变量。\(Z\)的均值\(\mu_z\)和方差\(\sigma_z\)分别为:
\[\mu_z={\rm E}[Z]={\rm E}[X+iY]={\rm E}[X]+i{\rm E}[Y]=\mu_x+i\mu_y=\mu+i\mu
\]
\[\sigma_z={\rm E}[(Z-\mu_z)(Z-\mu_z)']={\rm E}[|(X-\mu_x)+i(Y-\mu_y)|^2]={\rm E}[(X-\mu_x)^2]+{\rm E}[(Y-\mu_y)^2]=\sigma^2_x+\sigma^2_y=2\sigma^2
\]
将\(\mu_z\)和\(\sigma_z\)代入\((6)\)可以得到\(Z\)的概率密度函数为
\[f_Z(z)=\frac{1}{\pi\sigma_z^2}e^{-\frac{(z-\mu_z)^2}{\sigma_z^2}} \tag{7}
\]