通信基础-常用的一些分布

CSCG: Circularly Symmetric Complex zero-mean white Gaussian noise 循环对称复高斯噪声

CSCG:循环对称复高斯

数学定义: \[ \textbf{Z}\sim\mathcal{CN}(0,1)\to \Re{\textbf{Z}} \perp \!\!\! \perp\Im{\textbf{Z}} \\and\\ \Re{\textbf{Z}}\sim \mathcal{N}(0,1/2)\quad \Im{\textbf{Z}}\sim\mathcal{N}(0,1/2) \]

含义

  • Circularly: means the variance of the real and imaginary parts are equal.
  • Gaussian: means the probability distribution of the amplitudes of the noise samples is Gaussian

Matlab

1
(randn(m,n)+1i*randn(m,n))*sigma/sqrt(2)

其中sigmal是方差

正态分布的再生性

设随机比昂量\(X\)\(Y\)相互独立且分别服从正态分布\(N\sim(\mu_1,\sigma_1^2)\)\(N\sim(\mu_2,\sigma_2^2)\),则\(Z=X+Y\)服从正态分布\(N\sim(\mu_1+\mu_2,\sigma_1^2+\sigma_2^2)\)

Rayleigh分布

复高斯分布的模服从瑞利分布: \[ f(x)=\frac{x}{\sigma^{2}} e^{-\frac{x^{2}}{2 \sigma^{2}}}, x>0 \]

Rayleigh分布和复高斯分布的关系

如果\(\textbf{A}\sim\mathcal{CN}(0,\sigma^2)\),即\(\Re{\textbf{A}}\sim\mathcal{N}(0, \sigma^2/2)\)\(\Im{\textbf{A}}\sim\mathcal{N}(0, \sigma^2/2)\),则\(|\textbf{A}|\sim\)参数为\(\sigma^2/2\)的瑞利分布: \[ f_{|\textbf{A}|}(x)=\frac{x}{\sigma^{2}/2} e^{-\frac{x^{2}}{\sigma^{2}}}, x>0 \]

simulation

因为没有找到Rayleigh分布和复高斯分布的关系。使用matlab进行仿真验证。

untitled

非中心卡方分布

非中心卡方分布由若干独立同方差的均值不全为0的高斯随机变量平方和得到。

image-20220720161758149

值得注意的是,例如OneL文章中以及其他参考文献中所述,常常写为\(\mathcal{X}^2(a,b)\),其中,\(a\)为自由度,\(b\)为参数\(\lambda/\sigma^2\)

所以,此时有期望\(E=a+b\)(事先归一化了方差,所以这里比上式少了一个\(\sigma^2\)

方差\(V=2(a+2b)\)(事先归一化了方差,所以这里比上式少了一个\(\sigma^4\)


参考书目:《数字通信》P46

非中心卡方分布的PDF为: \[ p(x)= \begin{cases}\frac{1}{2 \sigma^2}\left(\frac{x}{s^2}\right)^{\frac{n-2}{4}} e^{-\frac{s^2+x}{2 \sigma^2}} I_{\frac{n}{2}-1}\left(\frac{s}{\sigma^2} \sqrt{x}\right) & x>0 \\ 0 & \text { otherwise }\end{cases} \] 其中: $$ \[\begin{align} &s = \sqrt{\sum\limits_{i=1}^nm_i^2}\\ &I_\alpha(x)=\sum_{k=0}^{\infty} \frac{(x / 2)^{\alpha+2 k}}{k ! \Gamma(\alpha+k+1)}, \quad x \geq 0 \end{align}\] $$ \(I_\alpha(x)\)modify Bessel function of the first kind and order \(\alpha\)\(\Gamma(x)\)是gamma function。

对于\(x>1\),有常用近似\(I_0(x)\approx\frac{e^x}{\sqrt{2\pi x}}\)

其CDF为: \[ F(x)= \begin{cases}1-Q_m\left(\frac{s}{\sigma}, \frac{\sqrt{x}}{\sigma}\right) & x>0 \\ 0 & \text { otherwise }\end{cases} \] 其中\(2m=n\)\(Q_m(a,b)\)generalized Marcum Q function.

matlab实现

generalized Marcum Q function 和modify bessel function of the first kind and order \(\alpha\)可以直接用matlab函数:

1
2
marcumq(a,b,m) % calculate the generilized Marcum Q function
besseli(alpha,x) % calculate the modify bessel function of the first kind and order alpha

CDF和PDF推导

例如之前遇到的典型问题,三个随机变量分别服从参数为\(\lambda_1\),\(\lambda_2\),\(\lambda_1\)的指数分布,记为\(\Gamma_1\)\(\Gamma_2\)\(\Gamma_3\),即: \[ \Gamma_1\sim\mathbf{E}(\lambda_1)\\ \Gamma_2\sim\mathbf{E}(\lambda_2)\\ \Gamma_3\sim\mathbf{E}(\lambda_3)\\ \] 则其和\(\Gamma=\Gamma_1+\Gamma_2+\Gamma_3\)的CDF为: \[ \mathrm{F}_\Gamma(\gamma)=\text{Pr}(\Gamma\leq \gamma)=\text{Pr}(\Gamma_1+\Gamma_2+\Gamma_3\leq \gamma) \]\(\Gamma^\S=\Gamma_1+\Gamma_2\),则上式变为: \[ \begin{align} &\text{Pr}(\Gamma_1+\Gamma_2+\Gamma_3\leq \gamma)=\text{Pr}(\Gamma^\S+\Gamma_3\leq \gamma)\\ =&\text{Pr}(\Gamma^\S\leq \gamma-\Gamma_3)\\ =&\mathrm{F}_{\Gamma^\S}(\gamma-\Gamma_3)\\ =&\int^\gamma_0\mathrm{F}_{\Gamma^\S}(\gamma-\gamma_3)\mathrm{f}_{\Gamma_3}(\gamma_3)d\gamma_3 \end{align} \] 因为: \[ \begin{align} &\mathrm{F}_{\Gamma^\S}(\gamma^\S)=\text{Pr}(\Gamma^\S\leq\gamma^\S)\\ =&\text{Pr}(\Gamma_1+\Gamma_2\leq\gamma^\S)\\ =&\text{Pr}(\Gamma_1\leq\gamma^\S-\Gamma_2)\\ =&\int^{\gamma^\S}_0\mathrm{F}_{\gamma_1}(\gamma^\S-\gamma_2)\mathrm{f}_{\Gamma_2}(\gamma_2)d\gamma_2\\ =&\int^{\gamma^\S}_0(1-e^{-\lambda_1(\gamma^\S-\gamma_2)})\lambda_2e^{-\lambda_2\gamma_2}d\gamma_2\\ =&-\frac{\lambda_1}{\lambda_1-\lambda_2}e^{-\lambda_2\gamma^\S}+1+\frac{\lambda_2}{\lambda_1-\lambda_2}e^{-\lambda_1\gamma^\S} \end{align} \] 所以: \[ \begin{align} &\text{Pr}(\Gamma_1+\Gamma_2+\Gamma_3\leq \gamma)\\ =&\int^{\gamma}_0\left(\frac{-\lambda_1}{\lambda_1-\lambda_2}e^{-\lambda_2(\gamma-\gamma_3)} +1+\frac{\lambda_2}{\lambda_1-\lambda_2}e^{-\lambda_1(\gamma-\gamma_3)}\right)\lambda_3e^{-\lambda_3\gamma_3}d\gamma_3\\ =&e^{-\lambda_3\gamma}c_3+e^{-\lambda_2\gamma}c_2+e^{-\lambda_1\gamma}c_1+1 \end{align} \] 其中,有: \[ \begin{align} &c_3=\frac{\lambda_2\lambda_3}{(\lambda_1-\lambda_2)(\lambda_1-\lambda_3)}-\frac{\lambda_1\lambda_3}{(\lambda_1-\lambda_2)(\lambda_2-\lambda_3)}-1\\ &c_2=\frac{\lambda_1\lambda_3}{(\lambda_1-\lambda_2)(\lambda_2-\lambda_3)}\\ &c_1=-\frac{\lambda_2\lambda_3}{(\lambda_1-\lambda_2)(\lambda_1-\lambda_3)} \end{align} \]

其PDF为: \[ -\lambda_3e^{-\lambda_3\gamma}c_3-\lambda_2e^{-\lambda_2\gamma}c_2-\lambda_1e^{-\lambda_1\gamma}c_1 \]

高斯分布的任意高阶矩

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
clc;clear;close all;

% 阶数为n
for n=2:2:16
% n=6;
k=n/2;
% s代表sigma 没有平方!!!!
syms s
if rem(k, 1) ~=0
disp('Error: make sure n is even plz')
else
disp(n)
moment = factorial(2*k)*s^(2*k)/(2^k*factorial(k))
end
end

image-20221116100727821

参考Digital Communications-Fifth EditionJohn G. Proakis


通信基础-常用的一些分布
https://lcjoffrey.top/2022/04/11/CSCG/
作者
Joffrey
发布于
2022年4月11日
更新于
2022年11月16日
许可协议