随机过程是所有样本函数的集合,从另外的角度看随机过程是随机变量的延伸,是在任意时刻的值是一个随机变量。也可称随机过程看成时间进程中处于不同时刻的随机变量的集合。
设 ξ ( t ) \xi(t) ξ(t)为一个随机过程,它在任意时刻 t 1 t_1 t1的值 ξ ( t 1 ) \xi(t_1) ξ(t1)是一个随机变量,我们将 ξ ( t 1 ) \xi(t_1) ξ(t1)小于或等于某一数值 x 1 x_1 x1的概论记作:
F 1 ( x 1 , t 1 ) = P { ξ ( t 1 ) ≤ x 1 } F_1(x_1,t_1)=P\lbrace \xi(t_1)\leq x_1\rbrace F1(x1,t1)=P{ξ(t1)≤x1}
则将:
f 1 ( x 1 , t 1 ) = ∂ F 1 ( x 1 , t 1 ) ∂ x 1 f_1(x_1,t_1)=\dfrac {\partial F_1(x_1,t_1)} {\partial x_1} f1(x1,t1)=∂x1∂F1(x1,t1)
称为一维概率密度函数,同样的我们可以将上列推广到n维:
F n ( x 1 , x 2 , ⋯ , x n ; t 1 , t 2 ⋯ , t n ) = P { ξ ( t 1 ) ≤ x 1 ξ ( t 2 ) ≤ x 2 ⋯ , ξ ( t n ) ≤ x n } f n ( x 1 , x 2 , ⋯ , x n ; t 1 , t 2 ⋯ , t n ) = ∂ n F n ( x 1 , x 2 , ⋯ , x n ; t 1 , t 2 ⋯ , t n ) ∂ x 1 ∂ x 2 ⋯ ∂ x n F_n(x_1,x_2,\cdots,x_n;t_1,t_2\cdots,t_n)= P\lbrace \xi(t_1)\leq x_1 \xi(t_2)\leq x_2 \cdots, \xi(t_n)\leq x_n \rbrace \\ \\ f_n(x_1,x_2,\cdots,x_n;t_1,t_2\cdots,t_n)= \dfrac {\partial^n F_n(x_1,x_2,\cdots,x_n;t_1,t_2\cdots,t_n)} {\partial x_1 \partial x_2 \cdots \partial x_n} Fn(x1,x2,⋯,xn;t1,t2⋯,tn)=P{ξ(t1)≤x1ξ(t2)≤x2⋯,ξ(tn)≤xn}fn(x1,x2,⋯,xn;t1,t2⋯,tn)=∂x1∂x2⋯∂xn∂nFn(x1,x2,⋯,xn;t1,t2⋯,tn)
n越大,则对随机过程的描述越充分
均值
随机过程的均值(数学期望)定义为:
a ( t ) = E [ ξ ( t ) ] = ∫ − ∞ ∞ x f 1 ( x , t ) d t a(t)=E[\xi(t)]=\int_{-\infty}^\infty xf_1(x,t)dt a(t)=E[ξ(t)]=∫−∞∞xf1(x,t)dt
表示随机过程n个样本函数曲线的摆动中心
方差
σ 2 ( t ) = D [ ξ ( t ) ] = E { [ ξ ( t ) − a ( t ) ] 2 } \sigma^2(t)= D[ \xi(t) ]=E\lbrace [ \xi(t)-a(t) ]^2\rbrace σ2(t)=D[ξ(t)]=E{[ξ(t)−a(t)]2}
协方差
B ( t 1 , t 2 ) = E { [ ξ ( t 1 ) − a ( t 1 ) ] [ ξ ( t 2 ) − a ( t 2 ) ] } = R ( t 1 , t 2 ) − a ( t 1 ) a ( t 2 ) \begin{aligned} B(t_1,t_2)&=E \lbrace [ \xi(t_1)-a(t_1) ] [ \xi(t_2)-a(t_2) ] \rbrace \\ &=R(t_1,t_2) -a(t_1)a(t_2) \end{aligned} B(t1,t2)=E{[ξ(t1)−a(t1)][ξ(t2)−a(t2)]}=R(t1,t2)−a(t1)a(t2)
相关函数
R ( t 1 , t 2 ) = E [ ξ ( t 1 ) ξ ( t 2 ) ] R(t_1,t_2)=E[\xi(t_1)\xi(t_2)] R(t1,t2)=E[ξ(t1)ξ(t2)]
如果一个随机过程的统计特性与时间起点无关,称为严平稳(狭义平稳)
但我们研究的一般为宽平稳(广义平稳),它的一维概率密度函数与t无关,二维概率密度只和时间间隔有关,即:
{ E [ ξ ( t ) ] = a R ( t 1 , t 2 ) = R ( τ ) \begin{cases} E[\xi(t)] &= a \\ R(t_1,t_2)&=R(\tau) \end{cases} {E[ξ(t)]R(t1,t2)=a=R(τ)
随机过程任意一次都经历了随机过程的所有可能状态,满足各态历经一定是平稳过程,其时间均值和时间相关函数分别定义为:
{ a ‾ = lim T → ∞ 1 T ∫ − T / 2 T / 2 X ( T ) d t R ( τ ) ‾ = lim T → ∞ ∫ − T / 2 T / 2 x ( t ) x ( t + τ ) d t \left\{ \begin{aligned} \overline{a} &= \lim_{T\to\infty} \dfrac{1}{T} \int_{-T/2}^{T/2} X(T)dt \\ \\ \overline{R(\tau)} &=\lim_{T\to\infty} \int_{-T/2}^{T/2} x(t)x(t+\tau)dt \end{aligned} \right. ⎩⎪⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎪⎧aR(τ)=T→∞limT1∫−T/2T/2X(T)dt=T→∞lim∫−T/2T/2x(t)x(t+τ)dt
如果平稳过程满足下列条件,则各态历经:
{ a = a ‾ R ( τ ) = R ( τ ) ‾ \left\{ \begin{aligned} a &= \overline{a} \\ R(\tau) &= \overline{R(\tau)} \end{aligned} \right. {aR(τ)=a=R(τ)
我们知道稳态随机过程的自相关函数为:
R ( τ ) = E [ ξ ( t ) ξ ( t + τ ) ] R(\tau)=E[\xi(t)\xi(t+\tau)] R(τ)=E[ξ(t)ξ(t+τ)]
具有以下性质:
对于确定功率信号,功率谱密度定义为:
P X ( f ) = lim T → ∞ ∣ X T ( f ) ∣ 2 T P_X(f)=\lim_{T\to\infty}\dfrac{|X_T(f)|^2}{T} PX(f)=T→∞limT∣XT(f)∣2
对于随机信号而言,每一个样本对应一个功率谱密度,所以应该取所有样本得统计平均,即:
P ξ ( f ) = lim T → ∞ E ∣ X T ( f ) ∣ 2 T P_\xi(f)=\lim_{T\to\infty}\dfrac{E|X_T(f)|^2}{T} Pξ(f)=T→∞limTE∣XT(f)∣2
与确知信号相同,平稳随机信号自相关函数和功率谱密度也互为一对傅里叶变换关系,也称为维纳——辛钦定理:
{ P ξ ( ω ) = ∫ − ∞ ∞ R ( τ ) e − j ω τ d τ R ( τ ) = ∫ − ∞ ∞ 1 2 π P ξ ( ω ) e j ω τ d ω \left\{ \begin{aligned} P_\xi(\omega)&=\int_{-\infty}^{\infty} R(\tau)e^{-j\omega\tau}d\tau \\ \\ R(\tau)&=\int_{-\infty}^{\infty} \dfrac{1}{2\pi} P_\xi(\omega)e^{j\omega\tau}d\omega \end{aligned} \right. ⎩⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎧Pξ(ω)R(τ)=∫−∞∞R(τ)e−jωτdτ=∫−∞∞2π1Pξ(ω)ejωτdω
或:
{ P ξ ( f ) = ∫ − ∞ ∞ R ( τ ) e − j ω τ d τ R ( τ ) = ∫ − ∞ ∞ P ξ ( f ) e j ω τ d ω \left\{ \begin{aligned} P_\xi(f)&=\int_{-\infty}^{\infty} R(\tau)e^{-j\omega\tau}d\tau \\ \\ R(\tau)&=\int_{-\infty}^{\infty} P_\xi(f)e^{j\omega\tau}d\omega \end{aligned} \right. ⎩⎪⎪⎪⎪⎪⎨⎪⎪⎪⎪⎪⎧Pξ(f)R(τ)=∫−∞∞R(τ)e−jωτdτ=∫−∞∞Pξ(f)ejωτdω
由上面得定理我们可以得到以下结论:
n维随机过程均服从正态分布,则称为高斯过程,其n维正态分布表示如下:
f n ( x 1 , x 2 , ⋯ , x n ; t 1 , t 2 , ⋯ , t n ) = 1 ( 2 π ) n / 2 ∏ σ n ∣ B ∣ 1 / 2 e x p [ − 1 2 ∣ B ∣ ∑ j = 1 n ∑ k = 1 n ∣ B ∣ j k ( x j − a j ) ( x k − a k ) σ j σ k ] f_n(x_1,x_2,\cdots,x_n;t_1,t_2,\cdots,t_n)= \\ \\ \dfrac{1}{(2\pi)^{n/2}\prod\sigma_n|B|^{1/2}} exp \left[ -\dfrac{1}{2|B|} \sum_{j=1}^{n} \sum_{k=1}^{n} |B|_{jk} \dfrac{(x_j-a_j)(x_k-a_k)}{\sigma_j\sigma_k} \right] fn(x1,x2,⋯,xn;t1,t2,⋯,tn)=(2π)n/2∏σn∣B∣1/21exp[−2∣B∣1j=1∑nk=1∑n∣B∣jkσjσk(xj−aj)(xk−ak)]
其中:
a k = E [ ξ ( t k ) ] σ 2 = E [ ξ ( t k ) − a k ] 2 ∣ B ∣ = ∣ 1 b 12 ⋯ b 1 n b 21 1 ⋯ b 2 n ⋮ ⋮ ⋯ ⋮ b n 1 b n 2 ⋯ 1 ∣ a_k=E[\xi(t_k)] \\ \\ \sigma^2=E[\xi(t_k)-a_k]^2 \\ \\ |B|= \left\vert \begin{matrix} 1&\quad b_{12}&\quad\cdots&\quad b_{1n}\\ b_{21}&\quad 1&\quad\cdots&\quad b_{2n}\\ \vdots&\quad\vdots&\quad\cdots&\quad\vdots \\ b_{n1}&\quad b_{n2}&\quad\cdots&\quad 1 \end{matrix} \right\vert ak=E[ξ(tk)]σ2=E[ξ(tk)−ak]2∣B∣=∣∣∣∣∣∣∣∣∣1b21⋮bn1b121⋮bn2⋯⋯⋯⋯b1nb2n⋮1∣∣∣∣∣∣∣∣∣
∣ B ∣ j k |B|_{jk} ∣B∣jk为代数余因子, b j k b_{jk} bjk为归一化协方差函数:
b j k = E [ ξ ( t j ) − a j ] E [ ξ ( t k ) − a k ] σ j σ k b_{jk}=\dfrac{E[\xi(t_j)-a_j] E[\xi(t_k)-a_k]}{\sigma_j\sigma_k} bjk=σjσkE[ξ(tj)−aj]E[ξ(tk)−ak]
重要性质
高斯过程的一维概率密度函数为:
f ( x ) = 1 2 π σ k e x p ( − ( x k − a k ) 2 2 σ k 2 ) f(x)= \dfrac{1}{\sqrt{2\pi}\sigma_k}exp \left( -\dfrac{(x_k-a_k)^2}{2\sigma_k^2} \right) f(x)=2πσk1exp(−2σk2(xk−ak)2)
定义误差函数:
e r f ( x ) = 2 π ∫ 0 x e − t 2 d t erf(x)=\dfrac{2}{\sqrt\pi}\int_{0}^{x}e^{-t^2}dt erf(x)=π2∫0xe−t2dt
可得误差函数
为研究输入过程是平稳的,输出过程是否也是平稳的。输入与输出的关系可以表示为卷积,即:
v o ( t ) = v i ( t ) ∗ h ( t ) = ∫ − ∞ ∞ v i ( τ ) h ( t − τ ) d τ v_o(t)=v_i(t)*h(t)=\int_{-\infty}^{\infty}v_i(\tau)h(t-\tau)d\tau vo(t)=vi(t)∗h(t)=∫−∞∞vi(τ)h(t−τ)dτ
或:
v o ( t ) = h ( t ) ∗ v i ( t ) = ∫ − ∞ ∞ v i ( t − τ ) h ( τ ) d τ v_o(t)=h(t)*v_i(t)=\int_{-\infty}^{\infty}v_i(t-\tau)h(\tau)d\tau vo(t)=h(t)∗vi(t)=∫−∞∞vi(t−τ)h(τ)dτ
对于随机过程也满足:
ξ o ( t ) = ∫ − ∞ ∞ ξ i ( t − τ ) h ( τ ) d τ \xi_o(t)=\int_{-\infty}^{\infty}\xi_i(t-\tau)h(\tau)d\tau ξo(t)=∫−∞∞ξi(t−τ)h(τ)dτ
如果随机过程的频谱密度在中心频率 f C f_C fC附近相对窄的频带范围 Δ f \Delta f Δf内,且满足 Δ f ≪ f c \Delta f\ll\,f_c Δf≪fc,称为窄带随机过程,可表示为:
ξ ( t ) = a ξ ( t ) c o s [ ω c t + φ ξ ( t ) ] \xi(t)=a_\xi(t)cos[\omega_c\,t + \varphi_\xi(t)] ξ(t)=aξ(t)cos[ωct+φξ(t)]
其中:
我们也可以展开为:
ξ ( t ) = ξ c ( t ) cos ω c t − ξ s ( t ) sin ω c t \xi(t)=\xi_c(t)\cos\omega_c t- \xi_s(t)\sin\omega_c t ξ(t)=ξc(t)cosωct−ξs(t)sinωct
其中:
{ ξ c ( t ) = a ξ ( t ) cos φ ξ ( t ) 同 向 分 量 ξ s ( t ) = a ξ ( t ) sin φ ξ ( t ) 正 交 分 量 \begin{cases} \xi_c(t)&=&a_\xi(t)\cos\varphi_\xi(t) \quad同向分量 \\ \xi_s(t)&=&a_\xi(t)\sin\varphi_\xi(t) \quad正交分量 \end{cases} {ξc(t)ξs(t)==aξ(t)cosφξ(t)同向分量aξ(t)sinφξ(t)正交分量
ξ c ( t ) \xi_c(t) ξc(t)与 ξ s ( t ) \xi_s(t) ξs(t)
a ξ ( t ) 与 φ ξ ( t ) a_\xi(t)与\varphi_\xi(t) aξ(t)与φξ(t)
由上一点可知:
f ( ξ c , ξ s ) = f ( ξ c ) f ( ξ s ) = 1 2 π δ ξ 2 e x p [ − ξ c 2 + ξ s 2 2 σ ξ 2 ] f(\xi_c,\xi_s)=f(\xi_c)f(\xi_s)= \dfrac{1}{2\pi \delta_\xi^2}exp \left[ -\dfrac{\xi_c^2+\xi_s^2}{2\sigma_\xi^2} \right] f(ξc,ξs)=f(ξc)f(ξs)=2πδξ21exp[−2σξ2ξc2+ξs2]
由雅可比行列式变换得:
f ( a ξ , φ ξ ) = a ξ f ( ξ c , ξ s ) = a ξ 2 π δ ξ 2 e x p [ − a ξ 2 2 σ ξ 2 ] f(a_\xi,\varphi_\xi)=a_\xi f(\xi_c,\xi_s) = \dfrac{a_\xi}{2\pi \delta_\xi^2}exp \left[ -\dfrac{a_\xi^2}{2\sigma_\xi^2} \right] f(aξ,φξ)=aξf(ξc,ξs)=2πδξ2aξexp[−2σξ2aξ2]
a ξ a_\xi aξ$的一维概率密度函数服从瑞利分布:
f ( a ξ ) = a ξ δ ξ 2 [ − a ξ 2 2 σ x i 2 ] a ξ ≥ 0 f(a_\xi)=\dfrac{a_\xi}{\delta_\xi^2} \left[ -\dfrac{a_\xi^2}{2\sigma_xi^2} \right] \quad a_\xi\geq0 f(aξ)=δξ2aξ[−2σxi2aξ2]aξ≥0
φ ξ \varphi_\xi φξ一维概率密度函数服从均匀分布:
f ( φ ξ ) = 1 2 π 0 ≤ φ ξ ≤ 2 π f(\varphi_\xi)=\dfrac{1}{2\pi} \quad 0\leq\varphi_\xi\leq2\pi f(φξ)=2π10≤φξ≤2π
即均值为0,方差为 σ ξ 2 \sigma_\xi^2 σξ2的窄带平稳高斯过程其包络一维概率分布是瑞利分布,相位是均匀分布,且两者独立统计
设正弦波加窄带高斯噪声的混合信号为:
r ( t ) = A cos ( ω c t + θ ) + n ( t ) r(t)=A\cos(\omega_ct+\theta)+n(t) r(t)=Acos(ωct+θ)+n(t)
其中:
n ( t ) = n c ( t ) cos ω c t − n s ( t ) sin ω c t n(t)=n_c(t)\cos\omega_ct- n_s(t)\sin\omega_ct n(t)=nc(t)cosωct−ns(t)sinωct
为窄带高斯噪声,则,混合信号展开可得:
r ( t ) = [ A cos θ + n c ( t ) ] cos ω c t − [ A sin θ + n s ( t ) ] sin ω c t = z c ( t ) cos ω c t − z s ( t ) sin ω c t = z ( t ) c o s [ ω c t + φ ( t ) ] \begin{aligned} r(t) &= [A\cos\theta+n_c(t)]\cos\omega_ct- [A\sin\theta+n_s(t)]\sin\omega_ct \\ &=z_c(t)\cos\omega_ct-z_s(t)\sin\omega_ct \\ &=z(t)cos[\omega_ct+\varphi(t)] \end{aligned} r(t)=[Acosθ+nc(t)]cosωct−[Asinθ+ns(t)]sinωct=zc(t)cosωct−zs(t)sinωct=z(t)cos[ωct+φ(t)]
则r(t)的包络和相位得:
z ( t ) = z c 2 ( t ) + z s 2 ( t ) z ≥ 0 φ ( t ) = arctan z s ( t ) z c ( t ) 0 ≤ φ ≤ 2 π z(t)=\sqrt{z_c^2(t)+z_s^2(t)}\qquad z\geq0 \\ \\ \varphi(t)=\arctan\dfrac{z_s(t)}{z_c(t)} \qquad 0\leq\varphi\leq2\pi z(t)=zc2(t)+zs2(t)z≥0φ(t)=arctanzc(t)zs(t)0≤φ≤2π
当我们给定 θ \theta θ角时,包络线得概率密度函数为:
f ( z ) = z σ n 2 exp [ − 1 2 σ n 2 ( z 2 + A 2 ) ] I 0 ( A z σ n 2 ) z ≥ 0 f(z)=\dfrac{z}{\sigma_n^2}\exp \left[ -\dfrac{1}{2\sigma_n^2} (z^2+A^2) \right] I_0 \left( \dfrac{A_z}{\sigma_n^2} \right) \qquad z\geq 0 f(z)=σn2zexp[−2σn21(z2+A2)]I0(σn2Az)z≥0
得到的概率密度函数为广义瑞利分布,又称莱斯分布,其中:
I 0 ( x ) = 1 2 π ∫ 0 2 π exp [ x cos φ ] d φ I_0(x)= \dfrac{1}{2\pi} \int_{0}^{2\pi} \exp [x\cos\varphi]\,d\varphi I0(x)=2π1∫02πexp[xcosφ]dφ
为第一类零阶修正贝塞尔函数,当大于零时 I 0 ( x ) I_0(x) I0(x)单调上升,且有 I 0 ( 0 ) = 1 I_0(0)=1 I0(0)=1
当信号很小,即信噪比 γ = A 2 2 σ n 2 → 0 \gamma=\dfrac{A^2}{2\sigma_n^2}\to 0 γ=2σn2A2→0,分布退化为瑞利分布
当信号很大,分布接近高斯分布
如果噪声的功率谱密度在所有的频率上为一常数,其单边带表示为:
P n ( f ) = n 0 2 ( − ∞ < f < + ∞ ) ( W / H z ) P_n(f)=\dfrac{n_0}{2}\qquad (-\infty
则这样的噪声3为白噪声
则其自相关函数为:
R ( τ ) = n 0 2 δ ( τ ) R(\tau)=\dfrac{n_0}{2}\delta(\tau) R(τ)=2n0δ(τ)
易得其平均功率为无穷大
如果白噪声通过了低通滤波器,则称输出的噪声为低通白噪声,其功率谱密度为:
P ( f ) = { n 0 2 ∣ f ∣ ≤ f H 0 e l s e P(f)= \left\{ \begin{aligned} &\dfrac{n_0}{2}\qquad&|f|\leq f_H\\ \\ &0 \qquad &else \end{aligned} \right. P(f)=⎩⎪⎪⎨⎪⎪⎧2n00∣f∣≤fHelse
可以看出功率谱密度为门函数
假设理想带通滤波器传输特性为:
H ( f ) = { 1 f c − B 2 ≤ ∣ f ∣ ≤ f c + B 2 0 e l s e H(f)= \left\{ \begin{aligned} &1\qquad f_c-\dfrac{B}{2}\leq|f|\leq f_c+\dfrac{B}{2}\\ &0\qquad else \end{aligned} \right. H(f)=⎩⎨⎧1fc−2B≤∣f∣≤fc+2B0else
则输出的噪声功率谱密度为:
P ( f ) = { n 0 2 f c − B 2 ≤ ∣ f ∣ ≤ f c + B 2 0 e l s e P(f)= \left\{ \begin{aligned} &\dfrac{n_0}{2}\qquad f_c-\dfrac{B}{2}\leq|f|\leq f_c+\dfrac{B}{2}\\ \\ &0\qquad else \end{aligned} \right. P(f)=⎩⎪⎪⎪⎨⎪⎪⎪⎧2n0fc−2B≤∣f∣≤fc+2B0else
自相关函数为:
R ( τ ) = n 0 B sin π B τ π B τ cos 2 π f c τ R(\tau)=n_0B\frac{\sin\pi B\tau}{\pi B\tau}\cos2\pi f_c\tau R(τ)=n0BπBτsinπBτcos2πfcτ
由于通常滤波器 B ≪ f c B\ll f_c B≪fc,故也把带通白噪声叫做窄带高斯白噪声
其平均功率易得:
N = n 0 B N=n_0B N=n0B