next up previous
Next: GAUSSIAN PROCESSES Up: Random Processes II Previous: Random Processes II

LINEAR SYSTEMS


\begin{displaymath}y(t)=x(t)\ast h(t)\end{displaymath}


\begin{displaymath}y(t)=\int h(t-\tau)x(\tau)d\tau\end{displaymath}

What can we say about y(t), y(t1), x(t2) etc. when x(t) is stationary.

Thm.

1.

\begin{displaymath}m_y=m_x\int_{-\infty}^a h(\tau)d\tau\end{displaymath}

2.

\begin{displaymath}R_{XY}(\tau)=R_X(\tau)\ast h(-\tau)\end{displaymath}

3.

\begin{displaymath}R_Y(\tau)=R_X(\tau)\ast h(\tau)\ast h(-\tau)\end{displaymath}

Def.

1.

\begin{displaymath}E(Y(t))=\int_{-a}^{\infty}h(t-\tau)E(x(\tau))d\tau=m_x\int_{-\infty}^{\infty}h(\tau)d\tau\end{displaymath}

2.

\begin{eqnarray*}R_{XY}(t_1,t_2)&=&E\left[\int_{-\infty}^{\infty}x(t_1)x(t_2-\si...
...infty}R_X(\tau-\sigma)h(-\sigma)d\sigma\\
&=&R_X(\tau)*h(-\tau)
\end{eqnarray*}


3.

\begin{eqnarray*}R_Y(t_1,t_2)&=&\int_{-\infty}^{\infty}\int_{-\infty}^{\infty} E...
...o)h(\sigma)h(\rho)d\rho d\sigma\\
&=&R_X(\tau)*h(\tau)*h(-\tau)
\end{eqnarray*}


Punchline $\rightarrow$ WSS in $\rightarrow$ WSS out
Also true that SSS in $\rightarrow$ SSS out, but we did not prove it

Power spectra
Def.
Define xT(t)=x(t)(u(t+T/2)-u(t-T/2))

XT(f)=F[xT(t)]


\begin{displaymath}S_X(f)=E\lim_{T\rightarrow\infty} \frac{\vert X_T(f)\vert^2}{T} =\lim_{T\rightarrow\infty} \frac{\vert X(f)\vert^2}{T}\end{displaymath}

Now need to relate RX(t1,t2) to SX(f)

Thm. (Weiner Khinchin)

If $R_X(t+\tau,t)$ is integrable on any interval of length $\vert\tau\vert$ with $\vert\tau\vert<\infty$, i.e.

\begin{displaymath}\left\vert\int_a^{a+\tau} R_X(t+\tau ,t)dt\right\vert<\infty \qquad \forall a\in R\end{displaymath}

then

\begin{displaymath}S_X(f)=F\left[<R_X(t+\tau,t)>\right]\end{displaymath}

where

\begin{displaymath}<R_X(t+\tau,t)>=\lim_{T\rightarrow\infty}\frac{1}{T} \int_{-T/2}^{T/2} R_X(t+\tau,t)dt \end{displaymath}

Please look but don't touch

Corollaries

1.
X(t) stationary and $\tau R_X(\tau)$ finite for $\vert\tau\vert<\infty$

\begin{displaymath}S_X(f)=F(R_X(\tau))\end{displaymath}

Pf.
W-K $\rightarrow$

\begin{displaymath}\lim_{T\rightarrow\infty}\frac{1}{T}\int_{-T/2}^{T/2} R_X(\tau)d\tau=R_X(\tau)\end{displaymath}

because $\tau R_X(\tau)$ finite, says

\begin{displaymath}\int_a^{a+\tau}R_X(t+\tau,t)dt=\int_a^{a+\tau}R_X(t)dt=\tau R_X(\tau)<\infty\end{displaymath}

2.
X(f) cyclostationary w/ period

\begin{displaymath}\left\vert\int_0^{T_0} R_X(t+\tau,t)dt\right\vert<\infty \end{displaymath}

then

\begin{displaymath}S_X(f)=F(\bar{R_X}(\tau))\end{displaymath}


\begin{displaymath}R_X(\tau)=\frac{1}{T_0}\int_{-T_0/2}^{T_0/2} R_X(t+\tau,t)dt\end{displaymath}

pf. conditions satisfied for W-K if satisfied over a single period and

\begin{eqnarray*}\lim_{T\rightarrow\infty}\frac{1}{T}\int_{-T/2}^{T/2}R_X(t+\tau...
...tau,t)dt\\
&=&\frac{1}{T_0}\int_{-T_0/2}^{T_0/2}R_X(t+\tau,t)dt
\end{eqnarray*}


Useful Stuff

1.

\begin{displaymath}P_X=\left.\int_{-\infty}^{\infty}S_X(f)df=<R_X(t+\tau,t)>\right\vert _{\tau=0}\end{displaymath}

2.
PSD's of each sample function are equal for an ergodic process
3.
Turns out that SX(f) is nonnegative and even

LTI systems again
Remember

\begin{displaymath}m_Y=m_X \int_{-\infty}^{\infty}h(\tau)d\tau\end{displaymath}


\begin{displaymath}R_{XY}(\tau)=R_X(\tau)*h(-\tau)\end{displaymath}


\begin{displaymath}R_Y(\tau)=R_X(\tau)*h(-\tau)*h(\tau)\end{displaymath}

So

\begin{displaymath}m_Y=m_X H(0)=\left.\int_{-\infty}^{\infty}e^{-j2\pi f\tau}h(\tau)d\tau\right\vert _{f=0}\end{displaymath}


SXY(f)=SX(f)H*(f)


SYX(f)=SX(f)H(f)


SY(f)=SX(f)|H(f)|2

NICE PICTURE

Sums

z(t)=x(t)+y(t)


mZ=mY+mX


\begin{displaymath}R_Z(\tau)=R_X(\tau)+R_Y(\tau)+R_{XY}(\tau)+R_{YX}(\tau)\end{displaymath}


SZ(f)=SX(f)+SY(f)+2Re[SXY(f)]

Uncorrelated?

\begin{displaymath}R_{XY}(\tau)=m_X m_Y\end{displaymath}

and if mY or mX is zero

SZ(f)=SX(f)+SY(f)


next up previous
Next: GAUSSIAN PROCESSES Up: Random Processes II Previous: Random Processes II

1999-02-06