I remember a result about the cross-correlation function of a stationary signal and that same signal, filtered through a linear time-invariant filter. There might have been an assumption that the signal was Gaussian. What is the cross-correlation function?

$$E\left\{ x(t) y(t) \right\} = \,\,\,?$$

where $y(t)$ is the output of an LTI filter with the input $x(t)$.


If the linear system (does not have to be time-invariant) is bounded-input bounded-output (BIBO) stable (meaning that if $|x(t)| \leq M$ for all $t$ for some finite positive number $M$, then there is another finite positive number $M^\prime$ such that $|y(t)| \leq M^\prime$ for all $t$), then when the input is a finite-variance random process (a.k.a. second-order proecess), one can interchange the order of the expectation operation and the integration/sum for the convolution. That is, if $$Y(t) = \int_{-\infty}^{\infty} h(s)X(t-s)\,\mathrm ds$$ where $h(t)$ is the impulse response of the LTI system, then $$\begin{align*} E[Y(t)] &= E\left[\int_{-\infty}^{\infty} h(s)X(t-s)\,\mathrm ds\right ]\\ &= \int_{-\infty}^{\infty} E\left[h(s)X(t-s)\,\mathrm ds\right ]\\ &= \int_{-\infty}^{\infty} h(s)E\left[X(t-s)\right ]\,\mathrm ds\\ \end{align*} $$ and so if the input is a WSS process with $E[X(t)] = \mu_X$ for all $t$, $$E[Y(t)] = \mu_X\int_{-\infty}^{\infty} h(s)\,\mathrm ds = \mu_Y$$ is also a constant. In fact, $\{Y(t)\}$ is also a WSS process. If we define the crosscorrelation function $R_{X,Y}(\tau)$ as $R_{X,Y}(\tau) = E[X(t-\tau)Y(t)]$, then $$\begin{align*} R_{X,Y}(\tau) &= E\left[X(t-\tau)\int_{-\infty}^{\infty} h(s)X(t-s)\,\mathrm ds\right]\\ &= \int_{-\infty}^{\infty} h(s)E[X(t-\tau)X(t-s)]\,\mathrm ds\\ &= \int_{-\infty}^{\infty} h(s)R_X(\tau-s)\,\mathrm ds. \end{align*}$$ In short, $$R_{X,Y} = h*R_X.$$ For completeness, $$R_Y = \tilde{h}*h*R_X = (\tilde{h}*h)*R_X = R_h*R_X$$ where $\tilde{h}(t) = h(-t)$ for all $t$ is the time-reversed impulse response and $R_h = \tilde{h}*h$ is the autocorrelation function of the deterministic signal $h(t)$. Translated to the frequency domain, this gives the power spectral density relationship $$S_Y(f) = |H(f)|^2 S_X(f).$$

|improve this answer|||||

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.