2
$\begingroup$

The Bussgang theorem states that if a stationary Gaussian process $X(t)$ with covariance function $R(\tau)$ is passing any non-linear function, affecting only the amplitude, $g(x)$, then the covariance of $y(t)=g(X(t))$ is $CR(\tau)$, for some C.

Now, let the process be instead cyclo-stationary with covariance $R(\tau_1,\tau_2)=R(\tau_1-\tau_2,\tau_2)=R(\tau_1-\tau_2,\tau_2+T)$, where $T$ is the period. Is it true that the covariance of $Y(t)=g(X(t))$ is $CR(\tau_1,\tau_2)$, possibly with another constant $C$?

$\endgroup$
  • $\begingroup$ Do you really mean what you write? what does ``affecting only the amplitude'' mean? The statement can't be true as written for stationary processes - take $g(x)=x^2$ to see that, as in that case $Ey^2(t)y^2(s)=R(0)^2+2R(|t-s|)^2$. Further, the wikipedia page linked is also not clear, and in any case does not deal with the covariance of $y$ but rather with the cross-correlation of the $x$ and $y$ processes. In short, this question needs cleanup. $\endgroup$ – ofer zeitouni Nov 12 '13 at 5:57
2
$\begingroup$

The correct statement is the following: $$E(Y(t)X(s))=F(t) E(X(t)X(s))$$ for some function $F(t)$ which is periodic. To see that, follow verbatim the proof in the appendix of http://www3.alcatel-lucent.com/bstj/vol61-1982/articles/bstj61-7-1519.pdf

$\endgroup$
  • $\begingroup$ I think the link is broken. Could you please update the link @ofer? Thanks in advance! $\endgroup$ – Shanks Apr 11 '19 at 16:24
  • $\begingroup$ This is the bell systems technical journal, volume 61, number 7, page 1519. I don't have open access to it but it can probably be accessed through library (it is a wiley journal). You can try archive.org/details/… if you have the patience to sort through their user-unfriendly interface. $\endgroup$ – ofer zeitouni Apr 12 '19 at 18:23

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.