I've been driving myself absolutely crazy trying to do autocorrelation analysis on some data that I've got. I've read from multiple sources that the ACF of a random telegraph signal / random telegraph noise is given by an exponentially decaying function where the decay constant is the average rate of switching between states (cf. Example 7.3.2 here for example). Since I had trouble getting anything that even resembled exponential decay with real data, I decided to generate a fake signal that telegraphs between the value 0 and 1 with lifetimes centered binomially around two characteristic values for the 0 and 1 state. The autocorrelation function for this signal is still not at all exponential. In each case, the ACF ends up strongly oscillating.

I'm working in Mathematica, using CorrelationFunction to get the autocorrelation function. Based on my reading, this is a standard definition for autocorrelation.

So my question is can someone show me an example with a numerically obtained ACF from a random telegraph signal that decays exponentially? Am I missing something stupid here? Am I using an incorrect definition for the autocorrelation function?

  • $\begingroup$ Could you post links to figures of your artificial telegraph signal and its autocorrelation function? $\endgroup$
    – Matt L.
    Commented Jul 1, 2013 at 16:36
  • $\begingroup$ What is the definition given for a random telegraph signal? Do the content, QSO length, WPM, fist weight and carrier frequency all vary, and with what distribution? $\endgroup$
    – hotpaw2
    Commented Jul 1, 2013 at 16:58

1 Answer 1


A semi-random telegraph signal is one for which $X(0) = 0$ and $X(t)$ equals $0$ or $1$ according as there have been an even number or an odd number of arrivals in a Poisson Process in the interval $(0,T]$. A simple physical model would be the least significant bit (LSB) of a digital counter that is counting the Poisson arrivals. A random telegraph signal is almost the same except that $X(0)$ is equally likely to have value $0$ or $1$. (Feel free to translate this to signals having values $\pm 1$ if that is more convenient or more in accordance with your notation).

With this as background, you are encountering problems because your simulation is not modeling the random telegraph signal correctly. You should instead simulate a Poisson process and obtain the random telegraph signal from that.

  • $\begingroup$ If I may ask, for large time increments, does a Bernoulli process approximate a telegraph process? Please see this question for details. $\endgroup$
    – sam wolfe
    Commented Jul 31, 2022 at 20:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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