I have a binary time series $\{X(n), n= 0,1,2,\cdots\}$, $X(n)\in \{0,1\}$ that can be written as: $$X(n)=S(n)+\epsilon(n)$$ Where $S(n)$ is a random stationary binary signal and $\epsilon(n)$ is a zero-mean binary white noise process (uncorrelated with $S(n)$). My question is what is the definition of a binary (or discrete) white noise?


Assuming that the $S(n)$ are also binary (taking on values in $\{0,1\}$) as are the $X(n)$, then I suspect that the $\epsilon(n)$ are also meant to be taking on values in $\{0,1\}$ and that $+$ in $X(n) = S(n) + \epsilon(n)$ is intended to be a modulo 2 sum or Exclusive-OR sum which might be better written as $$X(n) = S(n)\oplus \epsilon(n)\tag{1}$$ or, in real-number arithmetic as $$X(n) = S(n) + \epsilon(n) - 2S(n)\epsilon(n).\tag{2}$$

The model for this white noise process is an IID sequence of Bernoulli random variables with parameter $p$ that are independent of the $S(n)$ series.

Yes, I know that you said uncorrelated, but uncorrelated Bernoulli random variables are also independent random variables.

Readers of stats.SE and time-series books will undoubtedly be horrified at the nonlinear equation $(2)$, but this model is quite commonly used in the commumnications and information theory literature under the name Binary Symmetric Channel with crossover probability $p$ and readers of dsp.SE will be quite familiar with the model.


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.