I have a signal $s$, of which the PSD I call as $S$.


$$ S(\omega) = \frac{1}{N} \left|\sum_{n = 0}^{N-1} s(n) \exp(-j\omega n)\right|^2 $$

I have a closed form expectation of the above mentioned $S$:

$$ \mathbb{E}[S(\omega)] = F(\Theta) $$ where $F$ is the expectation with some parameters $\Theta$.

However, the signal is corrupted with a white Gaussian noise of variance $\sigma_n^2$

$z = s + n$, where $n \sim \mathcal{CN}(0, \sigma_n^2)$.

I want to write the probability distribution of $p(\Theta | Z )$ , if the PSD of $z$ is $Z$.

I know that the PSD is exponentially distributed at each frequency point. So, in principle, $S$ is distributed exponentially with expectation $F$ and similarly, the noise is also exponentially distributed. I have used a probability that I write below, however, it gives me a big variance in my estimates of $\Theta$, although the bias is good enough.

$$ \log(p(\Theta | Z )) = -\sum_{\omega} \left( L \log(F + \sigma_n^2) + \frac{\sum_{l= 1}^{L} Z_l(\omega)}{F + \sigma_n^2} \right) $$

where $Z_l$ are just different realizations of $Z$. Am I doing something wrong?



Your Answer

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

Browse other questions tagged or ask your own question.