- Suppose $w$ is a vector containing $L$ i.i.d. normally distributed samples.
- Suppose $v$ is a vector containing $N < L$ i.i.d. normally distributed samples.
- Suppose the elements of $w$ and $v$ are independent.
The correlation between $w$ and $v$ is now calculated using a sliding correlator (the shorter vector $v$ is slid across the larger vector $w$) using
$R(j) = \sum_{n=1}^{N} v_n w_{n+j}$
Question
What is the mean and variance of the correlation $R$ in terms of the means and variances of $w$ and $v$? (Experimental results indicate that $R$ also has a normal distribution.)
Simpler case
In the case where $v$ and $w$ have equal lengths ($L=N$) and the correlation is calculated using the inner (dot) product (no sliding) $P = \sum_{n=1}^N v_n w_n$, it can easily be shown that $P$ is normally distributed with mean
$\mu_P=\mu_v \mu_w$
and variance
$\sigma_P^2 = \sigma_v^2 \sigma_w^2 + \sigma_v^2\mu_w^2+\sigma_w^2\mu_v^2$
due to independence.
The sliding correlator however introduces correlation between the samples of $R$ and the above expressions for mean and variance no longer hold. How does one go about expressing the statistics for $R$?
