What is the autocorrelation of $x(t) = \delta(t)$?

Can you explain to me how to calculate it?


2 Answers 2


Well, by definition of the $\delta$ distribution, you have:

$\int_{-\infty}^{\infty} f(t) \delta(t-T)\, \textrm{d}t = f(T)$

The autocorrelation of a function $g(t)$ can be computed via:

$\int_{-\infty}^{\infty} g^{*}(t)g(t + \tau)\, \textrm{d}t$, with $g^*$ as the complex conjugate of $g$. Since $\delta(t)$ is real-valued, this is conjugation can be skipped. So you are left with:

$\int_{-\infty}^{\infty} \delta^{*}(t)\delta(t + \tau)\, \textrm{d}t = \int_{-\infty}^{\infty} \delta(t)\delta(t + \tau)\, \textrm{d}t = \delta(-\tau)$.

The first = sign comes from the autocorrelation of the real-valued $\delta$, the second from the definition of the $\delta$-distribution.

So, the autocorrelation function of the $\delta$-distribution is the distribution itself. A eigenfunction of the autocorrelation function, so to say ;)

Think about it, this does make sense: the only perfect match is achieved with no time shift, ie at $\tau = 0$. All other shifts would end up with one of the arguments of the $\delta$ being different from 0, hence with the $\delta$-function being 0 there.

BTW: $\delta(-\tau) = \delta(\tau)$, since the function/distribution is symmetric.

  • $\begingroup$ Perfect, thank you very much! $\endgroup$
    – Kinka-Byo
    Oct 21, 2019 at 19:00
  • 4
    $\begingroup$ I am not sure mathematicians would agree. $\endgroup$ Oct 21, 2019 at 19:13
  • $\begingroup$ At first, I was in line with your comment, but apparently this can be well defined $\endgroup$ Oct 21, 2019 at 20:32
  • 1
    $\begingroup$ The first equation $$\int_{-\infty}^\infty f(t)\delta(t-T)dt = f(T)$$ is valid only when $f(t)$ is an ordinary function that is continuous at $t=T$. So, applying this to conclude that $$\int_{-\infty}^\infty\delta(t)\delta(t+\tau)dt = \delta(-\tau)$$ is not very convincing. $\endgroup$ Oct 22, 2019 at 20:58
  • $\begingroup$ @DilipSarwate I fear you are right. $\endgroup$
    – M529
    Oct 23, 2019 at 18:09

Looking at documents like Lecture notes on Distributions, Hasse Carlsson, or Convolution dans l'espace $\mathcal{D}'_+(\mathbb{R})$, convolution of distributions can be defined under some technical conditions. However, when one of the operand has a compact support, as $\delta(t)$ does, the convolution is well-defined. From Wikipedia:Distribution-Convolution:

Distribution of compact support: It is also possible to define the convolution of two distributions S and T on $\mathbb{R}^n$, provided one of them has compact support.

If you prefer classical books, there is: Francois Trèves, Topological Vector Spaces, Distributions and Kernels, 1967:

Francois Treves, Topological Vector Spaces, Distributions and Kernels

Adding that $\delta(t)$ is the neutral element, then $\delta(t)$ is its own autocorrelation.

This lazy version avoids to write cumbersome Dirac products under integral signs. Nota: I do admit that I initially thought it was way less simple.

  • 2
    $\begingroup$ We do get into trouble for $\tau=0$ though, don't we? $\endgroup$
    – Matt L.
    Oct 22, 2019 at 12:41
  • $\begingroup$ Intuitively yes, but the standard distribution theory seems to be OK with that. While the square of a Dirac is not defined. $\endgroup$ Oct 22, 2019 at 14:28
  • 1
    $\begingroup$ But for $\tau=0$ we get exactly that, the square of the Dirac impulse, which is indeed undefined, with or without integral, as far as I know. $\endgroup$
    – Matt L.
    Oct 22, 2019 at 16:13
  • $\begingroup$ As far as I remember, operations on distributions are defined with respect to integration over a set of test functions. Don't mean to convince your with an analogy, yet it's a bit (more complicated) like the Cauchy principal value (a distribution), which can get sense over Schwartz functions $\endgroup$ Oct 22, 2019 at 21:25
  • 1
    $\begingroup$ Would you care to write a fuller explanation of how you came to change your mind from two years ago when you wrote that $(\delta(t))^2$ is undefined? $\endgroup$ Oct 23, 2019 at 0:54

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.

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