I would like to calculate $x(t)$, when only $y(t)$ with

$y(t) = A x(t-a) + B x(t-b)$

is known.

Since this is a linear shift invariant operation (convolution), the inverse relation must be of the form

$x(t) = (y * g)(t) = \int_{-\infty}^\infty y(t-\tau) g(\tau) \mathrm d \tau$

The question is: what is $g(\tau)$? I am sure this problem has been solved before, but I cannot find any suitable references.

Also I would be interested in solutions to the equation with three or more terms (the shifts can be equidistant though).

I know that there may be (some) situations where this equation has no solution, but I think not in the majority of cases.


You can use the theorem that convolution in the time domain is the multiplication of the Fourier transforms. So $X(\omega)=Y(\omega)G(\omega)$ where the capital letters represent the Fourier Transforms of the lower case letters.

We can also use the shift theorem. What is $Y(\omega)$? $Y(\omega)=Ae^{-i\omega a}X(\omega) + Be^{-i\omega b}X(\omega)$

Factoring out the $X(\omega)$ we can solve for $G(\omega)$ If you find a inverse transform of $G$ then you have found $g$.

By the way, are you sure there's an easily expression for $g$? Based on what you're taking the inverse transform of, I'm not sure there is.

  • $\begingroup$ Thanks, but I would like to have an (as explicit as possible) expression for $g$. $\endgroup$ – Andreas H. Nov 14 '14 at 3:57
  • $\begingroup$ @AndreasH. In general, this cannot be done. Suppose that $x$ is a triangle waveform, and that $a$ and $b$ shift $x$ by 1/2. Let $A=B=1$. Then $y=0$. So was $x$ a triangle waveform shifted by 1/2? Or was $x=0$, which would yield the same result? One can never know. $\endgroup$ – NicNic8 Jan 29 at 17:45

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.