If I have an equation of a signal in the frequency domain given by


and I want to convert this to the time domain, can I do this using the inverse Fourier transform? I think the result should be an exponential, but when I tried to calculate this on mathematica I got a strange result.

  • $\begingroup$ Why don't you show us what Mathematica gave you? This looks like the square modulus of a typical signal's Fourier transform, not the transform itself. Did Mathematica give you a damped sinusoid? $\endgroup$ – garyp Apr 11 '18 at 11:14
  • $\begingroup$ I just used the mathematica expression $FourierTransform[2.5/(Pi*((x - 97.5)^2 + 2.5^2)), x, t]$ which returns $ 0.795775 exp((-2.5 + 97.5 I) t) (0.501326 exp(5. t) HeavisideTheta[-t] + 0.501326 HeavisideTheta[t])$ $\endgroup$ – JJH Apr 11 '18 at 11:59
  • 1
    $\begingroup$ The result looks reasonable as it reflects the distinction of case for $t>0$ and $t<0$ (the Heaviside function shows this) which has to be applied to make the complex path integral convergent. $\endgroup$ – Frederic Thomas Apr 11 '18 at 15:13
  • $\begingroup$ All ok. Except that $L(x)$ looks like a power spectral density (PSD) rather than a spectrum. If that's correct, the Fourier transform will give the autocorelation function of the original signal, not the original signal itself. I'd expect the autocorrelation function to be a damped sinusoid. And that's what Mathematica produced. $\endgroup$ – garyp Apr 11 '18 at 16:46

You can do the Fourier integral $\int L \exp(i\omega t) \frac{d\omega}{2\pi}$ easily by the application of the theorem of residues. First you look for the poles in the complex plane, second you choose the integration path (either a half-circle in the upper or in the lower complex plane, actually a distinction of cases has to be done for $t>0$ and $t<0$) checking that the integrand converges to zero on the half-circular part at infinity), and finally can apply the formula of the residue theorem which is essentially

$\oint f(\omega) d\omega = 2\pi i \sum\limits_{k=1,\ldots n} Res(f,\omega_k)$ with $f(\omega) = \frac{1}{\pi}\frac{1}{2\pi} \frac{a \exp(i\omega t )}{(\omega-\omega_1)(\omega-\omega_2)}$ with $\omega_1 = \omega_0 +ia$ and $\omega_2= \omega_0 -ia$ and $a = \frac{1}{2}\Gamma$

where $\omega_k$ is the k-th pole which is surrounded by the integral path. As the chosen half-circle (the integration path) only encircles one pole the expression Res(f) gives: $Res(f,\omega_{1/2}) = f(\omega)\cdot (\omega-\omega_{1/2})|_{\omega=\omega_{1/2}}$ 1 or 2 has to be chosen depending on which pole is encircled by the integration path.


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.