I was reading "Discrete and continuous Fourier transforms: analysis, applications and fast algorithms" written by E.Chu and, at some point, I found something that I could not completly understand.

If $x(t)$ is not continuous, then it is possible that 2 different functions share the same transform $X(f)$. Therefore, the function $x(t)$ cannot be uniquely determined by inverse trasforming $X(f)$ unless it is continuous or it satisfies 2 more conditions:

1- $x(t)$ has only a finite number of maxima and minima on any finite interval

2- $x(t)$ has on any finite interval at most a finite number of discontinuities, each of which is a jump discontinuity.

In the latter case, the inverse Fourier transform of $X(f)$ produces $\hat{x}(t)$, which agrees with $x(t)$ at every $t$ at which $x(t)$ is continuous, and $\hat{x}(t)$ equals the average of the left-hand limit $x(t^-_{\alpha})$ and the right-hand limit $x(t^+_{\alpha})$ at every $t_{\alpha}$ at which a jump discontinuity occurs.

So, here my doubts.

First of all, I am not sure I got the meaning of the statements related to the 2 conditions listed. This is probably due to the fact that I am familiar only with signals coming from structural tests (mainly accelerations), therefore I think it could be really helpful if someone of you can provide me some example I can focus on to understand the concept (for example, how can be possible to have infinite maxima and minima in a signal? And what are the discontinuities it is talking about?).

Second, measured signals (for example, the aformentioned acceleration measurements coming from a test) are not continuous (since they are sampled at a given sampling frequency), so if I want to apply the IFFT algorithm to retrieve the time signal does it mean I have always to check if the 2 other conditions are satisfied? And how can I check them? Visual inspection? Or do I have to look for some specific features the signal must have?

Are there classes of signals which always satisfy the 2 conditions?

  • $\begingroup$ It is possible to theoretically formulate pathological functions which can't exist as signals in the real world, and have too great a bandwidth to be sampled anyway. But some of them have an FT anyway. $\endgroup$
    – hotpaw2
    Commented Jan 4, 2015 at 0:19

1 Answer 1


The conditions you mention only apply to the continuous-time Fourier transform, and they only have theoretical relevance. All signals occurring in engineering applications will satisfy these conditions. But if you want an example of a function with an infinite number of maxima and minima on a finite interval, here it is:


This function has an infinite number of minima and maxima on the interval $(0,a]$ for any real-valued $a>0$.

And as for your measured signals, they are all discrete-time signals of finite length, and the above conditions do not apply. So you don't need to worry about them when using the DFT/FFT, because the DFT is defined by a finite sum, unlike the continuous-time Fourier transform, which is defined by an improper integral.


Your Answer

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

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