I know that a function $f_a\in \text{L}^2(\mathrm{R})$ is an analytic signal if its Fourier transform is 0 for negative frequencies, i.e.

$$ \hat{f_a}(\omega)=0$ if $\omega<0 $$

We can characterize an analytic function by its real part $f = \text{Re}[f_a]$, and the Fourier transform of this function is

$$ \hat{f_a}(\omega)=\left\{\begin{matrix} 2\hat{f}(\omega) & \text{if }\omega\geq0 \\ 0 & \text{if }\omega<0 \end{matrix}\right. $$

Now if we have a discrete signal $f[n]$ of size $N$, we can also compute $f_a[n]$ by setting the negative frequency components of its discrete Fourier transform to 0.

Now why is it that if we want $\text{Re}[f_a]=f$, then we must have:

$$ \hat{f_a}[k]=\left\{\begin{matrix} \hat{f}[k] && \text{if }k=0,N/2\\ 2\hat{f}[k] && \text{if }0<k<N/2\\ 0 && \text{if } N/2<k<N \end{matrix}\right. $$

Could that be explained?

  • $\begingroup$ I thought a function was analytic on the line or plane if it can be approximated by a convergent series. $\endgroup$
    – Tom Kealy
    Sep 4, 2013 at 8:27
  • 2
    $\begingroup$ @Tom Kealy: The OP confuses analytic functions and analytic signals. $\endgroup$
    – chirlu
    Sep 4, 2013 at 10:26
  • $\begingroup$ @chirlu I mentioned that in my answer below. $\endgroup$
    – Tom Kealy
    Sep 4, 2013 at 10:54
  • $\begingroup$ Your characterization of an analytic signal in the continuous-frequency domain is incorrect. See Wikipedia's definition. $\endgroup$
    – Peter K.
    Sep 4, 2013 at 12:01

2 Answers 2


Analytic signals to get a but more tricky in the discrete domain because of the periodicity in the time domain and the circular nature of the convolution. This article treats that fairly well http://www.andrewduncan.ws/air/


It's not clear what you're asking, but I think you want to know why the DFT of a signal has extra conditions for it to real.

For the continuous case: the 2 is there because we need a function to be symmetric to have a real Fourier transform.

In the discrete case, the DFT obey the symmtery:

$$ X_{N-k} = X_{-k} = X_k^* $$ i.e. conjugate symmetry (half the values are just complex conjugates of the other half). Thus $X_0$ and $X_{N/2}$ are real numbers whilst the remainder of the DFT is specified by $N/2-1$ complex numbers (hence the 2 in the second line).

Also, the answers to this question.

offer counterexamples to your first statement - that analytic functions have vanishing negative frequencies.


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.