I saw this question on one of the sites related to DFT:

The analog signal $x(t)$ is band-limited to $40\textrm{ Hz}$. Suppose the signal is sampled at the rate of $100\textrm{ samples/second}$ and that at this rate $200\textrm{ samples}$ are collected. Then $200$ zeros are appended to the $200\textrm{ samples}$ to form a $400$-point vector. Then the $400$-point DFT of this vector is computed to get $X[k]$ for $0 \leq k \leq 399$.

  1. Which DFT coefficients are free of aliasing?
  2. The DFT coefficient $X[50]$ represents the spectrum of the analog signal at what frequency $f$? (Give your answer in $\textrm{Hz}$).

The answer's provided were:

  1. All of the DFT coefficients are free of aliasing. The sampling rate is more that twice the maximum signal frequency.
  2. The DFT bin width is $100/400$ or $0.25\textrm{ Hz}$. The $50^{\rm th}$ DFT coefficient corresponds to the frequency $50$ times $0.25\textrm{ Hz}$ or $12.5\textrm{ Hz}$

I'm not really sure if this is right because when you append zeros you are upsampling and the sampling rate becomes $400\textrm{ Hz}$. So when you take a $400$-point DFT the bins are seperated by $1\textrm{ Hz}$.

Please let me know if I am right or not. If I am wrong please explain the answer to me.


  • 1
    $\begingroup$ That is incorrect. When you append zeros you are not up-sampling your signal. Sampling rate is always the same, you only change the spacing of frequency bins and interpolating samples between them. $\endgroup$
    – jojek
    Feb 26 '15 at 9:22

The provided answers are correct. Appending zeros is not upsampling and therefore the sampling rate is still $R_\mathrm{s}=100\, \text{Samples/s}$. Accordingly, the frequency spacing after taking the N-DFT is $f_\mathrm{d}=(R_\mathrm{s}/N)=0.25\, \text{Hz}$.

Upsampling can be achieved by inserting $N_\mathrm{U}$ zeros after every sample followd by lowpass filtering.

  • $\begingroup$ So if I need a sampling frequency of 400 hz , I would have to insert 3 zeros after every sample.This would be up sampling.If I then take a 400 point DFT I will get a bin reseloution of 1 Hz $\endgroup$ Feb 26 '15 at 10:05
  • $\begingroup$ @JoseKurian Yes, exactly. Appending zeros to a signal corresponds to an interpolation in frequency domain. Zero-padding after every sample plus lowpass filtering corresponds to an interpolation in time domain. $\endgroup$
    – Deve
    Feb 26 '15 at 10:33

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.