# Does Zero padding cause noise in the high frequency region?

I have an OFDM system. I would like to introduce a fractional delay of (0.2 samples). Therefore, I increase the sampling rate of the OFDM signal by a factor of 10 (via frequency domain zero padding), and then I shift the signal by simply adding 2 zeros at the begining of the signal, and then I downsample the signal by a factor of 10. So that the fractional delay will be (2/10 = 0.2). However, I have noticed, that by doing this, after equalization and phase correction (due to this introduced timing offset), the performance of high-frequency subcarriers is lower than low-frequency subcarriers. My conclusion is that increasing the sampling rate by frequency domain zero-padding introduces some noise near Nyquist frequency. Is that correct? thanks

• Have you tried it without zero-padding? If you increase the sample rate, you need to filter the aliases before downsampling. It seems to be that the issue is due to your oversampling.
– Ben
May 9, 2022 at 17:22

Is that correct?

No. This zero padding just leads to interpolation with a (cyclic) sinc kernel. It affects all subcarriers the same (as you can see in your own DFT!).

So, this has to be a problem with how well your receiver synchronizes that offset. Typically, higher-frequency subcarriers are more sensitive to timing offset (which is quite logical, they make more phase difference in the same time).

• I thought that zero-stuffing led to interpolation, not zero-padding.
– Ben
May 9, 2022 at 20:20
• @Ben what Amro does (if I understood correctly is) 1. Take N samples 2. Do N-point DFT 3. add 9·N zeros to the end 4. do 10·N-point IDFT, and that's a sinc interpolation. May 9, 2022 at 20:28
• Hi all, thank you for your support. I have posted a new question regarding this subject. I attached my simple Matlab code and the results. I would be very grateful if you have a look on my new post and my code: dsp.stackexchange.com/questions/82963/… May 10, 2022 at 9:06
• @MarcusMüller could you please check this question when you have time, thanks dsp.stackexchange.com/questions/82963/… May 10, 2022 at 12:06
• Just skimmed the comments. Dan is simply right. You already have the solution in the comments! May 10, 2022 at 15:53

If you've got it in the frequency domain anyway, why not phase shift it by (delay)(frequency)? It may still do odd things (like, shift whatever's at the end of the sample to the beginning), but you'll use a lot fewer processor cycles.

• Hi, I have posted a new question regarding this subject, I attached my simple Matlab code and my results, I would be very grateful if you have a look at my new question and my code dsp.stackexchange.com/questions/82963/… May 10, 2022 at 9:09