I need to oversample my OFDM signal, however I have issues understanding which OFDM oversampling method is the recommended one.

First method :

Before taking the IFFT, the OFDM subcarriers are padded with $(\text{OSR} - 1) \times n_\text{subcarriers}$ additional null subcarriers, then the IFFT is taken. Therefore, we have oversampled each OFDM symbol separately before Parallel to serial (P/S) operation. We then add cyclic prefix. Now, when we put them together with the P/S operation, there may be discontinuities between the OFDM symbols which will generate some high frequency components. In conclusion, method 1 generates strong out of band radiation.

Second method :

We perform the IFFT with only a few null subcarriers, add cyclic prefix and perform the P/S operation. As the spectrum, is not yet oversampled, the discontinuities between different OFDM symbols are folded back onto the existing spectrum and do not generate out of band radiations. Now we can zero stuff the complete signal and use an interpolation filter. I believe that we have to be careful with the interpolation filter length, as the effective cyclic prefix length after interpolation will be $l_\text{CP new} = l_\text{CP old}\times \text{OSR} - l_\text{interp filt}$. This means that this method shortens the cyclic prefix length.

Third method :

This method is a combination of first method and F-OFDM. We zero-pad before the IFFT, take the IFFT, add cyclic prefix, then filter each OFDM symbol. Finally, the complete signal is constructed with the P/S operation. The advantage is that we have reduced out of band emissions, at the cost of a slight increase in EVM (due to loss of orthogonality in subcarriers). It also seems, from the example found in [1], that the PAPR of the F-OFDM method is raised.

What is the most typical method and what would be the most suited method for my case, considering that I want to reduce out of band emissions as much as possible without increasing PAPR too much ? Thank you very much.

[1] https://uk.mathworks.com/help/comm/ug/f-ofdm-vs-ofdm-modulation.html#mw_rtc_FOFDMVsOfdmModulationExample_00D53342

  • $\begingroup$ I never gotten the "there might be discontinuities" argument – there's really nothing special about the "jump" at the point you mention. All digital signals are by definition discontinuous. The "high-frequency components" are a myth here – the cylic prefix literally is just a windowed version of the original signal, meaning that its frequency "content" is just the original spectrum + leakage. So, do you actually observe such a broadening in spectrum, or is this just a concern you have? $\endgroup$ Aug 10, 2023 at 16:43
  • 1
    $\begingroup$ Does this answer help? dsp.stackexchange.com/a/83871/21048 $\endgroup$ Aug 11, 2023 at 19:20
  • $\begingroup$ @DanBoschen yes it helped thank you. So in the real world, we would oversample through interpolation with a FIR filter, a bit as described in the second method ? Should we worry about resistance to multipath using this approach of oversampling ? When the filter passes over the cyclic prefix, it will take samples from the previous OFDM symbol and therefore reduce the effective length of the cyclic prefix. $\endgroup$
    – Nicolate
    Aug 14, 2023 at 7:59
  • $\begingroup$ @Nicolate for OFDM specifically I would not necessarily oversample through interpolation- you are correct in being concerned about loss of orthogonality and degradation to EVM. Oversampling with zero bins and longer FFT is very straightforward; I am not convinced the discontinuity would create an excessive OOBE given the mask is set for an OFDM waveform - to the extent OOBE was an actual issue I would post filter with very careful consideration to EVM effects. My approach would be leaning toward your first method but I would want to validate that myself before providing an actual answer $\endgroup$ Aug 14, 2023 at 19:32
  • $\begingroup$ I plus one’d the question as this interests me, and if left unanswered and time falls in my lap I will confirm my suspicions $\endgroup$ Aug 14, 2023 at 19:34


Your Answer

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

Browse other questions tagged or ask your own question.