We currently have C/C++ FFT code with Hann Window function support. For a large data (over 10,000) we want to split this into 1024 samples with 50% overlap. For each 1024 sample, we window and fft.

What is the recommended way to combine all the results from these smaller overlapping results to obtain a presentable results for the over 10,000 data?

Update: For the poor explanation. The process is best illustrated by the following links



  • $\begingroup$ i wasn't the one to down-arrow you, but could you be more specific about what is is that you have done or can do, and then what it is that you're missing? maybe a little code or pseudo-code would be illustrative. $\endgroup$ – robert bristow-johnson Aug 2 '15 at 19:43
  • $\begingroup$ Why the overlap? Is the signal you expect a transient one? Overlap is usually done for better IFFT. $\endgroup$ – Moti Aug 3 '15 at 3:52
  • $\begingroup$ I have added links to illustrate the process. My thought is to combine the non-overlapping FFT output for the final result, but could not find any source of information to confirm that approach. $\endgroup$ – paulusnet Aug 3 '15 at 12:51

I'm not sure what your intent is, but the organization of the computation can be done using straightforward overlap-and-add or overlap-and-save methods.

If you're attempting to look at data statistics in the frequency domain, a good starting might be to look at Welch's method and its descendents.

| improve this answer | |
  • $\begingroup$ Thanks, I found out it falls under FFT Averaging and looking into that right now. $\endgroup$ – paulusnet Aug 7 '15 at 22:51

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.