0

I am a newbie in signal processing. I am testing the Hanning window on a signal. Obviously doing this window also means cut some energy of the signal, and when I do the DFT, the 0 frequency is not equivalent anymore to the average of the time history. Do you usually solve in some way this problem? Can someone suggest papers or books where noise-reducing strategies for the FT of a signal are investigated? I just don't know how to start, an online search didn't give me more than basic definition (so please don't send me the Wikipedia definition of the tapering window)

Best

jojeck
  • 11,107
  • 6
  • 38
  • 74
Luca Mirtanini
  • 432
  • 3
  • 12
  • Hm, I find difficult to relate Hanning window, the DC frequency, and noise reduction in the frequency domain. What kind of signals are you dealing with? – GKH Jan 19 '20 at 15:03
  • 2
    In a nutshell, you must normalize the energy by the energy of the window. In case of simple magnitude, this means dividing by the sum of window samples. This publication has great explainations. – jojeck Jan 19 '20 at 15:03
  • 1
    Excellent reference, thank you for reminding me it is in my "to read in a serious manner" list for a while – Laurent Duval Jan 19 '20 at 15:15

1 Answers1

1

I suspect you are talking about the Hann window, after Julius von Hann, and Blackman & Tukey. Of course, a non-constant window may induce energy variations, either used as a weighted average, or a convolutional filter. This can be compensated in several ways:

  • by normalizing the window amplitudes
  • by separating tipme-portions, or signal components, like removing the mean or low-frequencies before doing the processing
  • by modelling the signal and applying model-based corrections.

More would be needed on your signal to provide more precise answers. R. Lyons on a diagram on page xviii of his book Understanding Digital Signal Processing highlights the importance of windows.

Laurent Duval
  • 31,850
  • 3
  • 33
  • 101