7

I am a programmer but quit bad on math.. I recently read an article which mentioned frequency-domain entropy,he calculated that thing from a FFT power spectrum but didn't tell me how to do that. I cannot find enough information online to understand what it is . I'm appreciate for any helps..

Here is part of article: enter image description here

Marcus Müller
  • 34,677
  • 4
  • 53
  • 94
Yank
  • 718
  • 6
  • 17
  • I've never heard of entropy being calculated in the frequency domain. Entropy is used to calculate the average amount of bits required to convey information about a signal... this is a really watered down version of what it is, but that's what you can take away if you're in a hurry. For signals... whether they are waveforms, images, etc., it boils down to computing a histogram. I don't see why you should go into the frequency domain to calculate the entropy of a signal.. or why you should even go into the frequency domain all together. Can you link us to this article? I'm rather skeptical. – rayryeng May 23 '15 at 23:05
  • Hi , this is due to machine learning usage , i need to fetch the features from some data as more as possible(but without thinking their actual physical meaning), the article is A comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data , written by IEEE member in 2009 march. – Yank May 23 '15 at 23:09
  • But if you know mathematically how to calculate it please tell me :) – Yank May 23 '15 at 23:10
  • @Yank could you please give me the source of this article. – iamprem Dec 06 '15 at 03:24

1 Answers1

7

Frequency domain entropy, also known as a (Power) Spectral Entropy calculation is done in following steps:

  1. Calculate the FFT of your signal.

  2. Calculate the PSD of your signal by simply squaring the amplitude spectrum and scaling it by number of frequency bins.

enter image description here

  1. Normalize the calculated PSD by dividing it by a total sum.

enter image description here

  1. Calculate the Power Spectral Entropy using a standard formula for entropy calculation:

enter image description here

jojeck
  • 935
  • 9
  • 29
  • Thanks for answering ! The equation you explained is pretty clear,however I cannot find where these equation come from ,could you provide a link or something ..? – Yank May 26 '15 at 20:46
  • I am in travel now. It was one of the publications on PSE topic. – jojeck May 26 '15 at 20:47
  • really thanks ,start reading some of it ,once I found the article I will approve your answer:) – Yank May 26 '15 at 20:54
  • Good luck! Try to search for something like *extraction of spectral entropy feature for eeg signals*. That was definitely one of the BioMed & Info conferences.Obviously there is plethora of different papers with that. I used it only for capturing of the equations, since SO is picky about TeX. – jojeck May 26 '15 at 21:06
  • Wonderful!!! Is there a restriction with the base of logarithm? I found the same with log2 [here](http://dsp.stackexchange.com/questions/10137/spectral-entropy-calculation-in-matlab) – iamprem Dec 06 '15 at 01:45