Interpretation of FFT signal and determination of low pass cut-off frequency - acceleration signal

[StathPol]

Guys,

I have used DIAdem to plot the FFT of an IEPE acceleration signal sampled at 100 kHz with 100 kHz resonance frequency.

I have some difficulties for the interpretation. I would like to use a digital filter to filter out the high frequencies in the signal which is noise. However, how I determine the cut-off freqency? There are two peaks; one at 1 kHz and one near 4 kHz. Do I just say I picked 4 kHz because the peaks were not affected much (this is what i used in green colour? Thanks.

 

[dr2]

The cut-off frequency that you should be using is completely dependent on what you are trying to learn from your measurement.  Do you expect to have signals above 4kHz of interest?  If so, the a cut-off frequency of 4 kHz is a bad idea.

For me, if I've already done the FFT, why bother with the filter?  Just plot the FFT up to the frequencies of interst and ignore any higher frequencies (of course this pre-supposes that you are using a hardware filter to ensure that you don't have any aliasing in your signal).

[StathPol]

Nequist theorem, the maximum signal is 50 kHz as i sample at a rate of 100 kHz.

After 4 kHz it seems that there is noise, isnt it?

[Daikataro]

Also keep in mind that, depending on your application, Nyquist theorem might not hold true. I put together this simple VI to help illustrate the point, a simple sine wave generator sampling at 10kHz, where you start having data loss as soon as sampling rate is lower than 10 times the signal frequency, and gradually turns into a garbled mass of points that resembles a sine wave as it progresses, so if you want to reconstruct the signal, 2 times frequency might not be enough. It is further explained in this MATLAB thread

https://www.mathworks.com/matlabcentral/newsreader/view_thread/314114?requestedDomain=www.mathworks....

And this bit about oscilloscopes:

http://electronics.stackexchange.com/questions/39788/why-did-aliasing-occur-at-a-frequency-before-th...

This one hits close to home as I had issues with sampling rate a while ago, struggling with a similiar problem, until I further delved into Nyquist theorem.