frequency analysis of time varying data

[labmaster]

I want to add a routine for frequency analysis of before and after strong impulse on a body.

Some vibration sensors will be attached to the body.

In this case, I was told I need to use a time-varying method like wavelet or any other to analysis instead of simple FFT.

I installed wavelet package but I have no idea how to use it.

(Wavelet method seems to be popular in reduction of noise.)

Only what I want to know is the frequency elements in dynamics.

Can you recommend any tools (functions) for my purpose?

Are there some LabVIEW examples?

I am using LV2009 and some extra toolkits (developer suites).

labmaster.

[johnsold]

Wavelets can be used.  So can the Gabor transform. Or the STFT (Short Time Fourier transform)

But maybe you can do something simpler: Split the time domain data into before and after (the impulse) segments.  Then do FFT on each segment separately.  If the behavior is sufficiently stationary in each segment, the FFT will be fine.  If the frequency behavior changes during the segments, this will not work.

Lynn

[labmaster]

Hi,

I just tried STFT using the wizard in LabVIEW advanced signal processing toolkit.

The input data was "time waveform" on attachment, obtained real-DAQ experiment.

Some reasonable frequency could be identified with an artificially generated waveform (not shown here).

When I used STFT, why didn't identified the low frequency (<2kHz) of real-DAQ data?

In vibration analysis, I need the resolution in range of 2kHz.

However, how can I try with wavelet method with the same waveform?

labmaster.

[johnsold]

The time waveform in your image appears to be a single frequency at about 60 Hz, so why would you expect to see multiple frequencies on the spectrum?

Lynn

[labmaster]

Yes, that's my question in using STFT and real-DAQ data.

How can I identity the frequency, 60Hz?

Can you help me more in detail?

[labmaster]

I see the principle.

Too fast sampling rate for 60Hz.

labmaster.

[Coq_Rouge]

---

@labmaster wrote:

I see the principle.

Too fast sampling rate for 60Hz.

labmaster.

---

Nope......Then doing spectrum anylysis. Always use dB for Y-axis (magnitude) 

[labmaster]

I want to ask some curious things.

1. I have to analysis the spectrum of frequency in dynamic signal.

In case of presenting dB, how to know the frequencies?

2. In high sampling rate (1MHz), the low frequency element can't be analyzed. (too small window for 60Hz in my last example).

If so, generally, do I have to down sampling to ~1kHz for the 1MHz in wavelet or short time FT?

labmaster 

[Coq_Rouge]

---

@labmaster wrote:

I want to ask some curious things.

1. I have to analysis the spectrum of frequency in dynamic signal.

In case of presenting dB, how to know the frequencies?

2. In high sampling rate (1MHz), the low frequency element can't be analyzed. (too small window for 60Hz in my last example).

If so, generally, do I have to down sampling to ~1kHz for the 1MHz in wavelet or short time FT?

---

Take a look on how the Express VI you are using are made. The Express VI you are using is located here

C:\Program files\National Instruments\LabVIEW 2010\vi.lib\addons\Time Frequency Analysis\_express\_TimeFrequencySpectrogramConfig.llb

The VI you need to open is the configure.vi. Copy this VI to your user.lib folder before doing any work on it.

To answer your first question. The dB setting will change the Y axis scaling, and not the X axis scaling. Which is the frequncy axis 

The step between each frequncy bin will be( number of Frequency bins/(Fsample/2)). So using 1 MHz may reduce resolution. Perhaps 10 KHz will be bettter for you to use. 

[johnsold]

You can certainly measure 60 Hz with a 1 MHz sampling rate.  The key is that you need enough samples.  For FFT based analyses you need a minimum of about three cycles of the signal to get good results. So three cycles of 60 Hz is about 50 ms or 50000 samples at 1 MHz.  As Coq Rouge pointed out a much lower sampling rate will probably work.  At any rate you will need 50 ms or more of data.

What does the 60 Hz have to do with your impulse measurements?  Is your system resonant at that frequency or do you have power line frequency interference in your data?

Lynn