Correct Tukey window taper length?

Experimenter
Experimenter

I have another question.

 

While going through "Types of windowing" tutorials, I found tukey window and when I applied the same during my testing, then by changing the taper length from in "Acquisition parameters" while post processing,  the amplitude is changing as shown in here with attached picture. 

 

tukey_wind.png

 

How should this be intrepreted?  How does one set the taper length normally?

3 REPLIES 3

Re: Correct Tukey window taper length?

Siemens Experimenter Siemens Experimenter
Siemens Experimenter

tukey1.png

With a higher taper length on the Tukey Window, more time data is removed from the FFT calculation.  If there is an event (i.e. an impact or transient event) that exists outside the taper of the window, the spectral energy of that event is removed.  This phenomenon is most likely what you are experiencing in the data you submitted (pictured below).

 

graph.png

The possibility of missing impact events – which is the case in which a Tukey window is most used - on either end of the time sample increases when the taper percentage is higher.

 

In my opinion, I would use, and recommend using, a lower taper percentage most often when using this window to maximize the time data used in the FFT calculation.

Re: Correct Tukey window taper length?

Siemens Phenom Siemens Phenom
Siemens Phenom

With a taper length of 50%, the shapes of Tukey and Hanning windows are identical. Tukey and Hanning are designed for completely different usage.

 

Hanning: This window will be used for stationary type of measurement. It’s obviously used to avoid leakage due to not perfectly cyclic signals generating different levels first and last samples of time block. Of course, applying a window on a signals affect the amplitudes of the spectrum But on perfectly cyclic signals the amplitude attenuation is perfectly know and can be corrected. Testlab automatically applies the appropriate correction factor removing the impact of the window on the spectrum amplitude.

 

For impact test, Hanning is not a good options, because the most important part of the signal is concentrated at the beginning of the block where the Hanning window has the lowest amplitude. Rectangular, exponential of force-exponential windows are better choice. Force and force-exponential do not allow to apply any correction but it’s not necessary as most of the energy of the signal is concentrated where the window is closed to 1.

 

Tukey: This window has been developed for operational measurement of transient signals. Cleat impact testing is a perfect example. Defining properly a trigger for those signal is not easy. With small pre-triggers, there always a risk that beginning of signal will be lost. To avoid this, it’s a good idea to define the trigger to get the most important part of the block in the middle of the block. But because the trigger is difficult to define on such signals, the main part of the signal will most of the time not be exactly in the middle. An exponential window is not an option if main part of the signal is not very short and not at the beginning of the block. A Hanning window is not suitable for transient signal if the main part is not exactly in the middle. A rectangular window is not suitable for operational measurements. Operational measurement will always capture noise coming from other source than the cleat. Tukey is the perfect compromise between rectangular and Hanning. The taper length should be the highest possible to avoid leakage but to too high either to make sure that the relevant pact of the transient signal is fully windowed by 1 values.

 

Remark: Even if you use a taper length of 50%, Tukey and Hanning will not give you the exact same result in Testlab. This is normal. For Hanning we apply a correction factor to compensate for the lost energy due to the shape of the window. For Tukey we do not apply any correction as the relevant part of the signal is supposed to correspond to the 1 values of the window.

Highlighted

Re: Correct Tukey window taper length?

The whole discussion sounds good for me.