Hi

You need only window signals when they are not periodic between the frames of your FFT. Rectangular windowing is used, e.g. for analysis of rotating machinery where FFT is taken per revolution to produce machine order spectrum.

That said, rectangular windowing works fine for a smoothly running (linear) machine but one migh be better off using some sort of windowing function when the machine is not running as it should.

There are many window functions to choose from. They all are best in one situation or the other.

You might be interested also in DFT adapted periodic noise which is a nice way of measuring FRFs, in particular in situations where reverberation time is long (=damping is low).

As an anecdote. I must confess that I have frequently used the name Hanning as the name is quite widespread. As the story goes, first was the a window function suggested by Hamming. A window function, very similar but a bit better, was suggested by von Hann. As the two were very close in shape, the Hann window was called Hanning - as a joke. The name stuck but, it is incorrect.

https://en.wikipedia.org/wiki/Hann_function

You find most of what you need to know about window functions here

https://en.wikipedia.org/wiki/Window_function

This is also an excellent source of information on signal processing

http://cp.literature.agilent.com/litweb/pdf/5952-8898E.pdf

Sincerely

Claes

Anyway, If your signals are periodic or not and you apply an rectangle window - each sampling starting and stopping at any time does it - and the borders does not fit the position of  zero crossing of your signal,  then you might find additional frequency parts in your spectrum. Either you have to synchronize your sampling window for zero crossing borders or you use a window. Hann, Hamming, Blackman.... they all do their job for damping signal amplitudes at the borders. Mostly Hann or Hamming windows will be used for better damping signals at the ends.

Window functions trade sidelobe reduction for mainlobe broadening.  The right window to use depends on the nature of the problem and what you want to see in the spectrum.  If you have a single sinusoid and just want to know its frequency, then a rectangular window is just fine.  If you have multiple spectra and want to see, say, -40 dBc spectral components, then you will want to use a window that suppresses sidelobes by somewhat more than 40 dB so that you don't overwhelm the weaker spectral components.  There are techniques that give both narrow mainlobe and low sidelobes, and are called "apodization" techniques, but are nonlinear and have other problems...

Hi

Here is another good resource on windo functions.

http://www.bksv.com/doc/bo0436.pdf

Sincerely

Claes

Hi

Here is another good resource on window functions.

http://www.bksv.com/doc/bo0436.pdf

Sincerely

Claes

Thanks for all your helpful information.