The longer your measurement goes on for, the better the resolution will be will be, resolution bandwidth ~= 1/measurement time. However, eventually, things like the stability limit of your system will result in broadening of the line so increased measurement time will have no effect beyond this.

Hello,

The important questions you have to answer before you launch a test involving a Fourier analysis :

* Frequency range you need for your analysis -> [0;fN]. This parameter leads you to chose your minimal sampling frequency fs => fs>2*fN

* Frequency resolution you need Deltaf. This parameter leads you to choose your measurement time Tm (or window length) => DeltaF>1/Tm

You have two another parameters you can tune, two :

* number N of time windows you use to estimate your fequency transfer function by averaging

* the overlap you use between the number N of averaging time windows

Once you have answered to these different questions, you can chose all your experimental parameters.

Be aware to the aliase phenomena two (by analogic filtering or oversampling). This is not specific to Fourier analysis, but to the time domain discretization you will use.

In the case of non-stationary or short time transient responses, the long recordings are not available to improve resolution. If you use FRF for modal analysis, the frequency resolution should be enough to recognise close peaks (natural modes). The large frequency step may result in smooth or even lost small peaks and looks like additional damping.

If you can get access to them, read the Keysight application notes on spectrum analysis. Many of the same principles apply when using an FFT, which is spectrum analysis. The old application notes are as good or better than the modern ones. They used to be Hewlett Packard and then Agilent. Their technical publications are excellent.