Data length (time T for
loading 2048 sample data points) varies depending on the
frequency range. When changing the frequency range, the data
time to be FFT-processed changes, and energy therefore differs
proportionately.
In order to average and normalize energy, the value obtained
by dividing the power spectrum by T (i.e. the power spectrum for
each 1sec unit time) is taken into account, and the power
spectrum density is defined as the power spectrum per 1Hz.
With random signals, differences in the power spectrum value
occur even with the same signal due to FFT resolution
(equivalent to bandwidth). Power spectrum resolution (bandwidth)
is normalized to 1Hz and displayed, to remove as much as
possible the differences due to frequency range.
Energy spectrum density is defined as the power spectrum
density multiplied by the data length T.
With random signals, averaging involves the
use of power spectrum density, however with single events
(e.g. shockwaves), the energy has a time limit, and the energy
spectrum density for which the duration is taken into account is
therefore used.
These relationships are as follows. Assuming data length
(i.e. time) as T, and frequency resolution as Δf;
* Power spectrum: The spectrum found with a 2048-point FFT
(displays the rms amplitude for each frequency component)
* Power spectrum density: The Power spectrum / Δf
(displays power spectrum per unit frequency)
* Energy spectrum density: Power spectrum density x T
(displays energy spectrum per unit frequency)
T = 2048 / frequency range / 2.56
Δf = Frequency range / 800
|