As I mentioned above, you can express PSD as
V/. One of
my first measurements involving PSD called for just that (measuring
eigenmode shapes of MEMS devices by thermal noise excitation. The
mode shape was proportional to V not V2). So, rather than
square a bunch of FFTs, average them together, and then take the square
root, what I did was to just average together a bunch of FFTs. I figured
this saved me a lot of computation time and would give the same answer.
I was wrong. Squares and square roots are not linear operators. Now
if your data is pretty clean it will be close and you might not notice.
For example
= 1 and
(1 + 1 + 1) = 1
so it seems like this works, but with noisy data
= 1.0033
while
(1.1 + 1.0 + 0.9) = 1.
Now to drive home an earlier point, what happens if we calculate the PSD of a periodic signal? This is shown in 25. As you can see, the FFT magnitude is correct for each different length of signal, but the PSD is not. To reiterate: PSD (and only PSD) applies to random vibration. FFT magnitude (and only FFT magnitude) applies to periodic vibration. Do not confuse the two!
Also, if you've done a lot of vibration analysis, you know that for
a periodic signal, you can convert between displacement, velocity,
and acceleration in frequency (Fourier) domain by simplying multiplying
or dividing by . You can do a similar operation for PSD as
well, but the relations are now:
Pdisp = to convert
(m/s)2/Hz
to m2/Hz or
Paccel =
Pvel, to convert
(m/s)2/Hz
to
(m/s2)2/Hz.
The extra comes from the fact that the units are squared.
http://www.desy.de/~sahoo/WebPage/Ground%20Motion.htm
/' $I