FFT of acceleration sensor

[Stenley]

I have done some measurements with an acceleration sensor both on an actual car and on a 4post shaker.
For FFT I use an appropriate number of fast fourier transfers, both tried hamming and hanning windowing etc.
But although I sample with motec at 500 Hz, there is extensive low frequency noise at low frequencies (approx beneath 0.5 ~1Hz).
After searching the internet I found out that in the loglog diagram this is a -3 dB slope.
My question is, as I suspect this low frequency noise to interrupt with my measurement, can I get this noise out or lowered?
With my 4post shaker measurement I do not have this noise, but a perfect signal at low frequencies.

[Holmz]

I have done some measurements with an acceleration sensor both on an actual car and on a 4post shaker.
For FFT I use an appropriate number of fast fourier transfers, both tried hamming and hanning windowing etc.
But although I sample with motec at 500 Hz, there is extensive low frequency noise at low frequencies (approx beneath 0.5 ~1Hz).
After searching the internet I found out that in the loglog diagram this is a -3 dB slope.
My question is, as I suspect this low frequency noise to interrupt with my measurement, can I get this noise out or lowered?
With my 4post shaker measurement I do not have this noise, but a perfect signal at low frequencies.

I doubt that the windowing is the problem, although depending on the length of the FFT and whether there is any DC bias removal, or slope between the first and last point you do get bleeding between the FFT bins. But your Nyquist is at 250-Hz. How are you getting low frequencies?
Are you zero padding the data?
You will get better resolution if you increase the FFT size, but it is usually worthwhile to do some bias shifting and slope removal. I find that usually it is easier to export the data into MATLAB.
Any slow change in acceleration will have low frequency energy components, and it is hard to not have low frequency changes in acceleration.

(Wondering if you are doing cruise control,,,)
What are you doing with the data?

Which axis are you looking at? (Lateral, Longitudinal, vertical)
Do you have any DC component in there?
What does it look like with the car at rest.
And what does it look like when idling?
And what does it look like when at a steady state cruise?

The stopping and starting have lots of energy in the 1/2 to 5 Hz range, which is probably why it is easy to get car sick.
I would bet that the data is true. If you can do some power spectra in those other states (idle, rest, cruise), you may get some further insight. The shaker was a good idea.
For me it is apparent looking at the spectra to tell if the engine is idling.