Real-life test case
Finally, we want to test our method on real-life signals. The setup comprises 3 high sensitivity accelerometers (model CA YD-119, 1000pC/m/s²), a MMF model M72B3 charge amplifier, and a DT 9837A DAQ. The entire setup is located in a low vibration environment: our lab is located in a basement with heavy masonry walls, and the setup is resting on a heavy (20mm thick), steel plate.
Applying the same signal processing procedure as previously, the sensors noise spectral content (expressed as Amplitude Spectral Density of acceleration, i.e. mm/s²/(Hz)^(1/2)) is obtained as follows:
As expected, the 3 DUT’s being identical, they provide very similar estimates, with an exception around 12Hz. At this frequency, the physical signal content completely dominates the measurements (it comes from a highway bridge located about 250m away, continuously excited by road traffic), so that the estimate is poor and the corresponding values should be discarded. This is of course not a problem: since the noise is known to have a smooth spectral content, it is legitimate to discard the data and fill-in by linear interpolation (in log-log units).
Nota: since the DAQ has 4 input channels, the last one has been fitted with a 50ohm shunt, so that digitizer’s noise could be estimated. This confirms that the overall measurement noise levels are dominated by the transducer’s electronics.