# Reducing noise on several Transfer Function measurements

INTRODUCTION

I want to obtain the transfer function of certain system (a ground "path", in this case) using an impact hammer for the excitation and one accelerometer (measuring vertical direction only) for the response.

I performed 20 impacts and recorded those 20 force and acceleration signals since, in theory, that helps reducing noise by performing signal averaging.

I perform signal averaging in the frequency domain both for the auto and cross spectra of the force and acceleration signals. The way I do this is by calculating the arithmetic mean accross the 20 signals, for each frequency bin.

PROBLEM

The noise level is quite big compared to that of the excitation response as it can be seen in the next figure (no clear response to the excitation is observed), so I am not sure if signal averaging is enough to solve the issue.

QUESTIONS

1) Am I implementing signal averaging correctly?

2) Is there any other way to reduce the noise present in the signals using the information that having 20 recordings of the "same" transfer function may provide?

• Did you use a filter matched to the spectral density of the impact? Do you expect non-linear harmonics in the response? – Tony Stewart Sunnyskyguy EE75 Jan 28 '19 at 22:05
• @SunnyskyguyEE75, sorry I don't understand the first question, however, I did not use any filter. And no, I don´t expect non-linear harmonics in the response. – sdiabr Jan 29 '19 at 8:03
• Well the best way to reduce noise is use a filter matched to the shape of the input signal spectral shape then reject noise outside – Tony Stewart Sunnyskyguy EE75 Jan 29 '19 at 14:12
• If I understand correctly, you mean to use a force window to reject the samples after the impulse, which are only noise. That is true, however, the main noise problem remains in the acceleration signal. – sdiabr Jan 29 '19 at 14:40
• No the acceleration signal needs a special Low Pass Filter whose shape is somewhat similar to the Fourier spectra of the range of force signals Compare Fourier spectra of input and output to get a desired filter response – Tony Stewart Sunnyskyguy EE75 Jan 29 '19 at 14:42