I have data collected from an experiment of shaking a 4-storey steel frame using a shaker and responses were recorded using accelerometer sensors. I realised that in the dataset some of the signals behave randomly and some behave as impulse response signals, as shown in the figures below. I was planning to investigate the idea of using the Random Decrement Techine (RDT) (RDT link) in the random signals, however, as some of the signals are already impulse I wouldn't want to implement the RDT in impulse signals.

The data I am using is from the following experiment and link of the dataset can be found in the second link. The structure is in the following figure:

Four-Storey steel frame experiment

Experimental Data Sets section files

Four-Storey steel frame experiment

So, my question is if there is a way to differentiate the random signals to impulse signals? Also, if this can be done through Matlab or python?


Random signal

Impulse response

  • $\begingroup$ would it be possible to know more details? for instance 1.- what is the pulse repetition of the input pulsed signal 2.- if the structure is 'large' it may be that you are hitting way too often and what you are actually hearing is the same input signal bouncing back, because not enough time left to listen to whatever signal caused by the system WITHOUT input, which is in turn what actually gives the decay and may allow to calculate the damping factor you are after 3.- does the device under test has mechanical 'active' loads like rolling parts, loose boards? $\endgroup$ Commented Oct 2, 2022 at 15:43
  • $\begingroup$ @JohnBofarull See the updated question $\endgroup$
    – WDpad159
    Commented Oct 3, 2022 at 12:20
  • $\begingroup$ ok thanks for question update. With some data available I'd try check if what you are recording the the echo of the same input pulse. $\endgroup$ Commented Oct 3, 2022 at 16:24
  • $\begingroup$ @JohnBofarull You are welcome. I appreciate the assistance. $\endgroup$
    – WDpad159
    Commented Oct 3, 2022 at 16:49

1 Answer 1


Your best shot is probably to put an accelerometer directly on the shaker and use this as a reference signal.

You can than calculate the normalized cross correlation between any signal and the reference, which tells how much of the energy is causally related to the input (instead of being noise).

If you can't get a good reference signal, you can key of other signal features but that's more complicated and has to be tailored to your specific signal properties

  • $\begingroup$ Well, I think in the dataset I do have the input response because there is an accelerometer attached to the shaker which might be the first accelerometer sensor. So, are you saying that I do normalisation of the signals then implement cross correlation between input and each output responses? $\endgroup$
    – WDpad159
    Commented Oct 3, 2022 at 12:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.