I am calculating the frequency response function of a 2 degree of freedom system from simulation data in matlab using the tfestimate function.
The input is a impulse force and the output is displacement of the two masses. The sampling rate is 10 000 Hz
However I am struggling to understand the cause and meaning of the plot of the FRF, in general the FRF plot has the shape that I would expect it to have, but when I zoom in it has small "oscillations" throughout the entire plot. What is causing the FRF to have this form and what can be done to improve it?
Dear Johan Kruger ,
welcome!
What is the frequency of this oscillations.
I think it is related tp the sam0ling frequency. It is so that you get the discretized frequency response according to the sampling frequency. The separation between the samples is equal to fs/N, where N is the number of the samples in plot.
If you change the sampling frequency or N the pattern you obtain may change.
The discrete sampled signal frequency response is different from the continous time signal because sampling leads to the generation of copying the frequency response in frequency domain. In fact the fft is valid only for frequencies up to fs/2 other wise aliasing occurs.
You have have to take into consideration the sampling theorem of Nyquist when sampling the continuous signals.
Best wishes
Dear Johan Kruger ,
welcome!
What is the frequency of this oscillations.
I think it is related tp the sam0ling frequency. It is so that you get the discretized frequency response according to the sampling frequency. The separation between the samples is equal to fs/N, where N is the number of the samples in plot.
If you change the sampling frequency or N the pattern you obtain may change.
The discrete sampled signal frequency response is different from the continous time signal because sampling leads to the generation of copying the frequency response in frequency domain. In fact the fft is valid only for frequencies up to fs/2 other wise aliasing occurs.
You have have to take into consideration the sampling theorem of Nyquist when sampling the continuous signals.
Best wishes
Spectral leakage due to windowing. See link below:
https://en.wikipedia.org/wiki/Window_function
Remember, if you believe you are not using any windows, actually you are using a rectangular window.
You find your response in this link:
https://community.sw.siemens.com/s/article/what-is-a-frequency-response-function-frf
https://blog.prosig.com/2009/10/19/a-simple-frequency-response-function/
https://sites.chemengr.ucsb.edu/~ceweb/faculty/seborg/teaching/SEM_2_slides/Chapter%2013%20%283-7-05%29.pdf
Thank you for your responses, Abdelhalim abdelnaby Zekry the peaks of the oscillations on the FRF plot are about 3.66 Hz apart, the data was sampled at 10kHz, with 65536 points used for the FFT calculations (NFFT).
Simone Orcioni The attached graphs have been done with a rectangular window, however I have tried Hamming and Hanning windows of different sizes but neither seemed to provide a better result. Is there a good method to determine the window lengths and overlap?
Harald Kirchsteiger thank you for your suggestion, the reason for using a impulse is because the main objective is to use the impulse response to obtain a inverse FRF to be used for load identification.
Dear Johan Kruger ,
Dear would you please zoon on these very low frequency periodic in the envelope of the fft signal. Is the frequency of this ripple has relation to the sampling frequency of the number of points in the fft transform?
Best wishes
- Badges
- Science topic