How to choose the right timing constant for RMS (in ms) using FFT (MotionSP library)

JTam.1 · ‎2021-04-29

There are many options for choosing the timing constant for RMS when using FFT. How to choose the right one? Why is the purpose of the timing constant for RMS? Please advise. Thanks!

Eleon BORLINI · ‎2021-05-05

Hi @JTam.1 ,

uhm... that looks like a Hanning time window for FFT: if so, this defines the time frame for which the FFT is calculated, and not the time base. So basically for calculating the FFT you take the number of point (in the time buffer) that are contained in T, and calculate the FFT on these points.

A time constant for RMS is usually the timeframe containing the points for which the RMS is calculated (but always in time, no FFT).

By the way, could you link the source of your picture?

-Eleon

View solution in original post

Eleon BORLINI · ‎2021-04-30

Hi @JTam.1 ,

In general, the timing constant for RMS defines the time base for the processed signal, in the sense that in this interval a number of samples determined by RMS_time* 1/ODR is submitted to the RMS calculation.

So, the FFT df -or resolution- of the output of this calculation will be, for Nyquist, 2/RMS_time.

Not sure I have answered to your question, however...

I'll ask internally to our experts and come back to you, in case.

-Eleon

JTam.1 · ‎2021-05-03

Hi Eleon, If I use 50ms of RMS time constant, the number of samples will be 1.87 x 10^-6 (50ms * 1/26667Hz). Is it right? Please advise. Thanks!

Eleon BORLINI · ‎2021-05-05

Hi @JTam.1 ,

I see you posted a new post HERE. Can I refer to that one and close this one?

-Eleon

JTam.1 · ‎2021-05-05

Hi Eleon, I still don't know how to set it (Timing constant for RMS). Please help to describe it by example. Thanks!

JTam.1 · ‎2021-05-05

Timing constant for RMS

JTam.1 · ‎2021-05-05

Does the time constant for RMS mean below T? Please advise. Thanks!

Eleon BORLINI · ‎2021-05-05

Hi @JTam.1 ,

uhm... that looks like a Hanning time window for FFT: if so, this defines the time frame for which the FFT is calculated, and not the time base. So basically for calculating the FFT you take the number of point (in the time buffer) that are contained in T, and calculate the FFT on these points.

A time constant for RMS is usually the timeframe containing the points for which the RMS is calculated (but always in time, no FFT).

By the way, could you link the source of your picture?

-Eleon

JTam.1 · ‎2021-05-06

Source: Capacitive MEMS accelerometer for condition monitoring_White paper (st.com) (Figure 7)

Eleon BORLINI · ‎2021-05-06

Ok so my previous comment was correct.

I believe that the RMS calculation, and the relative time base, should be tuned on your application: if you need to detect low frequencies, you can afford to use a lot of samples for averaging and cleaning the noise effect. Otherwise, you could only use few samples to keep the resolution in frequency high.

-Eleon