Smoothing a signal from accelerometer

[palkin]

Hi all,

I'm acquiring a signal from an accelerometer using PCI 6251 card. Graph below shows the signal acquired. I need a smoother curve for this data..

I was confused trying to use the curve fitting from mathematics function palette ( it asks for X,Y etc) also in express curve fitting it says my signal matrix is singular. 

thanks a lot

[Coq_Rouge]

A trick that I often use is to sample with a much higher sample rate than needed.  Then I downsample the signal by averageing x samples. Say I need a sample rate equal 50Hz. But then I sample with 1000Hz and then average together every 20 sample. This can also be combined with didgital filters. The digital filters are apllied on the high sample rate signal 

[Henrik_Volkers]

What is signal and what is noise?

I would try a FIR lowpass filter with a Kaiser window and shift back the signal half order points to correct the phase (if needed)

Something like curve fitting is the Savatzgi-Golay filter, also worth a try 😉

[palkin]

The signal is from an accelerometer fitted inside of a helmet that is dropped from a certain height.

When I use just the moving average filter, i do get a smooth curve, but the peak value has also changed (along with some shift..that is OK)...

For my application, peak value is important so it should not change...

[palkin]

Henrik,

Fir filter does not alter the magnitude that much.. I used the express VI filter FIR with 29 taps...where can i find the kaiser window? i din't find in the mathematics menu

[Henrik_Volkers]

FIR Windowed Filter VI in signal processing/Filter  or hit Ctrl-Space and use quick drop to search 🙂 has an input to define the type of window used. 

As already mentioned (but misspelled) the Savitzky-Golay Filter  is a good tool to preserve maxima.

What sample rate are you using? 

What type of sensor are you using and what uncertainty are you expecting?  I ask because maybe you need to take look at the influence from the high pass of your sensor amplifier chain (assuming charge based sensors)

Message Edited by Henrik Volkers on 03-05-2010 08:26 AM

[palkin]

Henrik,

Initially I was sampling at 5k. Later we changed the sampling to 100kHz (since we are getting a very short impulse ) Now the signal is lot better.

we are using a piezoelectric accelerometer sensor with 1msec/g.

thanks for your time