How to Filter a Noisy Acceleration Signal?

I have recorded acceleration signal of a dynamical system with acceleration transducers, but these data are noisy. how can i remove this noise? Sampling frequency is 2 kHz, recording time 25 seconds, the system is a time varying system so the acceleration signal is non-stationary signal. the system had been excited with a random signal. i want to eliminate the signal noise totally if it is available, because my analysis method is very sensitive to noise, so i want to remove noise as much as possible. i think the systems first natural frequency range is between 130-170 Hz.(because system is time varying)

[Star Strider]

There are usually a number of high-frequency noise components in accelerometer signals, unique to each channel. A bandpass filter usually works, but there’s not enough information here (sampling frequency, the signals themselves) to provide a definitive reply.

[Javad Pourhosein]

sampling frequency is 2 kHz, recording time 25 seconds, the system is a time varying system so the acceleration signal is non-stationary signal. the system had been excited with a random signal. i want to eliminate the signal noise totally if it is available, because my analysis method is very sensitive to noise, so i want to remove noise as much as possible. i think the systems first natural frequency range is between 130-170 Hz.(because system is time varying)

Answer 1

[Image Analyst]

Do you want to identify and replace outliers, replacing with smoothed values? Or just totally eliminate outliers? Or do you want to smooth the whole curve, like with the Savitzky-Golay filter (sliding polynomial) in the Signal Processing Toolbox?

See http://www.mathworks.com/matlabcentral/fileexchange/3961

[Javad Pourhosein]

i want to eliminate outliers as much as possible and have a signal with high SNR.

[Image Analyst]

Thanks for Accepting the answer. Since you have no follow up questions I assume Brett's "deleteoutlier" function worked for you. I'm attaching two Savitzky-Golay demos (one for images and one for a noisy sine wave), just in case anyone wants them.

[Javad Pourhosein]

I saw the demos, thank you. can you explain more about savitzky-golay function? for example in "y = sgolayfilt(x,k,f,w" what are the k,f,w parameters. and how i can i choose the right value for them. because my signal is non-stationary and i can not tell what is the best, by just adjusting random values for these parameters.

[Image Analyst]

They're descibed here http://www.mathworks.com/help/signal/ref/sgolayfilt.html

x is your signal, k is the window width over which to do the polynomial fit, and f is the order of the polynomial. Don't worry about w or dim - just leave those off in most cases.

If k is bigger, it's taking into account more data so the signal will be smoother. Smaller k means less smoothing.

If f is smaller it will be fitting a line or quadratic and will be smoother. If f is larger you're fitting a higher order polynomial that will be better able to follow your signal and thus provide less smoothing.

[Javad Pourhosein]

Thank you for the answer.