[SciPy-user] Filtering high frquency noise

iCy-fLaME icy.flame.gm at gmail.com
Mon Apr 3 06:27:18 CDT 2006


Assume that the noise power is constant though out your measurement,
and relatively small compare to the signal level, using cubic spline
line interpolation with the weight option might be a fast solution.

The weight is the standard deviation of the noise, this is very easy
to find if you have a bit of signal clear of signals, otherwise it
doesnt take long to guess a close enough value for it.

The advantage of this method is you dont have to worry about the group
delay in with filters, or dispersion to your signal.

Take a look the my data as example, it might help you to choose what to do:
This is the orginal signal as received, 1024 points.
https://warwickultrasound.co.uk/smooth/Org_1024.png

This is the result after a moving average filter:
https://warwickultrasound.co.uk/smooth/Avg_1024.png

This is the result after a badly done FIR filter:
https://warwickultrasound.co.uk/smooth/FIR_1024.png
NB: this is a very bad example, i am sure a carefully choosen FIR
filter can do much better than this.

This is the result after using the cubic spline method i described above:
https://warwickultrasound.co.uk/smooth/Cub_1024.png


Looking at the defference between orginal and filtered signal, i
choose to use the cubic spline method, because the residual seem
uniform enough not to include any signals. i.e. only the noise are
filted out.

--
iCy-fLaME
The body maybe wounded, but it is the mind that hurts.


More information about the SciPy-user mailing list