band pass filter problem in endpoints.

[labmaster]

I want to apply the band pass filter (butterworth) to my singal (linear in large scale).

Irregarless of the parameter, there was a problem of ringing (over damped) in the origin (starting point).

To overcome this problem, someone suggest to extend the original data to other side (negative X) by reflection.

I believe it is right to show the singal from zero to end point.

But, in my trial in code (attachment), the boundary effect was not removed.

How to remove the end point effect?

메시지가 07-15-2008 08:06 PM에 labmaster에 의해 편집되었음

[johnsold]

All filters have a transient at the beginning of the data set. Your data starts with a very large DC offset which aggravates the transient.

Are you trying to extract the sawtooth waveform at times 18 to 28? If so, try doing a linear fit to the original data and subtract that from the original data before filtering. You still get a transient, but the periodic component is quite visible from time 13 to the end. Fitting to a polynomial of degree two or a decaying exponential might be even better. If you have a good model for the behavior of the system which generates the signal, that can guide your choice of a fitting function.

The attached image shows the result of subtracting a linear fit.

Lynn

[Ipshita_C]

Hi Labmaster,

What does your original signal look like? It is very important to know the nature of your original signal while applying a filter to it. Typically, square waves and similar signals have a ringing effect at the edge of the pulses. Theoretically, a square wave consists of an infinite number of harmonic sine waves. These are getting filtered out which produces the ringing at the rising and falling edges. One way to get around this is to choose a cutoff frequency that is as close to the Nyquist Frequency that you can get. Let us know more about your signal if that does not answer your question.

Ipshita C.