RE: [geonet]: Bandpass Filtering

[Bengert, Brian (Sudbury)]

You can find excelent references to this in most time series textbooks, or on the web.  Look under "Gibbs Phenomenon".   Basically multiplying the transform of a time series (A) in the frequency domain by a windowing function (W) is equivalent to convolving the time series with the transform of the windowing function in the time domain.  If you look at the transform of a simple bandpass filter (boxcar) in the time domain, you'll see that it is a sinc function (sin(t)/t).  This function has large sidelobes which produce this oscillation in your results.  To minimize this oscillation, you need to use a better filter in the Frequency domain.  Geosoft suggests either a Hanning, or a Butterworth.  Play around a bit to find the most suitable one for your problem.   Good luck Brian -----Original Message----- From: skyhunter [mailto:[EMAIL PROTECTED]] Sent: Wednesday, July 10, 2002 11:07 AM To: GEONET Subject: [geonet]: Bandpass Filtering Hi everyone,   I was wondering if anyone can explain to me in simple terms why a bandpass filter tends to oscillate. For example, when applying a bandpass filter to a simple step function between 0 and 1, the resulting bandpass filter oscillates between very negative and very positive numbers in the vicinity of the step-up and step-down. Does anyone have any ideas of how to prevent or minimize these oscillations?     Aaron Balasch Sky Hunter Technologies Inc. Suite 101, 1725 10th Avenue S.W. Calgary, Alberta T3C 0K1 email: [EMAIL PROTECTED] phone: 403-228-2175 fax: 403-244-7955