SAC Command Reference Manual

BANDPASS

SUMMARY

Applies an IIR bandpass filter.

SYNTAX

[B]AND[P]ASS {[BU]TTER|[BE]SSEL|C1|C2}
      {[C]ORNERS v1 v2}
      {[N]POLES n}
      {[P]ASSES n}
      {[T]RANBW v} {[A]TTEN v}

INPUT

BUTTER:Apply a Butterworth filter.
BESSEL:Apply a Bessel filter.
C1:Apply a Chebyshev Type I filter.
C2:Apply a Chebyshev Type II filter.
CORNERS v1 v2:Set corner frequencies to V1 and V2.
NPOLES n:Set number of poles {range: 1-10}.
PASSES n:Set number of passes {n=1: causal, n=2: zero-phase}.
TRANBW v:Set the Chebyshev transition bandwidth to v.
ATTEN v:Set the Chebyshev attenuation factor to v.

DEFAULT VALUES

BANDPASS BUTTER CORNER 0.1 0.4 NPOLES 2 PASSES 1 TRANBW 0.3 ATTEN 30.

DESCRIPTION

A set of Infinite Impulse Response (IIR) filters is available in SAC. These recursive digital filters are all based upon classical analog designs: Butterworth, Bessel, Chebyshev type I, and Chebyshev type II. These analog prototype filters are mapped to digital filters via the bilinear transformation, a transformation which preserves the stability of the analog prototypes. A complete description of this method of design can be found in the reference given below. However, it is not necessary to read that description, unless you want complete control over the more complicated Chebyshev filters.

Generally speaking, the Butterworth filter is a good choice for most applications, since it has a fairly sharp transition from pass band to stop band, and its group delay response is moderate. The Butterworth filter is the default filter type. It's 3 db point is at the designated cutoff frequency. The Bessel filter is best for those applications which require linear phase without twopass filtering. It's amplitude response is not very good. The SAC Bessel filters have been normalized so that their 3 db points are also at the designated cutoff freqency. The two Chebyshev filters are included for situations which require very rapid transitions from pass band to stop band. Although they have good magnitude discrimination, their group delay responses are the worst among the filters contained in SAC.

Some caution must be exercised in applying these filters. First, all recursive filters have non-linear phase, which can result in some dispersion of filtered waveforms. For applications where the phase of the resulting filtered waveform is important, a zero-phase implementation of the recursive filters is provided. Zero-phase filtering is possible by running the filter forward and backward over the data, instead of just forward over the data. This two-pass operation results in a effective filter magnitude response which is the square of the original magnitude response. It also results in a non-causal filter impulse response, which can leave a signal containing a sharp time onset with a ringing precursor. For this reason, you should not measure arrival times of data that has been filtered using this two-pass option. For cases where signal precursors cannot be tolerated, such as onset picking operations, it may not be a good idea to do two-pass filtering. Second, the filters can become numerically unstable if the width of the filter pass band is very small compared to the folding frequency of the data. The problem is only aggravated by increasing the number of poles in the filter. Situations that seem to require an exceptionally narrow band filter can be handled more reliably by decimation, filtering with a filter of more moderate band width, and interpolation to the original sampling rate. Recourse to this resampling strategy should be made when the filter band width drops below a few percent of the folding frequency.

Generally, the filter will have a sharper transition from pass band to stop band as the number of poles is increased. However, there are penalties for using a large numbers of poles. Filter group delays generally get wider as the number of poles increases, resulting in worse dispersion of the filtered waveform. Applications that appear to require more than three or four poles should probably be reconsidered.

The design of Butterworth and Bessel filters is particularly simple. You simply specify the cutoffs of the filter and the number of poles. Chebyshev filters are more complicated to design. In addition to cutoffs and number of poles, you must supply a transition band width, and a stop band attenuation factor for the analog prototype filter. The transition band width is the width of the region between the filter pass band and stop band. It is specified as a fraction of the analog prototype pass band width.

Due to the non-linear warping of the frequency axis of the bilinear transformation, the transition band width of the recursive digital filter may be smaller than that specified in the design. In SAC, the analog prototype filter cutoffs are compensated to ensure that they map to the requested cutoffs after the bilinear transformation is performed. The same is not true of the stop band edges. Consequently, if precisely located stop band edges are necessary, you must compensate for this shrinkage when choosing your cutoffs.

The stop band attenuation is specified as the ratio of the pass band gain to the stop band gain.

EXAMPLES

To apply a four-pole Butterworth with corners at 2 and 5 Hz.:

SAC> BANDPASS NPOLES 4 CORNER 2 5

To later apply a two-pole two-pass Bessel with the same corners.:

SAC> BP N 2 BE P 2

ERROR MESSAGES

  • 1301: No data files read in.
  • 1306: Illegal operation on unevenly spaced file
  • 1307: Illegal operation on spectral file
  • 1002: Bad value for
    • corner frequency larger than Nyquist frequency.

HEADER CHANGES

DEPMIN, DEPMAX, DEPMEN

LATEST REVISION

January 8, 1983 (Version 8.0) Magnitude Frequency Response of Chebyshev Type II Filter