Band-Pass Filter: Seismic Processing, Frequency Selection, and Noise Suppression
A band-pass filter is a signal-processing operator that transmits frequencies within a defined range (the passband) while attenuating all frequencies below the low-cut corner frequency and above the high-cut corner frequency of the filter. In reflection seismic acquisition and processing, band-pass filtering is among the most universally applied data conditioning operations: it removes low-frequency noise sources (ground roll, air blast, surface wave energy, equipment vibration) that dominate below 8 to 15 Hz in most land seismic environments, and removes high-frequency noise (shot noise, cable strum, aliasing artefacts, instrument electronics noise) that overwhelms the reflection signal above 80 to 120 Hz in typical WCSB land surveys. The result is a cleaner seismic record in which the signal-to-noise ratio within the reflection frequency band is improved before the data undergoes velocity analysis, stacking, migration, and attribute extraction. Band-pass filters are also applied in wireline acoustic logging, electromagnetic survey processing, and production data time-series analysis to isolate signals of interest from noise sources occupying different frequency ranges.
A band-pass filter is defined by four parameters: the low-cut corner frequency (f1) below which attenuation begins, the low-pass frequency (f2) below which the signal is passed at full amplitude, the high-pass frequency (f3) above which full amplitude signal is passed, and the high-cut frequency (f4) above which attenuation is complete. Between f1 and f2, and between f3 and f4, the filter transitions from full attenuation to full transmission according to the filter's rolloff characteristic, typically expressed in decibels per octave. Common specifications in WCSB seismic processing are given as trapezoidal filter panels such as 5-10-70-90 Hz, meaning the filter reaches full transmission at 10 Hz (with rolloff below 10 Hz reaching zero transmission below 5 Hz) and begins attenuating at 70 Hz (with rolloff reaching zero transmission at 90 Hz). The flatness of the passband, the steepness of the rolloffs, and the phase response of the filter are all design parameters that affect the quality of the processed data: a zero-phase band-pass filter applies symmetric time-domain effects and preserves the peak position of reflections at their true two-way travel time, while a minimum-phase filter shifts reflection peaks and must be accounted for in wavelet extraction and well-tie workflows.
Key Takeaways
- Filter design and rolloff characteristics: The steepness of the rolloff in a band-pass filter is measured in decibels per octave (dB/oct) or dB per decade, and reflects how quickly the filter attenuates frequencies outside the passband. A first-order Butterworth filter achieves 6 dB/oct rolloff; a fourth-order Butterworth achieves 24 dB/oct. Steeper rolloffs provide sharper frequency selection but introduce more group delay (phase distortion) in the transition bands, which distorts the wavelet shape and can create ringing artefacts in the filtered data near sharp onset events. In seismic processing, a balance between steep rolloff (for aggressive noise suppression) and minimal phase distortion (for wavelet integrity) is achieved using zero-phase implementations of the filter applied in the frequency domain rather than recursive time-domain filtering: the frequency-domain implementation multiplies the data spectrum by the filter's amplitude response, then applies the inverse Fourier transform, inherently producing a zero-phase output with no group delay distortion even for steep rolloffs.
- Frequency-dependent noise sources requiring band-pass suppression: In WCSB land seismic acquisition, the principal low-frequency noise sources are ground roll (Rayleigh surface waves with dominant frequency 5 to 20 Hz and apparent velocities of 300 to 600 m/s in shallow soil), near-surface guided waves in weathered layers (5 to 30 Hz), air blast from the seismic source (5 to 15 Hz), and cultural noise from roads, wind farms, and pipeline vibration (0.1 to 10 Hz). At high frequencies, primary noise sources include random environmental noise (wind, grass) with approximately flat spectra up to 200 Hz, cable strum on buried geophones excited by ground movement (resonances at 40 to 100 Hz), and acquisition aliasing from insufficient receiver spacing for high-frequency signal (aliased energy folds back into the data at frequencies determined by the spatial sampling rate). The band-pass filter's low-cut corner is set to suppress the specific dominant noise type at a given survey location, typically 5 to 15 Hz in the low cut and 75 to 100 Hz in the high cut for a standard Montney-target survey in northeast BC.
- Effect on seismic resolution and wavelet shape: The bandwidth of the applied band-pass filter directly controls the vertical resolution of the resulting seismic section, as narrower bandwidth produces longer-duration wavelets with larger side lobes that interfere with adjacent reflection events. A 5-10-80-90 Hz band-pass filter applied to a WCSB Duvernay dataset produces a zero-phase wavelet with a main-lobe width of approximately 16 ms, enabling separation of reflectors approximately 18 to 35 m apart depending on impedance contrast. Narrowing the passband to 5-10-50-60 Hz to suppress high-frequency noise in a noisy survey extends the wavelet main lobe to approximately 25 ms and raises the tuning thickness to approximately 28 to 55 m, with a corresponding loss of ability to resolve individual Montney or Duvernay sub-zones of 8 to 20 m thickness. The trade-off between noise suppression and resolution loss is the central challenge of band-pass filter parameter selection in WCSB thin-bed reservoir characterisation studies.
- Application in formation evaluation and production monitoring: Band-pass filtering is not limited to seismic applications. In wireline acoustic logging, the raw sonic waveform is band-pass filtered before semblance processing to isolate the mode of interest: a 5 to 15 kHz band-pass filter extracts compressional head waves from a monopole log while suppressing borehole flexural and Stoneley energy that would contaminate the P-wave slowness reading; a 1 to 5 kHz band-pass filter extracts Stoneley wave energy for permeability estimation in Montney tight siltstone intervals. In production data analysis, band-pass filtering of wellhead pressure time series is used to distinguish daily injection schedule variations (cycles of hours to days) from reservoir response signals (cycles of weeks to months) and long-term depletion trends (cycles of months to years), enabling separate analysis of short-term injectivity responses and long-term reservoir pressure behaviour from a single continuous pressure record.
- Adaptive and time-varying band-pass filters: A fixed band-pass filter applies the same frequency corners to all traces and time windows in a seismic dataset, which is sub-optimal because the signal band shifts with depth due to earth attenuation: the reflection signal from shallow horizons has more high-frequency content than the signal from deep horizons that have suffered additional attenuation. Adaptive or time-varying band-pass filters address this by automatically adjusting the filter corners as a function of travel time, typically lowering the high-cut corner at later times to match the earth-attenuated spectrum of the recorded signal at each depth level. Modern time-varying band-pass implementations use time-frequency analysis (such as the short-time Fourier transform or the S-transform) to estimate the local spectrum of the signal at each travel-time window, then apply a filter whose corners are updated every 50 to 200 ms to track the evolving signal band. This approach has been shown to improve signal-to-noise ratio by 2 to 4 dB relative to fixed filters in WCSB deep Montney surveys where the signal band narrows significantly from 30 Hz at 1,000 m to 50 Hz at 3,500 m depth due to Q attenuation in the Devonian carbonates.
Band-Pass Filter Panels and Parameter Selection in Seismic Processing
The selection of band-pass filter parameters for a WCSB seismic processing project is guided by three complementary analyses: frequency-wavenumber (FK) analysis of shot records to identify the dominant noise modes and their frequency content, analysis of the signal amplitude spectrum on a window-by-window basis from surface to target depth to quantify the earth-attenuation-limited signal band at each depth level, and display of multiple filter panels at different passband widths and corners to allow the interpretation team to visually assess resolution versus noise trade-offs at the target horizon before committing to final processing parameters. A typical WCSB processing workflow generates five to seven band-pass filter panels at the stack and migrated stack levels, spanning from a narrow filter (5-10-40-50 Hz) to a broad filter (5-10-90-100 Hz), allowing the processing geophysicist and the client's interpreter to identify the filter panel that best balances Montney or Duvernay event continuity with ground roll and noise suppression.
The final deliverable seismic volume is typically band-pass filtered at the conservative end of the optimal range rather than at the widest possible passband, to ensure that noise contamination in the amplitude attribute volumes extracted from the deliverable does not create false AVO or impedance anomalies that could mislead well placement decisions. In AVO analysis for Duvernay fluid discrimination, for example, a wide band-pass filter that extends to 90 Hz improves resolution but allows high-frequency noise to enter the AVO gradient calculation, which is particularly sensitive to even small amplitude variations across the offset range; a narrower 5-10-70-80 Hz filter produces slightly less resolution but more reliable gradient estimates that are better suited to distinguishing gas-condensate from brine-saturated intervals at the well-planning stage.
Band-Pass Filtering in Time-Lapse (4D) Seismic Monitoring
In time-lapse seismic monitoring of SAGD thermal recovery operations in the Cold Lake, Peace River, and Athabasca oil sands regions, consistent band-pass filtering between the baseline and monitor surveys is a critical repeatability requirement. If the baseline survey is filtered with a 5-10-80-90 Hz band-pass and the monitor survey is filtered with a different 5-10-70-80 Hz passband (perhaps because noise levels in the monitor acquisition were higher), the difference between the two volumes will contain a frequency-dependent amplitude artefact that mimics a reservoir saturation change signal and cannot be separated from genuine time-lapse amplitude differences caused by steam chamber growth. All major WCSB 4D seismic programmes, including those operated by Canadian Natural Resources Limited at Horizon, Cenovus at Foster Creek and Christina Lake, and Imperial Oil at Cold Lake, specify identical band-pass filter parameters for baseline and all monitor surveys as a non-negotiable repeatability condition. Where one survey has intrinsically lower bandwidth than another due to different acquisition geometry or weather conditions, both surveys are filtered to the narrower common bandwidth (the lowest common denominator) to ensure that all frequency-dependent processing artefacts are identical in both volumes.
The impact of band-pass filter choice on 4D SAGD amplitude differences can be quantified in a test on historical survey pairs: switching from a 5-10-80-90 Hz common filter to a 5-10-60-70 Hz common filter on a Christina Lake 4D dataset reduced the apparent average amplitude difference over the unswept region of the reservoir by approximately 40%, demonstrating that the wider filter's noise floor was contributing a false 4D signal that the narrower filter suppressed. This quantification is used to set the minimum repeatability specification (Normalised Root Mean Square, or NRMS) required for the 4D programme to be considered scientifically interpretable, with the band-pass filter parameters being one of the three primary controllable parameters (along with receiver coupling and source repeatability) in the NRMS budget for each SAGD 4D survey programme.