Band-Limited Function: Seismic Resolution, Wavelet Theory, and Spectral Constraints
A band-limited function is a mathematical function or time series whose spectral energy is confined to a finite range of frequencies, bounded below by a low-frequency cutoff and above by a high-frequency cutoff. Outside this frequency band, the function contains no energy; it is spectrally zero above and below its defined limits. In seismic exploration, every recorded signal is band-limited by the physics of the measurement: the seismic source generates energy only within its characteristic frequency range, the earth attenuates high-frequency energy as waves travel through rock (so deep targets arrive at the receiver with less high-frequency content than shallow targets from the same source), the recording instruments have finite bandwidth response, and anti-aliasing filters applied before digital sampling impose a hard upper frequency limit at the Nyquist frequency. Understanding band-limitation is therefore inseparable from understanding what a seismic dataset can and cannot resolve about the subsurface.
The implications of band-limitation for seismic interpretation are profound. A truly broadband signal approaching a Dirac delta function in the time domain could resolve reflectors of infinitesimal thickness. A band-limited signal cannot: its shortest resolvable time-domain feature is approximately 1 divided by the bandwidth in Hz, so a dataset spanning 10 to 80 Hz (70 Hz bandwidth) has a minimum resolvable event duration of approximately 14 milliseconds, corresponding to a bed thickness of approximately 17 to 35 m depending on formation velocity. For the Montney formation with P-wave velocity of 4,500 to 5,200 m/s, this means that the individual Montney A, B, and C silty shale sub-zones of 5 to 15 m thickness each lie below or near the tuning threshold of standard surface seismic, and their individual reflections cannot be separated without either broadband acquisition that extends the high-frequency limit beyond 100 Hz or constrained seismic inversion informed by well logs. The Nyquist theorem, which states that a band-limited function can be perfectly reconstructed from discrete samples taken at a rate of at least twice the highest frequency in the band, underpins the digital sampling design of all modern seismic acquisition systems and explains why the sample interval, not the recording bandwidth alone, determines the maximum frequency that can be captured in the digital record.
Key Takeaways
- Mathematical properties of band-limited functions: A band-limited function f(t) is one whose Fourier transform F(f) = 0 for all f outside the interval [f_low, f_high]. This constraint, in combination with the Nyquist-Shannon sampling theorem, means that the function is completely determined by its values at any set of discrete time samples spaced no more than 1/(2 x f_high) seconds apart. In practice, seismic data is sampled at 1 to 4 millisecond intervals (Nyquist frequencies of 500 to 125 Hz), which is far above the effective upper signal frequency after earth attenuation, giving comfort that the digital sampling is not the resolution-limiting step. The Fourier representation of a band-limited function also reveals that it cannot be localised in time arbitrarily tightly: the Heisenberg uncertainty principle for Fourier pairs states that the product of time duration and frequency bandwidth cannot fall below 1/4 pi, so a narrow bandwidth signal necessarily spreads over a long time window, producing the long-duration ringing wavelets seen in filtered seismic data compared to broad-bandwidth data.
- Seismic wavelet and band-limitation: The seismic wavelet recorded on a seismic trace is the band-limited impulse response of the earth's filter (the reflectivity series convolved with the source signature and instrument response, all band-limited). The zero-phase wavelet, where spectral phase is zero at all frequencies, has its energy centred on the time of the reflection and is symmetric in time, making it the optimal wavelet shape for structural and stratigraphic interpretation. The minimum-phase wavelet, where all spectral energy is concentrated at the earliest possible onset, is the natural output of impulsive explosive sources such as dynamite and must be converted to zero-phase by deconvolution during processing. The bandwidth of the wavelet directly controls its temporal duration: a 10 to 80 Hz (zero-phase) wavelet has a main lobe duration of approximately 17 ms and side lobes of approximately 12 ms each, while a 10 to 50 Hz wavelet has a main lobe of approximately 28 ms and more prominent side lobes, making it harder to separate closely spaced reflections from each other.
- Tuning thickness and the quarter-wavelength limit: The tuning thickness is the minimum bed thickness that produces two resolvable reflection peaks from the top and base of a thin layer, defined as approximately lambda/4 where lambda is the dominant wavelength of the seismic wavelet. Below tuning thickness, the top and base reflections overlap in time and their waveforms interfere, producing a single composite reflection whose amplitude and phase are functions of both the bed's impedance contrast and its thickness but whose reflector positions cannot be individually resolved. For a Montney B reservoir at 3,500 m depth with average P-wave velocity of 4,800 m/s and a dominant seismic frequency of 40 Hz, the dominant wavelength is 120 m and the tuning thickness is approximately 30 m. The Montney B pay zone in the Groundbirch area averages 8 to 14 m thick, placing it well below tuning thickness and explaining why individual Montney bench mapping from reflection amplitudes alone requires inversion or AVO analysis rather than simple pick-and-isochron methods.
- Spectral whitening and bandwidth recovery: Deconvolution and spectral whitening are processing operations applied to band-limited seismic data to recover attenuated high-frequency components and compress the wavelet toward a zero-phase spike. Spectral whitening applies a gain function in the frequency domain that boosts the amplitude of higher frequencies (which have been attenuated by earth Q filtering relative to lower frequencies) to produce a flat-amplitude spectrum across the passband, effectively recovering some of the bandwidth lost to earth attenuation without amplifying noise outside the signal band. The risk of over-whitening is that the noise floor at high frequencies is amplified along with the genuine signal, reducing the signal-to-noise ratio even as the bandwidth appears to increase; careful quality control of the whitened spectrum against the raw spectrum and against the noise level at high frequencies is required to ensure that whitening improves interpretable signal rather than simply creating a noisy high-frequency appearance to the data.
- Band-limitation in acoustic logging and its effect on Vp and Vs measurements: Wireline and LWD sonic tools are band-limited by their transducer resonance frequency and the geometry of the tool's transmitter-receiver spacing. A standard monopole sonic tool with a 3 kHz centre frequency and a 0.6 m receiver spacing samples the formation at a wavelength of approximately 0.5 m at Montney P-wave velocities of 4,800 m/s, providing depth resolution of approximately 0.25 m for changes in P-wave velocity. This is far better than surface seismic resolution, confirming the superiority of borehole acoustic measurements for identifying thin beds, fractures, and cemented intervals that are below the resolution limit of surface seismic. The band-limited nature of the sonic measurement does, however, create artefacts at sharp velocity contrasts: at a casing shoe where acoustic velocity transitions abruptly from steel (approximately 5,000 m/s) to formation (3,000 to 5,500 m/s), the band-limited tool response smears the transition over a length equal to approximately two receiver spacings (1.2 m), creating a zone of apparent velocity gradient where the true velocity contrast is a step function.
Shannon-Nyquist Theorem and Digital Seismic Sampling
The Nyquist-Shannon sampling theorem states that a band-limited continuous signal can be perfectly reconstructed from its samples if the sampling rate is at least twice the highest frequency present in the signal. In digital seismic acquisition, this theorem governs the choice of sample interval: a 2 ms sample interval (500 samples per second) can represent signal frequencies up to 250 Hz (the Nyquist frequency) without aliasing. Because the usable seismic signal band extends to at most 120 to 200 Hz in the best achievable marine broadband surveys, a 2 ms sample interval is more than adequate and is standard in most WCSB surveys. Some operators record at 1 ms intervals (1,000 Hz Nyquist) to provide margin for future spectral extension techniques that attempt to recover high-frequency information above the conventional signal band through model-based extrapolation, though the genuine signal improvement from 1 ms versus 2 ms recording in current WCSB land surveys is generally indistinguishable given that earth attenuation already limits recoverable signal to below 100 Hz at Montney depths.
The anti-aliasing filter applied before analogue-to-digital conversion is designed to prevent any energy above the Nyquist frequency from being sampled and creating aliased artefacts that fold back into the signal band at frequencies below Nyquist. For a 2 ms sample interval, the anti-alias filter must attenuate all signal above 250 Hz before sampling. Modern analogue-to-digital converters in seismic instruments achieve the required attenuation (typically 80 to 100 dB above the Nyquist frequency) with steep-rolloff Butterworth or elliptic filter designs that produce minimal group delay (phase distortion) within the signal band. The residual phase distortion introduced by the anti-alias filter is typically removed during processing as part of the instrument deconvolution step, which restores the zero-phase condition to the recorded wavelet before further signal processing operations are applied.
Implications for AVO Analysis and Rock Physics
Amplitude versus offset (AVO) analysis, which extracts information about formation lithology and fluid content from the variation of reflection amplitude with incidence angle on pre-stack seismic data, is critically sensitive to the band-limited nature of seismic recordings. AVO attributes such as intercept (A) and gradient (B) are derived by fitting a linear model to the amplitude-versus-sin-squared-theta relationship at each time sample across the common-midpoint gather. The quality of the AVO fit depends on the signal-to-noise ratio, which is directly controlled by the bandwidth of the data: narrower bandwidth produces longer-duration wavelets with more overlap between adjacent reflections (tuning interference), creating AVO artefacts where the measured amplitude variation reflects wavelet tuning rather than true formation AVO response.
For Duvernay AVO analysis aimed at distinguishing gas-condensate-saturated intervals from brine-saturated tight limestone within the same formation, the typical AVO anomaly (Class IIb or Class III) manifests as a reversal of polarity from near-angle to far-angle stacks that spans approximately 20 to 30 ms in the wavelet domain. If the seismic bandwidth is 15 to 55 Hz (40 Hz bandwidth), the wavelet duration is approximately 25 ms and the tuning width between the Duvernay top and base reflections at 8 to 12 m thickness is well below one sample interval, creating severe AVO cross-contamination between adjacent reflectors. Broadband processing extending to 75 Hz reduces the wavelet duration to approximately 15 ms, providing marginally better separation but still insufficient to fully isolate the Duvernay A, B, and C bench reflections from each other. For reliable Duvernay AVO on sub-tuning-thickness benches, seismic inversion (model-based or sparse-spike) that converts the band-limited wavelet response into a band-limited impedance estimate is required before AVO attributes are extracted, because the inversion step partially compensates for the tuning effect.