Band-Limited Function: Seismic Bandwidth, Resolution, and Wavelets

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 minimum frequency (the low cut or low-end rolloff) and above by a maximum frequency (the high cut or Nyquist limit). Outside this frequency band, the function contains no energy. In seismic exploration and borehole seismic data acquisition, every recorded signal is band-limited by nature: the seismic source generates energy only within its characteristic frequency range, the earth attenuates high-frequency energy as waves travel through rock, the recording instruments have finite response bandwidths, and anti-aliasing filters applied before digital sampling impose a hard upper frequency limit equal to the Nyquist frequency. Understanding band-limited behavior is fundamental to seismic data processing, interpretation, and the resolution limits that govern what geologists can detect in the subsurface.

The concept connects directly to the Nyquist-Shannon sampling theorem, which states that a continuous signal can be perfectly reconstructed from discrete samples only if the sampling rate is at least twice the highest frequency present in the signal. Because seismic data is always band-limited before sampling, the theorem guarantees that the sampled data can theoretically represent the original continuous signal within the recorded frequency band. However, the finite width of that band imposes an irreducible limit on time-domain resolution: a truly band-limited function cannot be a perfect spike in time, but instead spreads out in time as a wavelet whose duration is inversely related to the bandwidth of the signal. This wavelet character of the seismic trace is the central physical constraint governing bed resolution and stratigraphic interpretation.

Key Takeaways

• A band-limited function has non-zero spectral energy only within a defined frequency range; seismic traces are band-limited because the source, the earth, and the recording system each impose independent frequency constraints that together define the usable bandwidth of the recorded data.
• Typical seismic bandwidths span 10 to 120 Hz for land acquisition and 5 to 150 Hz for marine air gun surveys, representing roughly 3 to 3.5 octaves; this finite bandwidth means seismic data can only resolve geological features whose thickness exceeds approximately one quarter of the dominant wavelength.
• The Nyquist frequency, equal to half the sample rate, sets the absolute upper frequency limit that can be represented in digital seismic data; energy above the Nyquist frequency, if not removed by an anti-alias filter before sampling, folds back into the recorded bandwidth as aliased noise.
• Deconvolution, spectral whitening, and bandwidth extension processing techniques attempt to broaden the effective bandwidth of the data by compressing the wavelet and flattening the frequency spectrum, improving temporal resolution within the constraints of signal-to-noise ratio.
• The zero-phase wavelet is the processing target for most modern seismic datasets because it is symmetric in time, produces the maximum temporal resolution for a given bandwidth, and places the peak of the wavelet directly at the acoustic impedance contrast rather than offset from it.

How Band Limitation Arises in Seismic Data

The band-limited character of a seismic trace results from four independent processes that each impose frequency constraints on the recorded signal. First, the seismic source generates energy only across a finite frequency range. An explosive source such as dynamite produces a broadband impulse with energy from approximately 10 Hz to 250 Hz, but the shape of the generated pulse and the coupling of the explosion to the rock depends on the local geology, borehole condition, and charge size. A vibroseis source used in land acquisition generates energy across a controlled sweep range, typically 6 to 100 Hz or 8 to 150 Hz, with the bandwidth determined by the sweep design and the mechanical capabilities of the vibrator. Marine air gun sources generate energy from approximately 5 to 150 Hz, though the usable portion of the spectrum after deghosting and bubble removal is typically narrower.

Second, the earth attenuates seismic energy in a frequency-dependent manner. Attenuation, quantified by the quality factor Q, causes high-frequency energy to be absorbed more rapidly than low-frequency energy as the seismic wave travels through rock. A wave traveling through a rock with Q of 50 will lose half its high-frequency energy amplitude over a travel path of roughly one dominant wavelength, so the effective upper frequency of the seismic wavelet decreases progressively as reflections return from deeper targets. This earth filtering effect is one reason why deep seismic events appear smoother and lower in frequency content than shallow events on the same record. Compensating for earth attenuation through Q-compensation processing can restore some of the attenuated high-frequency energy, but is limited by the noise floor of the data. The relationship between Q and bandwidth loss has been studied extensively by the Society of Exploration Geophysicists and forms part of the theoretical foundation for seismic inverse-Q filtering methods.

Third, the recording instruments themselves have finite frequency response. Geophones used in land seismic recording have a natural resonant frequency (typically 4.5 Hz, 10 Hz, or 14 Hz depending on the application) below which sensitivity rolls off steeply, and a high-frequency response that attenuates above several hundred Hz due to mechanical and electronic characteristics. Hydrophones used in marine acquisition and in vertical seismic profile surveys have a broader and flatter frequency response but still show rolloff at the extremes of their operating range. The combination of geophone response and preamplifier filtering defines the instrument passband. Fourth, an anti-alias filter is applied electronically before the analog-to-digital conversion step in the recording system, cutting off all energy above the Nyquist frequency (half the sample rate) to prevent frequency aliasing in the digital data.

Nyquist Frequency, Sampling, and Aliasing

The Nyquist frequency is the maximum frequency that can be correctly represented in a digitally sampled signal, defined as half the sampling frequency. For seismic data sampled at 2 milliseconds, the sampling frequency is 500 Hz and the Nyquist frequency is 250 Hz. For data sampled at 4 milliseconds, the Nyquist frequency drops to 125 Hz. Modern land and marine 3D seismic surveys are typically recorded at 2 ms sample intervals for shallow targets and 4 ms for deeper surveys where the dominant frequencies are lower due to earth attenuation. VSP surveys designed to capture high-frequency borehole data may be recorded at 0.5 ms or 1 ms, with corresponding Nyquist frequencies of 1,000 Hz and 500 Hz respectively.

If energy above the Nyquist frequency reaches the analog-to-digital converter without being removed by the anti-alias filter, it undergoes frequency aliasing: the energy folds back into the recorded frequency band at a mirrored frequency calculated as 2 x Nyquist minus the original frequency. For example, energy at 300 Hz recorded with a 2 ms sample rate (Nyquist 250 Hz) would alias back to 200 Hz, contaminating the legitimate data at that frequency. In practice, the anti-alias filter in the recording system removes energy above approximately 80 to 90 percent of the Nyquist frequency with a steep rolloff, preventing aliasing at the cost of a small reduction in the usable high-frequency bandwidth. The concept of aliasing applies not only in the time-frequency domain but also in the spatial domain: the spacing between receivers (trace interval) must be less than half the apparent wavelength of the fastest dipping events in the data to avoid spatial aliasing.

The amplitude spectrum of a band-limited function, plotted as a function of frequency, shows non-zero values only within the passband and zero values outside it. The shape of the amplitude spectrum within the passband reflects the combined effect of all the filters the signal has passed through, including source signature, earth response, instrument response, and any processing filters applied. A perfectly flat amplitude spectrum within the passband would correspond to the ideal white spectrum; real seismic data typically shows a roughly bell-shaped or trapezoidal amplitude spectrum with a peak near the center of the passband and rolloff toward both the low-frequency and high-frequency ends.

Wavelets, Resolution, and the Ricker Wavelet

The seismic wavelet is the time-domain representation of the band-limited pulse that conveys a reflection from an acoustic impedance boundary in the subsurface. Because the seismic trace is band-limited, the wavelet is not a perfect spike but instead has finite duration and a characteristic shape that depends on the amplitude and phase spectra of the signal. The convolution model of the seismic trace states that the recorded trace is the convolution of the earth's reflectivity series (the sequence of reflection coefficients at each acoustic impedance boundary) with the seismic wavelet, plus noise. Deconvolution processing attempts to invert this convolution, estimating the wavelet and dividing it out to recover the reflectivity series as a sharper, higher-resolution trace.

The Ricker wavelet is the most commonly used mathematical model of a zero-phase seismic wavelet. It is defined in the time domain as a function of dominant frequency (f_peak), producing a symmetric pulse with a central peak flanked by two side lobes of opposite polarity. The Ricker wavelet is the second derivative of a Gaussian function and has a closed-form expression in both the time and frequency domains, making it analytically convenient for modeling and testing. Its amplitude spectrum is approximately bell-shaped, with peak amplitude at f_peak and rolloff toward both zero frequency and the Nyquist frequency. A Ricker wavelet with a dominant frequency of 30 Hz in rock with a P-wave velocity of 3,000 m/s (9,840 ft/s) has a dominant wavelength of 100 m (328 ft) and a Rayleigh resolution limit of 25 m (82 ft).

The zero-phase wavelet is the processing objective for most modern seismic surveys. In a zero-phase wavelet, all frequency components have zero phase shift, meaning the wavelet is perfectly symmetric in time and the peak amplitude of the wavelet coincides with the position of the acoustic impedance contrast it represents. Zero-phase data simplifies interpretation because the peak of a positive wavelet indicates a hard reflection (increase in acoustic impedance with depth) and the trough of a negative wavelet indicates a soft reflection (decrease in acoustic impedance). The acoustic log or sonic and density log from a nearby well, converted to a reflection coefficient series and convolved with the estimated wavelet, produces a synthetic seismogram that can be tied to the seismic data to calibrate the phase and frequency content of the recorded wavelet. An incorrect phase assumption in interpretation leads to systematic depth errors in picking reflections and can cause acoustic impedance inversion results to have the wrong polarity.