Aliasing
Aliasing is the signal distortion that occurs when a continuous signal is sampled at a rate insufficient to capture its full frequency content, causing signal components at frequencies above the Nyquist frequency (fN = 1 / (2 × Δt), where Δt is the sample interval) to be misrepresented as lower-frequency artefacts in the sampled record. The phenomenon was rigorously described by Claude Shannon in his 1949 sampling theorem, which proved that a signal of maximum frequency fmax can be perfectly reconstructed from its discrete samples only if the sampling frequency fs = 1/Δt is at least 2 × fmax. When this condition is violated, a signal at frequency fsignal > fN appears in the sampled data at the alias frequency falias = fs - fsignal (for the first-order alias when fsignal < fs). For example, in a seismic record sampled at 4 ms (Δt = 0.004 s, fs = 250 Hz, fN = 125 Hz), a genuine seismic event at 150 Hz aliases to 100 Hz (250 - 150), a 180 Hz event aliases to 70 Hz (250 - 180), and a 220 Hz event aliases to 30 Hz (250 - 220), all within the primary reflection frequency band and indistinguishable from genuine low-frequency signal without independent knowledge of the pre-sampling content. Aliasing is an irreversible distortion: once a signal has been sampled below its Nyquist rate and aliased frequency components are mixed with the genuine low-frequency signal, the two cannot be separated by any post-sampling processing. Prevention requires applying an anti-aliasing (alias) filter before digitisation that removes all energy above fN. In seismic exploration, aliasing occurs in both the temporal domain (insufficient time sampling by the recording instrument) and the spatial domain (insufficient geophone spacing to represent steeply dipping reflections), and each domain requires a different anti-aliasing strategy appropriate to the acquisition geometry and processing objectives.
Key Takeaways
- Temporal aliasing in seismic recording is prevented by the anti-aliasing filter in the recording instrument, but spatial aliasing depends on geophone spacing and reflection dip, and cannot be prevented by field hardware alone: The temporal alias filter in a seismic recording channel is an analogue or digital low-pass filter that attenuates all frequencies above fN before the ADC digitises the signal. Modern instruments achieve greater than 120 dB rejection at fN, making temporal aliasing a solved problem in contemporary seismic acquisition. Spatial aliasing, however, is determined by the relationship between receiver (geophone) spacing and the apparent velocity of events on the shot gather: the spatial Nyquist wavenumber kN = 1/(2 × Δx) in cycles per metre, where Δx is the spatial sample interval (geophone spacing). A 25 m geophone spacing gives kN = 0.02 cycles/m; an event with apparent velocity Vapp and temporal frequency f has spatial frequency f/Vapp, and aliases if f/Vapp > kN. For ground roll at 300 m/s and 40 Hz, the spatial frequency is 40/300 = 0.133 cycles/m, far above the 0.02 cycles/m Nyquist for 25 m spacing, so ground roll is spatially aliased on virtually all conventional WCSB land seismic surveys, making its separation from primary reflections in the F-K domain difficult.
- Spatial aliasing of dipping reflection events occurs when the geophone spacing is too large to represent the horizontal wavenumber content of the event at the highest temporal frequency in the data, and the dip angle at which aliasing begins decreases with increasing temporal frequency: For a reflection event of true dip α in a medium with velocity V, the apparent velocity on the surface is Vapp = V / sin(α) for a 2D survey geometry. The event aliases when f / Vapp > kN, i.e., when f > kN × Vapp = V/(2 × Δx × sin(α)). For a 25 m geophone spacing, V = 2,000 m/s, and α = 30 degrees dip, aliasing begins at f > 2,000 / (2 × 25 × 0.5) = 80 Hz. At shallower dips (10 degrees), aliasing begins at f > 231 Hz (above the normal seismic bandwidth, so no aliasing). At steeper dips (45 degrees), aliasing begins at f > 57 Hz, meaning that most of the reflection energy from this event is spatially aliased even at conventional seismic frequencies. In the steeply dipping Foothills of Alberta, where structural dips of 30 to 70 degrees are common, spatial aliasing of steeply dipping reflections is a persistent challenge in 2D and 3D seismic imaging that requires dense receiver sampling (Δx ≤ 10 m) for reliable F-K processing.
- In the seismic F-K domain, aliased energy wraps around the Nyquist wavenumber kN and appears at the opposite wavenumber, creating apparent dips in the opposite direction from the true event and contaminating migration velocities: On a 2D shot gather F-K display, aliased energy from a right-dipping event (positive wavenumber) folds into the left-dipping quadrant (negative wavenumber), appearing as a false dip opposite to the true event. If this aliased energy is included in a migration velocity analysis, it introduces spurious velocity picks that may cause the migrated image to focus dipping reflectors at incorrect positions or create migration artefacts (smiles, swings) in the final stack. F-K dip filtering to remove unwanted events such as ground roll must account for the aliased copies of those events that appear at folded wavenumbers: a reject fan targeting the right-dipping ground roll must include both the primary F-K lobe (positive wavenumbers at the ground-roll velocity) and the aliased lobe (negative wavenumbers where the aliased copies reside), otherwise the filter is incomplete and aliased ground-roll energy contaminates the stack.
- Spatial aliasing in 3D seismic crossline directions is a major limiting factor in Montney and Duvernay 3D acquisition, where crossline receiver line spacings of 100 to 200 m make crossline spatial aliasing occur at much lower frequencies than inline aliasing: In a standard WCSB Montney 3D survey design with 50 m inline trace spacing and 100 m crossline receiver line spacing, the inline spatial Nyquist is kN,inline = 0.01 cycles/m (aliasing above 40 Hz for a 30-degree event at 2,000 m/s) and the crossline Nyquist is kN,cross = 0.005 cycles/m (aliasing above 20 Hz for the same event). This means that virtually all dipping events are spatially aliased in the crossline direction at normal seismic frequencies, severely limiting the effectiveness of crossline F-K noise filtering and requiring pre-stack crossline interpolation using trace-interpolation algorithms (MWNI, POCS, Fourier reconstruction) that generate synthetic traces at 25 to 50 m crossline spacing before migration. The cost of crossline trace interpolation adds approximately CAD 80,000 to 160,000 to the processing budget for a 150 km² Montney 3D survey, but is considered essential for achieving adequate crossline resolution for AVO analysis and reservoir delineation.
- Well-log spatial aliasing occurs when logging speed is too high relative to the formation heterogeneity scale, causing apparent log smoothing or artificial low-frequency cycles that are misinterpreted as real geological variation: For a wireline density tool with 1 ft (0.3 m) depth sampling and an averaging volume of 0.5 m, the spatial Nyquist is kN = 1/(2 × 0.3) = 1.67 cycles/m. Formation beds with thickness below 0.3 m (thinner than the sampling interval) will be aliased: a 0.15 m shale lamina repeating every 0.30 m (3.33 cycles/m) aliases to 3.33 - 2 × 1.67 = 0 cycles/m (a constant value), appearing as a uniform average density rather than alternating shale-sand densities. This log smoothing is one of the reasons that log-derived porosity overestimates true porosity in thinly laminated Montney and Spirit River tight-gas sands where shale laminations are below 0.3 m, and Pickett crossplot interpretations based on aliased log data systematically underestimate water saturation, a known bias that completion engineers must account for when designing hydraulic fracture volumes in laminated Montney intervals.
Temporal Aliasing: The Fundamental Sampling Problem
The mathematics of temporal aliasing follows directly from the Fourier transform of a sampled signal. A signal x(t) sampled at interval Δt produces a sampled sequence x[n] = x(n × Δt) whose frequency spectrum X(f) is periodic with period fs = 1/Δt: Xsampled(f) = sum over n of X(f - n × fs). If X(f) is non-zero outside the range [-fN, fN], the periodic copies of the spectrum overlap, mixing frequency components from adjacent periods and making the original spectrum unrecoverable. This spectral overlap is aliasing, and it is irreversible because the mixed spectrum cannot be unmixed without knowing the original pre-sampling content.
The practical prevention strategy is therefore mandatory pre-sampling filtering: the alias filter removes all signal outside [-fN, fN] before the ADC samples the signal, ensuring that the periodic copies of X(f) do not overlap and the original signal can be perfectly reconstructed by the sinc-function interpolation (Whittaker-Shannon interpolation). Modern seismic recording systems implement the alias filter as a fourth- or fifth-order digital Butterworth filter within the sigma-delta modulator's decimation chain, achieving greater than 120 dB stop-band rejection at fN with linear phase response that preserves wavelet shape for subsequent processing.
Spatial Aliasing and Acquisition Design
Spatial aliasing affects both 2D and 3D seismic surveys and is the primary design constraint on receiver spacing in surveys targeting steeply dipping geological features. The conventional approach to spatial aliasing is to design receiver spacing small enough to avoid aliasing at the highest frequency of interest for the maximum expected structural dip in the survey area. For Foothills Alberta surveys targeting Devonian carbonates at dips of 40 to 60 degrees, this requires receiver spacings of 5 to 10 m to avoid aliasing above 50 Hz, compared to the 50 m spacing standard on flat-lying WCSB Montney and Duvernay surveys. The increase in trace density from 50 m to 5 m spacing multiplies the number of traces (and recording channels) by 100, increasing survey acquisition cost by 20 to 50 times for the same coverage area.
Practical compromises include using dip-moveout (DMO) correction to partially unmix aliased dipping events from flat events before F-K filtering, and using multi-channel trace interpolation algorithms (minimum weighted norm interpolation, POCS) to reconstruct missing spatial frequencies and synthesise traces at finer spacing than the acquisition grid. These processing approaches can recover some of the spatial information lost to aliasing but cannot perfectly restore the pre-aliasing content, and their effectiveness diminishes rapidly when the aliasing exceeds two to three octaves of spatial frequency above the Nyquist.
Fast Facts
Claude Shannon published the sampling theorem in 1949 in the Proceedings of the IRE as part of his landmark paper on information theory, although an equivalent theorem was independently published by Harry Nyquist in 1928 (the Nyquist rate) and Vladimir Kotelnikov in the Soviet Union in 1933 (the Kotelnikov theorem). In audio engineering, aliasing is commonly demonstrated by undersampling a high-pitched tone (for example, sampling a 600 Hz tone at 1,000 samples per second produces an alias at 400 Hz, easily audible as a different pitch). In seismic, aliased ground roll on 1970s and 1980s shot records produced characteristic "alias patterns" visible as apparent dips opposite to the true ground-roll direction, which were incorrectly identified as reflections by some early interpreters unfamiliar with the F-K aliasing phenomenon. The SEG Technical Standards Committee's recommended acquisition practice document (SEG-D, most recent revision 2016) requires that alias filter -3 dB frequency and roll-off slope be documented in every digital recording system parameter record as a mandatory quality-control field. The CSEG Geoconvention 2019 included a dedicated session on spatial aliasing solutions for Montney 3D acquisition, covering MWNI and compressive sensing techniques for recovering aliased crossline content from sparse receiver line surveys, reflecting the economic importance of reducing crossline trace spacing cost while maintaining interpretable image quality for AVO-sensitive Montney and Duvernay plays in the WCSB.