Stabilized Inverse Q Filtering¶
Stabilized inverse Q filtering corrects seismic dispersion and attenuation while limiting amplitude gain where signal is lost beneath noise.
Core Idea¶
Stabilized inverse Q filtering corrects seismic wavelet dispersion and compensates attenuation under an estimated earth quality-factor Q, while limiting high-frequency amplitude gain where signal has sunk beneath noise. Phase correction and amplitude compensation are distinct; an unrestricted inverse may sharpen a trace but create artifacts or worsen signal-to-noise. This method restores only recoverable information.[ref-56fe30737f69][ref-537f579403ee]
Scope of Application¶
Wang's original 2003 paper reports controlled synthetic comparisons and a real low-S/N 2D land line processed with Q estimated from vertical seismic profile downgoing waves. The later original paper clarifies that phase inversion is stable while amplitude gain needs stabilization, but its full text was not directly retrievable here. A ground-penetrating-radar application is not asserted without source support.[ref-56fe30737f69][ref-537f579403ee]
Clarity¶
The synthetic Fig. 1 compares Q=400, 200, 100, 50 and 25 traces: full exact inverse produces strong artifacts while the stabilized filter controls them. In the unlike land case, Wang's Table 1 reports over 300–2000 ms +36% spectral bandwidth, +27% S/N and +162% in his defined temporal-resolution indicator. The last number is not a direct measure of recovered geological truth for every reflector.[^ref-56fe30737f69]
Manages Complexity¶
The filter uses a Q estimate to separate dispersive timing from frequency-dependent amplitude loss, then checks gain against a noise-limited recovery boundary. Surface seismic may not yield reliable Q because source, receiver and interference effects contaminate estimates; the field case instead uses VSP information. Bandwidth and S/N must be evaluated together.[^ref-56fe30737f69]
Abstract Reasoning¶
If attenuation multiplies a component by A(f,t), an inverse multiplies by 1/A; as A becomes very small, both signal and noise are enlarged. Limiting that gain sacrifices some nominal high-frequency recovery but avoids false detail. Wang's synthetic full-versus-stable comparison shows this genuine two-sided cost; the field case shows one successful balance, not a universal guarantee.[^ref-56fe30737f69]
Knowledge Transfer¶
The trace–Q–phase–amplitude–stabilization–dual-metric diagnostic transfers from known-Q synthetic traces to VSP-calibrated land data, though Q uncertainty differs. It does not justify calling generic image sharpening or unverified radar compensation the same method. The strict Filter in Signal Processing parent does not imply recovery of signal beneath noise.[^ref-56fe30737f69]
[^ref-56fe30737f69]: Yanghua Wang, “Quantifying the effectiveness of stabilized inverse Q filtering”, Geophysics 68(1) (2003), pp. 337–345, original published full text in Imperial College repository, Fig. 1 and Table 1. [^ref-537f579403ee]: Yanghua Wang, “Inverse Q-filter for seismic resolution enhancement”, Geophysics 71(3) (2006), pp. V51–V60, original indexed abstract/preview; direct full text restricted.
Relationships to Other Abstractions¶
Current abstraction Stabilized Inverse Q Filtering Domain-specific
Parents (1) — more general patterns this builds on
-
Stabilized Inverse Q Filtering is a kind of Filter (Signal Processing) Domain-specific
Stabilized inverse-Q filtering is a signal-to-signal frequency-response filter.
Hierarchy path (1) — routes to 1 parentless root
- Stabilized Inverse Q Filtering → Filter (Signal Processing) → Transformation → Function (Mapping)
Neighborhood in Abstraction Space¶
Stabilized Inverse Q Filtering sits in a moderately populated region (58th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Domain-Specific Measurement Parameters (36 abstractions)
Nearest neighbors
- Reflection Seismology — 0.88
- Seismic Interferometry — 0.87
- Seismic Inversion — 0.87
- Infrasonic passive differential spectroscopy — 0.86
- Dim Spot — 0.85
Computed from structural-signature embeddings · 2026-10-08