Automatic Gain Control: Definition, Seismic Amplitude, and AGC

Automatic gain control (AGC) is a time-varying gain function applied to a seismic trace that continuously adjusts the trace amplitude to maintain a relatively constant output level over a sliding time window. Mathematically, the gain G(t) at each sample time t is computed as the ratio of a desired target level to the measured signal energy (RMS or average absolute amplitude) within a window of length T milliseconds centered on sample t: G(t) = Target Level / RMS(t), where RMS(t) = sqrt( (1/N) Σx²(t±T/2) ) and N is the number of samples in the window. The gain is then multiplied sample-by-sample into the original trace: x'(t) = G(t) x(t). The effect is to equalize amplitude differences between shallow, high-energy reflections and deep, low-energy reflections, making the entire time section visually uniform and easier to interpret structurally.

AGC was developed in the early days of analog seismic recording when dynamic range limitations of paper records and early magnetic tape systems made deep reflections essentially invisible without some form of time-varying gain. Modern 24-bit seismic recording systems have dynamic ranges exceeding 120 dB, making AGC technically unnecessary for the purpose of displaying faint deep reflections. Nevertheless, AGC remains ubiquitous as a display tool because it provides a rapid, visually appealing overview of structural features. The critical distinction every geophysicist must maintain is that AGC is appropriate for structural display but is destructive to the relative amplitude information required for AVO analysis, amplitude anomaly interpretation, acoustic impedance inversion, and fluid discrimination. Applying AGC before any of these quantitative workflows irreversibly destroys the very information being sought.

Key Takeaways

• AGC is a multiplicative, time-varying gain computed from the RMS or average absolute amplitude in a sliding window; long windows (500-1,000 ms) preserve broad amplitude trends, while short windows (50-200 ms) aggressively equalize amplitude and destroy more amplitude information.
• AGC is appropriate only for structural display and first-pass QC; it must never be applied before AVO analysis, acoustic impedance inversion, amplitude attribute extraction, or any workflow that requires preserved relative amplitudes.
• True amplitude (TA) processing, the correct alternative to AGC for quantitative interpretation, applies only physically justified corrections including spherical divergence compensation, surface-consistent amplitude corrections, and absorption (Q) compensation, leaving lithology-dependent and fluid-dependent amplitude variations intact.
• The Wiener-Khinchin theorem connects autocorrelation to the power spectral density, and similarly, time-variant spectral whitening, a frequency-domain alternative to AGC, equalizes the amplitude spectrum without distorting trace-to-trace amplitude ratios needed for AVO.
• In wireline log and LWD workflows, AGC-equivalent normalization is sometimes applied to sonic and resistivity logs for display purposes, but raw log values must always be preserved in the database for quantitative petrophysical calculations.

How AGC Works: Window Length and Gain Computation

The behavior of an AGC operator is determined primarily by the window length T. A long AGC window (500-1,000 ms) computes the gain over a broad time interval, which means the gain function changes slowly. Events within this long window that differ in amplitude by factors of 2 to 5 are equalized, but amplitude trends that vary on a scale longer than the window (for example, the general amplitude decay caused by spherical spreading over the full record length) are largely preserved. A short AGC window (50-200 ms) computes the gain over a narrow interval, causing the gain to react rapidly to local amplitude variations. This aggressive equalization can make individual wavelets appear nearly equal in amplitude regardless of their origin, making bright spots, dim spots, and polarity reversals, the classic direct hydrocarbon indicators (DHIs), essentially invisible. Processors sometimes use a window as short as 20-30 ms for specific display purposes such as examining thin-bed tuning effects in a restricted time interval, but such extreme AGC is never used on data intended for any form of amplitude interpretation.

The RMS-based gain implementation is the most common in commercial seismic processing software. The RMS amplitude within the window at each sample time is computed by summing the squares of all samples in the window, dividing by the sample count, and taking the square root. The gain is then the ratio of the desired target level (often normalized to 1.0 or to the median RMS of all traces) to this local RMS value. An alternative is the average absolute amplitude (mean absolute value) implementation, which is computationally simpler and slightly more robust to isolated high-amplitude noise spikes, since squaring is avoided. Both approaches produce visually similar results for typical seismic data, but the RMS approach is theoretically preferred because it is directly related to signal energy and to the autocorrelation at zero lag (R(0) equals the sum of squared amplitudes).

Edge effects at the beginning and end of the trace require special handling. When the sliding window extends beyond the trace limits, the standard approach is either to use a one-sided window (only samples within the trace are used) or to taper the gain smoothly toward the median gain value at the trace boundaries. If edge effects are not handled, the trace endpoints may receive anomalously high or low gain, producing visible amplitude artifacts at the top and bottom of the section. This edge-effect problem is analogous to the windowing problem encountered in short-time Fourier transforms and in the design of array sonic depth-of-investigation windows, where boundary samples require special weighting.

When AGC Is Appropriate and When It Is Harmful

The appropriate uses of AGC are well-defined and limited. AGC is suitable for structural seismic interpretation when the goal is to map fault positions, horizon geometry, unconformity surfaces, and stratigraphic architecture without needing to quantify amplitude levels. It is also appropriate for first-pass QC of raw or minimally processed field data, where the processor needs to assess whether reflections are present throughout the time section and whether the acquisition geometry produced any obvious gaps or noise contamination. In these structural and QC contexts, AGC provides a rapid visual normalization that makes the data legible without any need for carefully calibrated true amplitude processing.

AGC is explicitly harmful in all quantitative amplitude workflows. For AVO analysis, the fundamental observation is that reflection amplitude changes with offset (or angle) in a manner controlled by the elastic properties (P-wave velocity, S-wave velocity, density) of the reflecting interface. If AGC has been applied, the gain function varies differently at each offset because the amplitude envelope at each offset is different. The AGC gain therefore introduces a spurious offset-dependent amplitude trend that masquerades as an AVO effect. Even if the AVO analyst attempts to remove the AGC by dividing out an estimated gain function, the process is irreversible in the presence of noise: once the AGC has been applied, there is no way to recover the original pre-AGC amplitudes because the gain function is derived from the noisy data itself. For acoustic impedance inversion and seismic-to-well tie, AGC destroys the relationship between seismic amplitude and acoustic impedance contrast that makes inversion possible. For fluid discrimination using amplitude versus frequency (spectral decomposition) methods, AGC equalizes the frequency content in a manner that confounds the detection of frequency shadows or bright spots at low frequencies associated with gas saturation.

True Amplitude Processing: The Correct Alternative

True amplitude (TA) processing preserves relative trace amplitudes by applying only those gain corrections that have a demonstrable physical basis. The standard TA processing sequence consists of the following steps applied in order: (1) geometry assignment and data editing to remove noisy traces; (2) surface-consistent amplitude corrections that account for near-surface coupling differences between sources and receivers, derived by solving a least-squares system analogous to the surface-consistent deconvolution normal equations; (3) spherical divergence correction, which compensates for the 1/r² amplitude decay (or, more precisely, 1/(v² t) correction in practice) caused by the geometrical spreading of a wavefront expanding from a point source; (4) absorption (Q) compensation, which corrects for the frequency-dependent energy loss as the wavefield propagates through anelastic rock. After these physically motivated corrections, the amplitude of a reflection is proportional to the reflection coefficient at the interface, which is in turn determined by the contrast in elastic properties across that interface.

The distinction between spherical divergence correction (applied in TA processing) and AGC (not applied in TA processing) is subtle but critical. Spherical divergence correction applies a single deterministic gain curve t x v²(t) to every trace, where t is two-way travel time and v(t) is the RMS velocity, based on the physics of wavefront expansion. This correction is the same for every trace and does not depend on the local amplitude of the data. AGC, by contrast, adapts to the actual amplitude of each individual trace in each time window, making it data-dependent and therefore trace-dependent. The data-dependent nature of AGC is precisely what destroys inter-trace amplitude relationships, while the deterministic, trace-independent spherical divergence correction preserves them.