Linear prediction error filter design

Abstract

The design of a linear prediction error filter can be formulated as a four‐step procedure. These steps, in part, consist of a variational calculus approach that leads directly to the normal equations and an entropy conservation principle. From the normal equations, we obtain parameters called partial correlation coefficients that minimize the prediction error at the kth filter design step. Partial correlation coefficients determine the best prediction error filter for different seismic trace models (autoregressive, autoregressive‐moving average, and phase‐intercept‐shifted autoregressive traces). A result shows that simple phase distortion can increase the length of the prediction error filter and affect the estimates of the reflection sequence.