Convergence of Time-Domain Adaptive Maximum-Likelihood Filters for Stationary Data
Abstract
The behavior of an adaptively designed time-domain maximum-likelihood multichannel filter during convergence on stationary data was examined. The covariance matrices of a measured seismic short-period prewhitened noise sample were used to generate 3300 time points of 13 channel stationary Gaussian data having the measured correlation structure. Using these data, 29-point adaptive filters were computed and applied. Their performance was evaluated as a function of time and compared with the performances of the beamsteer filter and the maximum-likelihood filter generated from the measured matrices. Beginning with a beamsteer weighted initial filter, the filter was adapted for 3272 points. After 1000 adaptions, the adaptive filter was equally effective as the optimum filter in rejecting high-frequency noise. After 3272 adaptions, low-frequency noise rejection by the adaptive filter was much poorer than that of the optimum filter and not appreciably better than that of the beamsteer filter. Wide-band noise reduction obtained by the best adaptive filter was about 3.5 db worse than optimum. The loss in performance was probably caused by incomplete convergence. Estimates of the gradient measurement noise are in good agreement with those predicted by theory.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 26, 1970
- Accession Number
- AD0870777
Entities
People
- William H. Swindell
Organizations
- Texas Instruments