Enhancement of Teleseismic Body Phases with a Polarization Filter

Abstract

Summary A time domain polarization filter, originally proposed by Flinn, has been modified and is used to increase the signal to noise ratio of teleseismic body phases. Both the rectilinearity and direction of particle motion is obtained from the co-variance matrix for three components of ground motion over a small time interval. The estimate of rectilinearity is found by diagonalizing the matrix and forming a function that involves the ratio of the largest and intermediate principle axes of the matrix. The direction of polarization is found from the eigenvector of the largest principle axis. A set of time-varying operators are then obtained which act as a point by point gain control to modulate the digital seismic records. This non-linear filter is useful for enhancing P, pP, sP, PP, S and other compressional or shear phases. When applied to an array of three-component stations it appears to be possible to identify multiple events in the source function of some earthquakes. 1. Iotroductioo As any seismograph record is always noise contaminated, the detection and interpretation of seismic events requires knowledge of the characteristics of both signal and noise. Elastic body waves, which may be generally separated into P (compressional) and S (shear) phases, can be considered as non-dispersive group arrivals with maximum power in the 0.3 to 10-second period range. Surface Rayleigh and Love waves may be described as dispersive group arrivals with maximum power in the 2-to 100-s period range for earthquakes of moderate magnitude, the observable periods extending up to 57 min for larger teleseismic events. Superimposed on these signals is microseismic background noise as well as signal generated noise. Signal generated noise is the result of multiple reflections and refractions of P and S body waves at crustal interfaces and inhomogeneities and local conversion of body waves to surface waves; this type of noise originates primarily under the recording station. Microseismic noise, which is considered to consist mainly of fundamental and higher mode Rayleigh waves, has been shown to exhibit a sharp peak in the 540 8-s period range (Brune & Oliver 1959). As a result of this sharp peak in the microseismic noise spectrum, frequency bandpass filtering is often employed to improve the signal to noise ratio. Although this type of filtering is very effective in removing microseismic background from both long and short period data, it often cannot distinguish between signal and signal generated noise. Difficulty may still arise in the attempted identification of phases whose frequency characteristics are similar.