Adaptive Processing of Digital Broadband Seismic Data

The main purpose of this paper is to develop a step-wise approach to recursive parameter estimation for time-invariant autoregressive moving average (ARMA) models used to track slowly time variant-seismic noise. In these steps computational complexity is balanced against estimation accuracy. By updating with every new data sample, the recursions are well adapted to on-line implementation. They are designed to be insensitive to spurious additive glitches in the data. Assuming that the ARMA parameters vary slowly with time, the estimated parameters contain information about the long time behavior of the modeled process compared with the time duration of additive transient signals. In seismological applications these transients are thought to be earthquake signals. The estimated ARMA parameters are used a) for the design of robust prediction error filters with arbitrary prediction distance to reduce the microseismic noise while passing the earthquake signal widely undisturbed, and b) for automatic detection of earthquake signals. A three-step scheme for the detection of weak earthquake signals is developed: The first step is to clean the data from glitches (for example data transmission errors) by replacing these with predicted values. The second step involves conventional recursive bandpass filtering to focus upon relevant frequency bands. In the third step a detection variable is computed from the difference of time consecutive ARMA parameter vectors for the bandpass filtered traces.