Discrimination of Seismic Signals Using Autoregressive Signal-Processing Methods (Diskriminering av Seismiska Signaler med Autoregressiva Signalbehandlingsmetoder)

摘要

The work investigates the feasibility of seismological discrimination using pattern recognition based on spectral data. To be more specific an autoregressive (AR) batch-method is used. Various analysis-window-lengths, windowing functions and filter orders are examined. Since the signals cannot be assumed to be stationary, the time evolutions of the AR-coefficients are explored by moving the analysis window successively through each signal. The time evolutions show to be characteristic for specific event types. The studied events have been recorded at the Hagfors Seismological Observatory and consist of 21 presumed explosions and 23 presumed earthquakes. The explosions are divided into two groups depending on where they have taken place. Having specified an analysis window etc., each event is represented by a point in the space spanned by the AR-coefficients. The best separation between the earthquake population and the different explosion populations was achieved for different parts of the signals for the two cases.