Detection of Ripple Firing in Time and Frequency.
Abstract
Frequency and time domain approaches to detecting ripple firing on an array are developed and tested on simulated data and on 10 mining explosions in a quarry on Kola peninsula. The approaches are useful for isolating the periodic modulations in the spectra that may distinguish the signatures of mining explosions from those of earthquakes. All three methodologies exploit the assumption that approximately the same pattern should appear on all sensors after path and site effects have been eliminated. In the first frequency domain method an additive noise model is used to deconvolve the common signal from path and site effects. A cepstral analysis exhibits the delay structure. In the second frequency domain method, a multiplicative convolution model is used to derive a Cepstral F-Statistic, based on the stacked cepstral and the spectrum of the stacked log spectra. The time domain approach involves searching the multiplicative convolution model using seasonal ARMA models for the delay structure and the source. It is shown that all three approaches work reasonably well on contrived data but that frequency domain versions tend to be superior. The methods are then applied to four phases from 10 mining explosions on Kola, observed on 5 elements of the Norwegian array ARCESS. We conclude that some phases show evidence of rippling in the 100-200 millisecond range for all ten explosions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 30, 1997
- Accession Number
- ADA370858
Entities
People
- Robert H. Shumway
Organizations
- University of California