HR: 13:45h
AN: S22D-01 INVITED [PDF]
TI: Detection of Uncertain Signals
AU: * Harris, D B
EM: harris2@llnl.gov
AF: Lawrence Livermore National Laboratory, 7000 East Avenue, Livermore, CA 94550-9234 United States
AB:
Relative location of events with highly similar waveforms can be made extremely precise through the use of correlation
relative picks. Groups of events susceptible to correlation picking may be identified by cluster analysis using waveform
correlation as a clustering metric. Frequently, waveform correlation clustering is used to sift catalogs or lists of STA/LTA
detections for events that are correlation picking candidates. An alternative approach is to use correlation detectors to
identify groups of related events that are guaranteed to have similar waveforms. Correlation detectors have the additional
advantage of greater sensitivity than simple energy detectors, i.e. of much higher probabilities of detection at a fixed
false alarm rate under threshold detection conditions. They have the potential to detect smaller correlatable events, and to
automate the detection of such events.
The similarity of waveforms from related events declines due to variations in source mechanism, source time history and
source location. The performance of correlation detectors declines significantly as the uncertainty of the waveform to be
detected grows.
It is desirable to develop detectors that retain much of the sensitivity of correlation detectors while reducing the loss of
performance due to signal uncertainty. Subspace detectors offer one approach to manage this tradeoff. These algorithms
detect signals that fall within a subspace of desired signals, represented by a waveform basis. The basis can be chosen to
represent the range of uncertainty in the signals to be detected (or conversely, the range of knowledge available about the
signals). With this approach it is possible to generate a family of detectors that grade in small steps from a correlation
detector, when the signal to be detected is known perfectly, to a simple energy (STA/LTA) detector, when little is known
about the signal.
This presentation discusses empirical methods for designing subspace detectors, focusing on selecting the order of the
subspace representation to maximize the probability of detection at a fixed false alarm rate. The approach is illustrated
for the problem of detecting variable mining explosions.
DE: 7219 Nuclear explosion seismology
DE: 7294 Instruments and techniques
SC: Seismology [S]
MN: 2003 Fall Meeting