Subspace Detectors: Theory

Broadband subspace detectors are introduced for seismological applications that require the detection of repetitive sources that produce similar, yet significantly variable seismic signals. Like correlation detectors, of which they are a generalization, subspace detectors often permit remarkably sensitive detection of small events. The subspace detector derives its name from the fact that it projects a sliding window of data drawn from a continuous stream onto a vector signal subspace spanning the collection of signals expected to be generated by a particular source. Empirical procedures are presented for designing subspaces from clusters of events characterizing a source. Furthermore, a solution is presented for the problem of selecting the dimension of the subspace to maximize the probability of detecting repetitive events at a fixed false alarm rate. An example illustrates subspace design and detection using events in the 2002 San Ramon, California earthquake swarm