Wavelet-Based Intelligence that can Tell a Whale Song from an Earthquake Phase
Frederik Simons
Geosciences, Princeton University
We are developing floating pressure gauges in order to record waves generated by earthquakes at large distance. The instrument needs to be completely autonomous: neutrally buoyant at great depth, it communicates with a satellite when periodically resurfacing. We evaluate a variety of algorithms to discriminate common ocean noises (whale songs, ship noise, etc.) from the target signals, most using the discrete wavelet transform. I will briefly discuss the concept of wavelets in signal and image processing, and three types of wavelet transform algorithms, all derived from an iterated filter-bank philosophy. The most promising (fastest, lowest memory requirement, in-place computation, trivially invertible) is the so-called lifting scheme due to Sweldens and Daubechies. Incidentally, the same algorithm forms part of the JPEG2000 standard on the World Wide Web. Our wavelet-based approach lends itself to analysis and discrimination, as well as to denoising and compression for satellite transmission.