Nonlinear signal processing techniques for signal detection

Abstract

The performance of underwater acoustic signal detection schemes can be reduced by extremes of variability in the input signal. Linear detectors such as spectral amplitude thresholding or matched filtering often show degrading detection performance as the power in the signal of interest reduces compared to the noise. Recently, literature examining weak signal detection has focused on exploiting nonlinear system models such as the Duffing equation, to improve detection. The applications in the literature include non-destructive testing, tool wear indication and seismic activity detection. However, not much attention has been paid to their application to underwater signal detection, and importantly very little is published on robust and comprehensive detection performance assessment. Two nonlinear mechanisms found in the Langevin system and the Duffing system, Stochastic Resonance and a transition from chaotic to stable motion, are examined in this thesis, as signal conditioning tools. Using Receiver Operating Characteristics analysis the detection performance is, for the first time, comprehensively measured for different input noise distributions and for different nonlinear pre-processing system configurations. A novel replica correlation detector is devised, that exploits a property of the stable motion in the normalised Duffing system. The often claimed noise immunity of nonlinear systems is refuted; the findings in this thesis strongly show that performance degrades with increasing noise in a similar to linear detectors. However it is also shown that under certain configuration conditions the Duffing transition mechanism generates better detection performance than a benchmark linear detector, when the signal of interest is mixed with highly impulsive real biological snapping shrimp noise.

More information

Identifiers

ORCID for Paul White: orcid.org/0000-0002-4787-8713

Catalogue record

Export record