Can passive sonar signals be classified on the basis of their higher order statistics?

The objective of the work described was to ascertain whether passive sonar signals can be classified on the basis of the higher order statistics of their time series. Of particular interest was whether classification information was present in the higher order statistics which was not present in the power spectrum. In order to achieve the objectives of the work, it was necessary to collate an extensive databank of sonar signals. The resulting databank contains over 300 examples. The examples have been selected so that a range of signal-to-noise ratios, target speeds, environmental conditions, etc. are represented. The paper begins by describing the limitations of conventional higher order spectra (HOS) applied to real, nonstationary signals. To circumvent these limitations new higher order spectra, termed “phase only spectra” (or POS) are defined. Typical results comparing the HOS and POS are presented. The effect of underwater propagation on the phase coherence of signals from moving targets is then investigated. Phase coherence is essential for the existence of features in the HOS or POS. It is shown that phase coherence, even at low frequencies, can only be maintained for a few seconds. This in turn limits the integration period for the HOS, and hence the magnitude of higher order features which can be measured