Statistical tests for sonar data conditioning

Source signature identification by means of spectral line characterization is important in passive sonar surveillance. However, narrowband spectral lines may be distorted or masked by the spectral sensitivity for nonstationarities such as transients. Here, we use a Kolmogorov-Smirnov based statistical test for stationarity characterization of sonar data. If the test shows presences of transients a stepwise outlier rejection algorithm in the bispectral domain is used for removal of nonstationarities. The two methods are combined to form a data conditioner for spectral lines in experimental sonar data. We demonstrate the potential of this procedure as a complementary tool to conventional sonar signal processing.