Detection of $H$ -Phase Signals From Hydroacoustic Data Using Quadratic Classification

The Comprehensive Nuclear-Test-Ban Treaty Organization (CTBTO) operates a network of underwater hydrophones as part of the International Monitoring System (IMS). Data from this network are processed at the International Data Centre (IDC), Vienna, Austria. One of the objectives is to identify the signals that are due to an underwater explosion, the so-called H -phase signals. Data provided by IDC were examined to investigate new automated processing schemes that could significantly reduce the number of signals needed to be analyzed by human experts, while still detecting with high probability the rare H -phase signals. A variant of quadratic classification (QC) using four signal characteristics from the time, frequency, and cepstrum domains was applied to the problem. It was found that 97.5% of the received H -phase signals are detected by the automated QC process. These H -phase signals are among only about 1% of the signals which are allowed to form event solutions for further analysis by experienced experts.