A Novel Detector of Broadband Transient Signals

Abstract : A novel approach for detection of broad-band (BB) transients (DBBT) is proposed. DBBT is based on an analysis of the spectrograms, calculated from time series of one or more hydrophones. A spectrogram is divided into locally normalized time-frequency cells (TFC). Two TFCs, which are measured at the same time but at different neighboring frequencies, are said to include a signature of a broad-band transient signal if their empirical centres of masses estimated in a higher-dimensional feature space using a Gaussian kernel exhibit correlation higher than a predefined threshold. A spread coefficient p of a Gaussian kernel is chosen so that the correlation between the pairs of centres of masses of TFCs that do not include transient signatures appear to be uncorrelated. Tests of BB transient detection, carried out using a number of real-world data sets, demonstrate robustness of the proposed method. In particular, DBBT was able to detect the launch and explosion of a torpedo, the start and stop of small boat engine, and other events. Reusable values of size of TF cells, a p value, and correlation threshold were used. The same settings were also successfully used with two sets of torpedo and ship measurements artificially combined at different bearings within a framework of simulated beam-forming. The results were found to be relatively insensitive to the array shape and target bearing errors. A few tests that were carried out on artificially generated narrowband signals using modified DBBT were found to be too limited and therefore inconclusive.