Signal Detection in Underwater Sound by Dual-Tree Discrete Wavelet Transform

A use of a dual-tree discrete wavelet transform (DT DWT) to the problem of signal detection in underwater sound is proposed in this paper. By noted that there is no obvious distribution assumption can model underwater noise, an empirical model of noise constructed by the discrete wavelet transform (DWT) is applied to identify the signals. This DWT based model adapts with time and is treated as a baseline at any time to detect the signals. Though, the DWT has a high reputation as a tool for signal compression, due to lack of shift invariance, it has major variations in the distribution of energy between DWT coefficients in different subbands. For detection applications, the ill effects of translation need to be neutralized because the test statistic depends on the energy of signals. This paper proposes the use of a dual-tree discrete wavelet transform (DT DWT) to the problem of signal detection in underwater sound. This shift-invariant DT DWT can generate multi-resolution subspaces that keep more of their coefficient energy in each of these subspaces than discrete wavelet transform´s (DWT). The detection performance comparison, using the receiver operating characteristics (ROC) curves, of this proposed DT DWT-based detector and DWT-based method is presented.