S. M. Jesus
UCEH---University of Algarve, PT-8000 Faro, Portugal
Normal-mode modeling is a well-accepted representation for acoustic signals propagating in a number of environmental conditions. Detection of the spatial normal-mode structure makes possible signal localization enhancement against the noise that has no spatial structure. A simple algebraic argument allows one to separate the vector subspace that contains the signal from the vector subspace that contains the noise and to obtain a narrow-band estimate of the source location [S. M. Jesus, Signal Process. 28, 117--122 (1992)]. This subspace splitting algorithm has been extended for localizing broadband transient signals assuming that the signal only has a normal-mode structure. It is shown with synthetic data that the proposed broadband algorithm outperforms both the generalized minimum variance and the conventional processors. As an example, this processor has been used to localize short transient pulses collected in a 120-m depth shallow water area with a 62-m aperture vertical array. The experimental results show that stable and accurate localizations could be obtained during long time intervals. This shows that the sound field, received over a given frequency band, is relatively stable over time and is in agreement with the predictions given by a standard normal-mode propagation model.