In underwater surveillance sources, such as ships or submarines, are localized using the acoustic noise emitted by the source engines, propellers and other machinery. The acoustic signals propagate in the sea and are recorded with an array of acoustic sensors. Processing the recorded signals to obtain the locations of the sources is known as Direction of Arrival (DOA) estimation in the field of signal processing.A simple mathematical model relating the sensor array geometry to the DOA of the source exists when the frequency of the source signal is known. The model is directly applicable to a narrowband DOA estimation problem where the energy of the source signals is concentrated around a single carrier frequency. For underwater surveillance, however, the source signals are wideband which complicates the problem.This thesis reviews existing methods for wideband DOA estimation: Simple extensions of well known narrowband methods MVDR and MUSIC, the so called coherent methods and the most recent methods belonging into the sparse framework. An original idea for extending MVDR using a likelihood based combining of subbands, MVDR-LBC is developed.The thesis models the sensor signals as a sparse autoregressive process by linear prediction and the original algorithm GRLS. The sparse model is shown to be effective compared to the conventional non-sparse one. The model can be used to compress the data recorded in underwater surveillance.The wideband DOA estimation methods are tested with a number of simulations and with real data recorded in the sea. MVDR is shown to be robust and effective, the accuracy and resolution of which can be improved using MVDR-LBC. MUSIC provides good resolution, is computationally efficient and can be implemented quite simply. The coherent methods are the most complicated and need good pre-estimations for the source directions but can resolve close sources best.
展开▼
