Passive localization of sound sources in shalow water simulation studies.

Abstract

Array processing techniques for source localization in shallow waters can be classified as:

i) Time delay estimation (TDE) techniques, and ii) High resolution (HR) methods.

Time delay estimation techniques are two-step procedures. The time difference of arrivals (TDOAs) is first estimated, and then, using these estimates, the direction of arrivals (DOAs) is calculated. One interesting method belonging to this class, proposed recently by Huan, computes the TDOAs from the phase of the cross-spectrum and is found to be computationally efficient. It is shown in this thesis that Huan's algorithm fails to localize a sound source when the source and the receiver are close to the channel boundaries, and this is an inherent limitation of the algorithm. The reasons for this failure are discussed.

In contrast, the High Resolution (HR) techniques are one-step procedures which estimate DOAs directly. In this thesis, one such HR technique based on eigenstructure decomposition, called Multi-Image Subspace Algorithm (MISA), is studied in detail. MISA exploits information about shallow channel geometry and has already been shown to possess all the advantages of eigenstructure methods under asymptotic conditions. The present study focuses on the finite data performance of MISA. For a given array signal-to-noise ratio (ASNR), the effect of the number of snapshots on the received signal and the source/array position on detection probability has been investigated.

The behavior of MISA estimates as a function of data length has been compared with the performance measure. It is shown that for a given ASNR, beyond a certain minimum number of snapshots, the performance of MISA is independent of source/array location. For a smaller number of snapshots, the estimates follow the performance measure.

Finally, the sensitivity of MISA to channel undulations is also studied. The undulations, modeled as sinusoidal fluctuations, severely affect the MISA estimates, particularly as the range increases. The detection probability decreases drastically for increases on the order of 1/10th of a wavelength. This emphasizes the need for accurate knowledge of the channel parameters and geometry in order to effectively employ MISA.