| In this thesis, eigenanalysis based techniques are developed for locating narrowband signal sources in 3-D by processing data received by sensor arrays. A general two-step framework, based on the concept of subspace rotation, is presented that decouples time delay estimation and source localization. The first step performs data cleaning and generalized eigenvalue decomposition on the sensor data to estimate time delays. The second step applies the array configuration (planar or volume) and the assumed relative distance of the sources from the sensors (far-field or near-field) to compute source location from the time delays.; The second step, which is applicable to any set of time delays, is fixed for a given array pattern and the geometrical relationships between the sources and sensors. Many possibilities, however, exist for computing time delays. Two techniques are developed that assume that the master array is composed of two translationally equivalent subarrays. Rigorous derivations are provided for the far-field case, and evidence is presented for their applicability to the near-field case. When combined with the second step, the two time-delay estimation techniques yield two new methods--the Modified Direction-of-Arrival (DOA) Matrix (MDOAM) method and the Signal Space Matrix (SSM) method.; Both methods model sensor data in terms of multidimensional coordinates (2-D DOA angles for far-field and 3-D Cartesian coordinates for near-field sources). The MDOAM and SSM methods directly compute multiple source locations, so they do not require peak searching procedures nor a grouping of independently estimated coordinates as the DOAM and SR (far-field) methods do. By using eigenvectors of the newly derived covariance matrices instead of their eigenvalues, the usual constraints on subarray spacing is lifted.; The SSM and MDOAM methods can locate up to M - 2 sources with an M-sensor array. The sensor array must contain two translationally equivalent subarrays whose corresponding sensors have pairwise identical directional responses. Simulations have shown that the MDOAM and SSM methods perform well when the sensor data exhibits a low signal-to-noise ratio (SNR), even under relatively short observation times (e.g., 10 snapshots with SNR = {dollar}-{dollar}5 dB). |
