Parameters Estimation Of Signals Based On Array Signal Processing

The research of localization of spatial signal sources has mainly focused on estimation of DOA of far-field sources in the early times, however, when sources are located close to the array and in the Fresnel region of the array, the curvature of the wavefront cannot be neglected, i.e., the plane-wave approximation to the spherical wave-fronts is no more valid, in which near-field sources are introduced. Due to the non-linear character of the wavefront, so the wavefront must be characterized by both the azimuth DOA and range. When the signal frequency is unknown, localization of near-field sources becomes the problem of joint estimation of frequency, range and DOA of the near-field source. Near-field source localization has been a key problem in applications such as speaker localization using microphone arrays, radar, sonar, electronic surveillance, and seismic exploration.The dissertation focuses on parameters estimation of near-field sources, study the parameters estimation problem of far-field, near-field and their mixture scenario.The main works of the dissertation are as follows:Chapter 1 analysis the background and meaning of this research, introduce the present status of the research on near-field and far-field sources parameters estimation, then the framework of the dissertation are introduced.Chapter 2 introduce the data model of the far-field sources, then analysis the character of the parameter related about range of the sources, then give two classical subspace like algorithms of the DOA estimation of narrowband far-field sources: ESPRIT(Estimating Signal Parameter via Rotational Invariance Techniques) and MUSIC(Multiple Signal Classification) algorithm. Through the comparison of these two methods, we can see MUSIC algorithm gives slightly better performance, but ESPRIT algorithm has much lower computational complexity, so it is more applicable to the real system.Chapter 3 analysis the problem of near-field sources localization, and build the data model of near-field sources, mainly study the widely used ESPRIT-like method based on fourth-cumulant, this kind of algorithm need no parameter match or spectral peak searching procedure. Moreover, it can efficiently jointly estimate range DOA and frequency of near-field narrowband sources, however, the algorithm suffers from heavy loss of the aperture.Chapter 4 Using JADE(joint approximate diagonalization) in blind source separation for reference, propose an algorithm for parameters estimation of near-field sources based on joint diagonalization theory. Separation of sources signal and estimation of steering matrix can be achieved through the joint diagonalization of a serial of constructed cumulant matrix. the algorithm alleviates the aperture loss gives better performance.Chapter 5 in the practical applications, the observations collected by an array may either mixed near-field and far-field signals or multiple near-field signals or multiple far-field signals. To solve the high computational complexity and aperture loss problem of most algorithms for localizing mixed near-field and far-field sources, the proposed algorithm make use of the ESPRIT-like theory based on a non-symmetric cross array, which can jointly estimate frequency, range and two-dimensional DOA by constructing cumulant matrix with special sensor outputs. The algorithm alleviates the aperture loss, and gives better performance.At last, the summary of the dissertation is made and the discussion of future research areas to be further studied is pointed.