Narrowband Array Signal Processing Using Time-Frequency Distributions
L.A. Cirillo · Technischen Universität Darmstadt · 2008
In many engineering applications where sensor arrays are employed, such as radar, sonar, telecommunications, speech processing and medical imaging, the signals observed are often nonstationary. This dissertation addresses particular problems in narrowband array signal processing for nonstationary signals. By making use of spatial time-frequency distributions, one is able to effectively exploit the nonstationary nature of the source signals, with one caveat: the time-frequency localization of the sources should be known a priori, or must needs be estimated. The task of determining the time-frequency localization properties of signals from noise-contaminated sensor array measurements, is composed in this work as problems of `point selection' and `signature estimation'. A `point selection' scheme for automatically determining the time-frequency locations at which spatial time-frequency distribution matrices exhibit underlying diagonal or off-diagonal structure is proposed, based on multiple hypothesis testing. The tendered method is used to achieve blind source separation of nonstationary signals via joint diagonalization and joint off-diagonalization of a set of spatial time-frequency distribution matrices. Toward the goal of `signature estimation', a computationally attractive implementation of a time-frequency Hough transform is proposed. Statistical analysis of the method is conducted to determine the achievable estimation accuracy. The proposed approach is applied to direction-of-arrival estimation, based on the averaging of spatial time-frequency distribution matrices. The problem of micro-Doppler signature estimation is also examined. Micro-Doppler signatures arise, for example in radar, due to the vibrational or rotational motion of targets. The aforementioned signature estimation approach is shown to yield biased estimates of the micro-Doppler amplitude, and a bias correction procedure is given. The methods developed here are applied to data from a radar experiment for validation of the theoretical ideas. Near-field parameter estimation is also considered. When sources are in the near-field of an array, it is possible to perform passive localization in both range and direction. The use of spatial time-frequency distributions for near-field localization is investigated. A means of distinguishing between the time-frequency representations of far- and near-field sources is also proposed. Data from an experimental radar system is analyzed using the proposed techniques.