Study Of Two-Dimensional Directions Of Arrival Estimation Methods

The problem of two-dimensional (2-D) directions of arrival (DOAs) estimation has attracted a lot of attention, especially in fields such as radar, sonar, communications, and seismology. Many 2-D DOAs estimation methods have been proposed recently, the popular high-resolution techniques used to distinguish multiple closely spaced sources are eigenstructure based techniques. However, multidimensional searching based 2-D DOA estimation techniques require searching spectral peaks in a 2-D domain and are thus not amenable to real-time implementation. Another class of methods that adopt two one-dimensional processing requires complex pair matching. The aligning accuracy of pair matching degrades significantly in the case of low SNR, small angular separation, and other severe propagation environments.The main contributions of this dissertation are studying the property of some widely used arrays and proposing several 2-D DOAs estimation algorithms with neither pair matching nor peak searching. The main creative works are concluded as the following:Firstly, a novel decoupled method for 2-D DOAs estimation using two parallel ULAs is proposed. It decouples the 2-D DOAs estimation problem into two one-dimensional estimation problems and thus reduces the computation complexity. It can estimate more 2-D parameters of source signals with distinct angles than DOA matrix method and obtained more accurate estimates. Then it is solved by polynomial-rooting. A relevant Cramer-Rao bound (CRB) of the proposed method is derived and forward/backward spatial smoothing techniques are adopted to extend the proposed method to estimate 2-D DOAs of multiple highly correlated and coherent signals.Secondly, a novel joint diagonalization based DOA matrix method is proposed to estimate the higher 2-D DOAs of uncorrelated narrowband signals. The method constructs three subarrays by exploiting the special structure of the array, thereby obtaining the 2-D DOAs of the array based on joint diagonalization directly with neither peak search nor pairing. The new method can handle sources with common 1-D angles. Then the method is extended to a sensor array consists of triplets of identical sensors. The new array structure achieves very accurate estimates since the estimation accuracy can be improved by adjusting the array aperture.Thirdly, in some applications, the sources are stationary with different spectral contents. We can achieve array aperture extending, array calibration, and so on by exploiting the temporal structure of the signals. We propose a 2-D DOAs estimation method based on a joint diagonalization of several spatio-temporal covariance matrices. It is shown that performing a joint diagonalization of a combined set of these matrices provides an improved estimate of the 2-D DOAs over the aforementioned techniques in two aspects. First, signals with common 1-D angles or in any curved surface can be resolved. Second, robustness is increased at low signal to noise ratios (SNRs). At last, the method is extended to 2-D DOAs estimation with arbitrary array configuration.Fourthly, for non-Gaussian signals, higher-order (HO) statistics have many excellent properties such as suppress non-Gaussian noise, array calibration, array aperture extending and so on. Two novel joint diagonalization fourth-order cumulant DOA matrix methods are proposed to estimate the 2-D DOAs of uncorrelated narrowband signals in arbitrary Gaussian noise environment. Based on the special structure of the array, the methods constructs several subarrays by exploiting fourth-order cumulant, thereby obtaining the 2-D DOAs of the array directly with neither peak searching nor pair matching. The new method can handle sources with common 1-D angles.Finally, there are also many phenomena in signal processing which are decidedly non-Gaussian. Non-Gaussianity often results in significant performance degradation for system optimized under the Gaussian assumption. In order that the devices work well and exert fighting efficiency sufficiently in this bad situations, an exact and appropriate mathematical model needs to be established for impulsive interference or noise, what's more, signal processing methods and techniques are studied in the impulsive noise environments. Three novel joint diagonalization fractional lower-order moment (FLOM) DOA matrix methods are proposed, which can estimate the 2-D DOAs of the signals directly with neither peak searchings nor pair matchings. Moreover, they can handle sources with common 1-D angles. The new methods are robust against SαS noises and remedy the lack of the traditional subspace-based techniques employing second-order or HO moments cannot be applied in the implusive noise environments.