| Summary: | 碩士 === 嶺東科技大學 === 資訊科技應用研究所 === 100 === Source direction of arrival (DOA) estimation using an array of spatially distributed sensors is an essential and difficult task in a variety of applications, such as radar, sonar, communications, and seismic exploration. The problem has been an active research area for decades, such as minimum variance distortionless response (MVDR), multiple signal classification (MUSIC), and others. Most existing techniques are established with an assumption that noise is spatially white and its covariance is a constant times an identity matrix. Alternatively, if the noise is not white but its covariance can be estimated from noise-alone data, the data is then prewhitened and the algorithms are applied to the transformed data. However, in many practical situations, the noise fields may be unknown. The success of subspace-based DOA estimation is based on its ability to perform a complete separation of signal and noise (orthogonal) subspaces. Observation noise degrades the performance of DOA estimation by causing an incomplete separation of the two subspaces.
For the application of MUSIC estimator in dealing with the time division multiple access (TDMA) signals, this thesis first presented an efficient derivative polynomial rooting method in against to the effects of finite samples and relative larger noise. This approach can fast and precisely estimate DOA, by way of improving the computation processing load of the conventional spectrum searching method and the estimating bias of the conventional polynomial rooting method. Since single subspace feature extraction fails to achieve satisfactory results under unknown noise fields. Therefore, no single subspace feature extraction method outperforms others under all circumstances, but applying a dual-space feature extraction method can overcome the limits of single subspace. A robust dual-space feature extraction is put forward based on minor component analysis (MCA) and independent component analysis (ICA) to increase the DOA estimate accuracy. Especially, the proposed feature extraction always starts the procedure as a MCA and ends as an ICA. A feature extracting procedure is presented here, which combines the MCA subspace demixing procedure exploiting individual noise-subspace projection and Newton’s iteration algorithm for complex-valued FastICA is utilized to extract the specific feature of the Gaussian noise component from mixtures so that the estimated component is as independent as possible to the other non-Gaussian signal components. Once the new noise basis vectors are obtained, which is as orthogonal as possible to the estimated noise basis vectors especially for highly correlated and impulse noises, the noise subspace can be reconstructed for the subspace-based estimation algorithm. Finally, comparison with the conventional MCA-based subspace methods is presented via computer simulations to support the effectiveness of the proposed method.
|
|---|
