Subspace methods for direction-finding and wavenumber estimation.

摘要

Many subspace-based parameter estimation methods have been proposed for applications in location of wave sources, harmonic analysis, commercial and military communication, electromagnetic and sonar imaging, seismic wave analysis, and electromagnetic telescopes. In this thesis new subspace methods for high-resolution parameter estimation are suggested. They depend not on optimization techniques, but on simple matrix computations.; The subspace-based companion matrix (SUBCOM) approach to estimation of 1-D direction-of-arrivals (DOAs) and sinusoidal frequencies, which has been suggested by this thesis research, opened the door to a new way of deriving computationally efficient subspace methods. The improved SUBCOM method is suggested in this thesis. This improved SUBCOM method is computationally more efficient than the total least squares (TLS) ESPRIT method. Results of simulation shows that the improved SUBCOM method outperforms the TLS ESPRIT.; Three new subspace methods for estimation of 2-D DOAs and wavenumbers are suggested. They are called 1-D based 2-D parameter estimation via signal-selectivity of signal-subspace (PESS), Pairing-PESS, and Direct 2-D PESS. They estimate the steering matrix of special form by computing a matrix of eigenvectors. Results of simulation show that the 2-D PESS methods provide as good estimates as the matrix enhancement and matrix pencil (MEMP) method which is a 1-D based 2-D method using the additional pairing algorithm. Results of simulation on a test case show that the Direct 2-D PESS method performs better than MEMP, and the Direct 2-D PESS method significantly outperforms MUSIC.; Estimation of DOAs of coherent signals by an RBF neural network is considered. Results of simulation are very good. As a possible application of 2-D wavenumber estimation, the enhancement problem of incomplete images, which may occur when digital video signals are transmitted in compressed form, is considered. Results of simulation show that the visual quality of the incomplete image is greatly enhanced.