Date on Master's Thesis/Doctoral Dissertation
Electrical and Computer Engineering
Frequency estimation; Harmonic retrieval; Perturbation
Radio frequency allocation--Mathematics; Wireless communication systems--Mathematical models; Frequency spectra--Mathematical models
Multidimensional frequency estimation is a classic signal processing problem that has versatile applications in sensor array processing and wireless communications. Eigenvalue-based two-dimensional (2-D) and N -dimensional ( N -D) frequency estimation algorithms have been well documented, however, these algorithms suffer from limited identifiability and demanding computations. This dissertation develops a framework on eigenvector-based N -D frequency estimation, which contains several novel algorithms that estimate a structural matrix from eigenvectors and then resolve the N -D frequencies by dividing the elements of the structural matrix. Compared to the existing eigenvalue-based algorithms, these eigenvector-based algorithms can achieve automatic pairing without an extra frequency pairing step, and tins the computational complexity is reduced. The identifiability, performance, and complexity of these algorithms are also systematically studied. Based on this study, the most relaxed identifiability condition for the N -D frequency estimation problem is given and an effective approach using optimized weighting factors to improve the performance of frequency estimation is developed. These results are applied in wireless communication for time-varying multipath channel estimation and prediction, as well as for joint 2-D Direction-of-arrival (DOA) tracking of multiple moving targets.
Liu, Jun, "Eigenvector-based multidimensional frequency estimation : identifiability, performance, and applications." (2007). Electronic Theses and Dissertations. Paper 842.