Methods for analysis and classification of non-stationary signals based on eigenvalue decomposition of Hankel matrix

Abstract

The natural signals observed in real life are non-stationary and have a time-varying frequency spectrum. Time-frequency analysis (TFA) methods are used to analyze such signals as they provide joint time and frequency information. The signal decomposition methods have gained popularity in TFA after the development of the Hilbert-Huang transform method. Eigenvalue decomposition of the Hankel matrix (EVDHM) is one of these methods from the literature, which has been successfully applied to various applications related to the analysis and classification of non-stationary signals. However, these methods have certain limitations. For instance, they cannot separate the signal components overlapping in the frequency domain. Additionally, the decomposed components are obtained from the components corresponding to eigenvalue pairs using frequency-domain information. Furthermore, they do not preserve the mutual information among decomposed components of the multichannel signals. Lastly, these methods are iterative in nature and become computationally expensive. This thesis addresses the limitations of EVDHM-based methods for signal analysis and classification tasks.