Sliding Eigenvalue Decomposition for Non-stationary Signal Analysis

Nowadays, decomposition of multi-component signals has gained popularity in time-frequency analysis (TFA) of non-stationary signals. Eigenvalue decomposition (EVD) is one such technique which decomposes signals into mono-components. In this paper, a new approach named sliding EVD for non-stationary signal decomposition has been proposed. The sliding EVD comprises short duration EVD of signals and an unsupervised grouping of obtained components. This proposed algorithm surpasses other EVD based techniques by successfully decomposing the signals which are overlapped in frequency domain and separated in time-frequency domain. Later, Hilbert spectral analysis has been used on decomposed mono-components for obtaining time-frequency distribution (TFD). At the end, proposed method has been compared with Hilbert Huang transform and is found to be providing better TFD.