Decomposition of multichannel multicomponent nonstationary signals by combining the eigenvectors of autocorrelation matrix using genetic algorithm

摘要

Multichannel multicomponent signals can be decomposed into individual signal components by exploiting the eigendecomposition of the corresponding autocorrelation matrix. Recently, we have shown that such decomposition is possible even in the particularly challenging case of non-stationary components with significantly overlapped supports in their time, frequency, and joint time-frequency domains. Each signal component can be recovered as a linear combination of the eigenvectors of the autocorrelation matrix, by minimizing its time-frequency concentration measure. However, as the local minima of the concentration measure do exist for each signal component and for each combination of signal components, such minimizations can be challenging and numerically demanding, particularly when considering the associated decomposition procedure which should, for each component, iteratively remove the influence of other components. To confront these challenges, we present a multichannel multicomponent nonstationary signal decomposition procedure which exploits a carefully tuned genetic algorithm for the minimization of the concentration measure of eigenvectors, each comprising the linear combination of the overlapped signal components. Concentration measures are calculated in the time-frequency domain. The presented theory is verified by numerical examples. (C) 2020 Elsevier Inc. All rights reserved.