Abstract
A new supervised approach for decomposition of single channel signal mixtures is introduced in this paper. The performance of the traditional singular spectrum analysis (SSA) algorithm is significantly improved by applying tensor decomposition instead of traditional singular value decomposition (SVD). As another contribution to this subspace analysis method, the inherent frequency diversity of the data has been effectively exploited to highlight the subspace of interest. As an important application, sleep EEG has been analysed and the stages of sleep for the subjects in normal condition, with sleep restriction, and with sleep extension have been accurately estimated and compared with the results of sleep scoring by clinical experts.