Linear and Instantaneous Mixtures Following Eigenvalue Decomposition

Abstract

The physical signiflcance of independent component analysis is not clear, and it does not have only one solution. Aiming at the disadvantages, a novel linear and instantaneous mixture algorithm method based on eigenvalue decomposition of observation signals autocorrelation matrix is proposed. Assumptions of this new method are difierent from the traditional blind source separation model and independent component analysis algorithm. For difierent types of mixed matrix, this new method gets difierent information of source signals. The numerical simulation results show that, it can separate multiple unrelated source signals in despite of Gauss measurement noise and can also achieve noise reduction efiect after separation and reconstruction of observation signals.