LAUSR

Empirical wavelet transform (EWT) has been successfully utilized for decomposing multi-component signals into intrinsic mode functions. However, it suffers from the spectrum subdividing problem when signals contain non-stationary components which overlap in both the time and frequency domains. In this paper, a morphological filtering enhanced empirical wavelet transform (EEWT) methodology is presented for mode decomposition of non-stationary signals. Instead of dividing spectrum in terms of the local maxima–minima segmentation principle, the proposed scheme will smooth the spectrum spikes of a signal with morphological filtering so as to keep different intrinsic mode functions in the corresponding spectrum segments. The proposed method is compared to the classical EWT and the EEWT. The experimental results demonstrate that the proposed method is able to achieve better performance of spectrum segmentation and higher resistance to noise than the EWT and EEWT techniques for both synthetic and speech signals.