Resolving the mode-mixing problem of EMD in the presence of noise

Empirical mode decomposition has become an established tool for time-frequency analysis and has been widely used. However, a major problem is that its performance of EMD may be affected by intermittence or noise, known as the mode-mixing problem. In order to overcome the mode-mixing problem in the empirical mode decomposition (EMD) algorithm, an adaptive pre-processing technique is proposed. In this work, B-spline least squares approximation is first studied and employed before the use of EMD to eliminate the noise which may result in mode mixing. After that, a knot placement iteration algorithm using the extrema time location is put forward to enhance the adaptive property of the proposed method. Simulations of linear and non-linear signals show that it is capable of significantly reducing mode-mixing problem caused by noise. Comparisons between the proposed method and EEMD method are carried out, indicating that the proposed method is superior to existing methods in accuracy.