The Implementation Of EMD Algorithm Based On DSP

Signal analysis is the study of the fundamental nature; it is an integral part of our theoretical research and practical engineering applications. EMD method is a new approach that analyses non-linear, non-stationary signals in recent years. It can effectively solve the problem that Fourier analysis and short-time Fourier transform, Wigner-Ville distribution and wavelet transform based on Fourier Transform can not analyze non-linear, non-stationary signal. Using this method can analyze complex signal, finally analyzed into a limited number of IMF. This method can effectively extract the trend of a data sequence or remove the mean of the data sequence.At present, the use of EMD is limited to software. It has many advantages but dependent on computer's performance greatly. The experimental results showed that computer performed a few seconds or even longer to a period of data. Therefore, the use of software algorithms on practical application does not have a good real-time. It is urgent to find an alternative to solve the computing long time problem.EMD algorithm implement technology based on DSP is studied in this paper. Power, processor, memory and AD are designed. The overall algorithm flow is also designed based on the realization process of EMD. FFT and initialization procedures are compiled on CCS. Extreme points and the criteria of IMF are studied based on EMD algorithm. An improved method for EMD is showed in this paper that introduced the FFT transform to filter the IMF component, to determine the number of low-frequency component. At the last, an experiment of known signal and non-stationary signal is made. The experimental results show that method determined the number of IMF with FFT is feasible.