The Radiated EMI Noise Analysis And Diagnosis Methods Based On Blind Source Separation

As more and more electronic equipments are widely used in research and manufacture anddaily life, the problem of electromagnetic interference(EMI) become more important. Manycountry has established their EMC standard and make it executed compulsively, meanwhile, thestudy of EMC are going on widely and deeply. Among these works, the use of the technical ofmodern signal processing(MSP) on the analysis and research of EMI noise signal and on thediagnose and locating the noise source is a hotspot.This paper is to discuss blind source separation technique of modern signal processing inradiated EMI noise analysis.First,it introduces the background and the research status of radiatedEMI,also ablout BSS.Then we has a deep research on some common blind source separationalgorithms, including independent component analysis fastICA algorithm, nonparametric densityestimation ICA algorithm and STFT-based convolution blind source separation algorithm,proofing them in the instantaneous mixtures and convolution mixtures case with the simulationdata and EMI radiated noise experimental data. Experimental results show that fastICAalgorithm, nonparametric density estimation ICA algorithm can separate independent componentof instantaneous mixed sources, but for existence of convolution mixed case in experimental data,these two algorithms is poor, can only separate a signal source. Separation algorithm based onSTFT of convolution, for whether the data for the simulation experimental data or EMI radiationnoise,can achieve better separation. Therefore, for blind source separation of radiation EMI noiseanalysis, the convolution algorithm is more efficient. Finally, through the wavelet packet featureextraction techniques and BP neural network classifier for blind signal separation,the separatedsignals are classified,and prove the feasibility and correctness of these methods throughexperiments.