Research Of Noise Source Identification And Separation Algorithms

Noise has been receiving increasing recognition as one of our critical environmental pollution problems besides air and water pollution. Noise and vibration measurement is of a great importance for the noise detection, environment protection, labour protection etc. The best way to control noise is at an adequate sound source. This approach decreases the complexity of work and help people research new products that generate undesired sound of lower level. The identification of acoustic source accurately is a fundamental problem in noise control. In the practical project, if the contribution of multi-source-noise to the whole was identified, and then the noise level can be reduced accordingly.A new approach to acoustic noise identification was proposed by introducing modern spectrum estimation and Grey Relational Analysis (GRA). Modern spectrum was used to recognize the main noise source and GRA was used to recognize the similarity among different curves of power spectrum. Then Artificial Neural Network (ANN) and Support Vector Regression (SVM) were used to train the GRA data individually. The ranking of the noise sources were obtained by the testing data on the basis of their individual contribution to the overall noise. Finally decision fusion was used to combine the results of the ANN and SVM.Due to the limited source noise we can measure, blind source separation (BSS) was used to separate the individual noise source from the whole. Based on the correlation between the recovered signal and the original individual noise signal, the place of the individual signal can be decided, and then the method mentioned above can be used again.The results of simulation signals confirmed the feasibility and validity of the method proposed in this dissertation and it will play an important role in noise control, signal source identification, blind source separation and other fields.