Research And Application On Noise Separation Based On Independent Component Analysis

With the improvement of modern industry and the rapid development of national economy, the use of air conditioner is getting higher and higher.The noise caused by the vibration of air conditioner unit contains important information of the running state of air conditioner.But all kinds of noises are combined together to make it difficult to extract accurate information.Therefore, in order to monitor the condition of air conditioner,it is needed to find a kind of noise separation method to extract the acoustic signal corresponding to each component from the mixed noise signals.ICA is a developed signal processing method recently.This method is not based on any prior knowledge, and uses the statistical characteristics of the signal.This paper first briefly introduces the background and significance of the subject.The theoretical basis of ICA is systematically expounded, including probability and statistics knowledge,information theory, the basic model,isolability conditions etc.Then, two commonly used algorithms for ICA are implemented:Informax and FastICA.In the MATLAB platform,separation experiments with the sine mixed signal,audio mixed signal, noise mixed signal were carried out to use the two algorithms respectively.The effectiveness of these two algorithms in separating the mixed signals is compared and analyzed from different aspects.Finally, the application of FastICA based on negative entropy maximization in the noise of air conditioner is discussed further,and the application model of noise separation of air conditioner is set up.The noise of different components was extracted by experiment and good separation effect was obtained.The feasibility of the FastICA algorithm in the noise separation of the air conditioner is verified.