Research On Blind Source Separation

Blind Source Separation(BSS)does not need to know the relevant information of the source signal and the transmission channel.It can only perform signal estimation and recovery from the receiving end.Due to its broad application prospects in fault diagnosis,voice enhancement,and remote sensing,the research has been rapidly growing since 1980,and has become a hot topic in signal processing disciplines.However,it also exposed some problems in practical applications,such as limited separation performance.This article first introduced the background,development and research status of BSS,followed by the basic theory of BSS basic model,pretreatment method,separation criteria and evaluation indicators,and then used adaptive variable adaptive blind source separation(EASI)and Single Channel Blind Source Separation(SC-BSS)is an object,and studies are conducted in both positive and underdetermined directions.The specific research includes:(1)For crosstalk error-based variable step size EASI(VEASI)algorithm,the crosstalk error can't accurately judge the separation effect when the energy of each component of the source signal differs greatly or the global matrix is a dominant matrix of some non-line elements.The variable-step-size EASI(IVEASI)algorithm for crosstalk error and crosstalk error can improve the separation performance by integrating the crosstalk error and the separation degree adjustment step.(2)Aiming at the convergence speed caused by the random selection of the initial matrix in the IEASIA algorithm,an IVEASI algorithm based on the RW-MCSO algorithm is proposed.The mutation factor and stochastic inertia weight were introduced in the chicken group algorithm to update the position of the chicken and rooster.The RW-MCSO algorithm was proposed.The RW-MCSO algorithm was used to optimize the initial separation matrix of the IVEASI algorithm and the convergence speed of the IEASIA algorithm was improved.(3)For the problem of end-point and modal aliasing in EMD-based SC-BSS algorithm,combined with variational modal decomposition(VMD),an SC-BSS algorithm based on VMD is proposed.Firstly,the single-channel observation signal is decomposed into a series of eigenmode function(IMF)components by using VMD,and the single-channel signal and its IMF component constitute a multi-dimensional signal.The principal component analysis method is used to estimate the number of sources,and the multi-channel is reconstructed according to the number of sources.The signals are mixed and then separated using the BSS method.Compared to the EMD method,the VMD method has improved both in the correlation coefficient and the root mean squared error.