Reaserch On Blind Separation And Its Application In Array Signal Processing

Blind source separation(BSS)is hot topi c in the area of signal processing.Based on the characteristic of signa ls dependence or sparse,it can separate signals from mixtures only using observation data.In recen t years,BSS are app lied in array sig nal processing by some researchers.It not only can separate signals from mixtures,but also can obtain the DOA o f separated signals.BS S is significant for array signal blind processing.However,som e important issues are still unsolved,such as array signal blind separation when SNR is low,wideband array signal blind separation,array signal blind separation when the m odel is dism atch,and so on.In this dissertation,we investigate the problem of blind source separa tion and its a pplication of array signals,involving narrowband signa l separation and DOA estima tion,wideband dependent signal blind separation and DOA estim ation,wideband deterministic signal separation and DOA estimation,array signal blind separation and DOA estimation when the array model is dism atch.The prim ary contributions and o riginal ideas included in this dissertation are summarized blow:In chapter 2,a novel method for narrowb and array signals blind separation and DOA estimation is proposed,which adapt to low SNR environment.Firstly,assume the sources are independent,the eigenvalue and ei genvector of array covariance matrix can be obtained by eigenvalue decom pose.Secondly,the num ber of sources K can be estimated according to eigenvalu e.Then,we can extract partial observation data what contain small noise component using K eigenvector.Based on the extracted data,sources can be effectively separated by J ADE algorithm.Finally,the relative DOAs o f signals are estimated by combining separated matrix and array structure.The proposed method also can effectively reduce noise com ponents over noisy environment.It could be directly applied to radar,sonar and radio surveillance for separating signals and estimationg relative DOAs of signals.In m ulti-path environment,the array m anifold matrix has changed so that the above DOA estimation method cannot be available.To overcome the problem,we newly introduce an extended DOA estimation method which combines independent component analysis(ICA)and sparse reconstruction algorithm.First,the a rray manifold is es timated by ICA algorithm.Then,we see the steering vector in array manifold as a single time sample of array output,and estimate the DOA of multi-path signals by sparse reconstruction.In chapter 3,a novel method for wideband independent signals blind separation and DOA estimation is proposed.Firstly,the wi deband mixed signals are transfer into instantaneous mixture in different freque ncy bins by Short Tim e Fourier T ransform(STFT).Secondly,all of frequency bins are cu t into several sections according to the situation of signals mixed in frequency domain.For mixed section in frequency domain,the mixture in each frequency bin is separated by the Joint Approxim ate Diagonalization of Eigenm atrices(JADE)t echnology.In this section,for different frequency bins,the separated signal com ponents can be m erged by relative DOA of signal.Different sections can be m erged by relative coefficient.Estimate DOAs of t he sources by utilizing the directiv ity pattern which is obtain ed according to unm ixed matrix and array m anifold matrix.The dire ctions of nulls are DOAs of the sources.When the number of the sources is 2,we can easily estimate DOA by searching for the minimum value of the direction parttern fu ction.When the number of the sources is more than 2,the num ber of null should be more than one.So we modify the DOA estimation method.The k means clustering method is used to estimate DOAs of relative separated components.The proposed m ethod not only can separate wideband mixed signals effectively,bu t also recovery the signals' amplitude correctly.Besides,the DOAs of relative separated com ponent can be obtained.Another advantage is that the number of antenna needed could be equal to the number of the sources.Compare to wideband beamforming method,the number of antenna needed of the proposed method is less.In chapter 3,the problem of DOA estim ation and blind separation for wideband and determistic signals is inves tigated.Based on the expectation-m aximization(EM)algorithm,a novel blind source separation and DOA estimation m ethod is presented.Assume the noise is gauss random noise.Firstly,we establish a likelihood function in the frequency domain by jointly using all fr equencies information.Then,based on this function,the E-step and M-st ep in the fram e of broadband EM algorithm are derived.Finally,we jointly estim ate DOAs and sources through EM iteration com pute.Moreover,through analyses the convergence of EM algorithm,we set adapted angle search space for DOA estim ation,which reduc e the com putations of our algorithm.Compare to the traditional m ethod,more effective information are applied in our algorithm.Therefore,the proposed algorithm has more advantages for DOA estim ation and source separation.In chapter 3,a novel DOA estim ation and blind source separation algorithm in the presence of m utual coupling and sens or position error is presented for unknown deterministic signals.In order to highlight the relationship between the array output and array error coefficients,we introduce a nove l model of the arra y output with the unknown array error coefficients.B ased on this model,we use the Space alternating generalized expectation-Maximization(SAGE)algorithm to jointly estimate the sources,DOA and the m utual coupling coefficients.First,the narrowband case is considered.Based on the small perturbation assumption,we propose a new augmentation scheme so that estimate DOA and perturbation parameters using SAGE algorithm.The E-step and M-step of SAGE algorithm in this case are derived.Then,the algorithm is extended to the wideband case.The wideband SAGE algorit hm is derived in frequency dom ain by joint all frequency bins.Co mpare with m any existing m ethods,our method requires neither calibration sources nor initial calibration information.And our proposed method inherits the characteristics of good converg ence and high estim ation precision of the SAGE algorithm.Moreover,we can jointly estim ate the sources,DOA and sens or location errors for wideband signals,what is not reported in public literature.