Study On DOA Estimation Of Microphone Array System Under The Impulsive Noise Environment

With the rapid development of computer technology and Internet of Things,the acoustic information collection tends to fusion mode of array sensors,which effectively improves the perception ability to the environment.And,direction of arrival(DOA)estimation is an important research of microphone array signal processing technology,which is widely used to the acoustic localization systems such as battlefield monitoring and video conference.The traditional DOA estimation technology is generally based on the assumption that noise follows Gaussian distribution,which can not effectively describe heavy-tailed impulsive noise.Influenced by ground clutter and human factors,there is a large number of impulsive responses in the acoustic environment,which leads to the performance degradation or even failure of DOA estimation based on the second-order statistics.Besides,with the rapid development of microphone array technology,most of acoustic signals received by microphone sensors are wideband signals,and the feature of acoustic signals is more complicated.Considering DOA estimation based on narrowband model can not satisfy the practical requirements,it is necessary to conduct the research of the wideband DOA estimation.In order to improve the accuracy of DOA estimation of microphone array system,this paper makes a systematic and in-depth research under the condition that the performance of DOA estimation is degraded with acoustic impulsive noise,small snapshots and low signal-to-noise ratio(SNR),and the wideband DOA estimation depends on the preliminary angles.The main contents and innovations of this paper are as follows:1)Two DOA estimation algorithms based on generalized maximum complex correntropy:The generalized maximum complex correntropy criterion is defined,which can deal with complex-valued signal and restrain impulsive noise.In order to solve the performance degradation of DOA estimation based on the second-order statistics under the impulsive noise environment,the median-difference correntropy(MDCO)based on the generalized maximum complex correntropy criterion and the complex-valued quasi-Newton based on the generalized maximum complex correntropy criterion(QN-GMCCC)are proposed respectively.(1)MDCO algorithm constructs the cost function of median-difference correntropy as the weight factor of the signal covariance in time domain.Besides,according to Kullback-Leibler(KL)divergence principle,an adaptive kernel width selection strategy is designed to adaptively deal with impulsive noise.The experimental results show that MDCO algorithm can effectively suppress impulsive noise and improve accuracy of DOA estimation.Compared with QN-GMCCC algorithm,MDCO algorithm has less computational complexity and can be effectively applied to embedded system.(2)QN-GMCCC algorithm uses the subspace decomposition to solve the GMCC-loss minimization of residual fitting error matrix of signal,and adopts the alternative iterative method to transform the generalized maximum complex correntropy problem into the convex optimization.Finally,the complex-valued quasi-Newton method is used to solve the GMCC-loss function,which can guarantee the positive definiteness of second-order partial derivative.The experimental results show that QN-GMCCC algorithm can effectively suppress impulsive noise and realize the second-order superlinear global convergence.Besides,QN-GMCCC algorithm has a high-precision of DOA estimation in the case of small snapshots.2)DOA estimation algorithm based on iterative reweighted variational Bayesian learning(OG-WVBL):In order to solve the problems of small snapshots,low SNR and difficult identification of outliers under the impulsive noise environment,OG-WVBL algorithm transforms DOA into the joint variational Bayesian learning of sparse outliers and DOA estimation.Using the sparse prior distribution of signal and impulsive noise,OG-WVBL algorithm solves the sparse vector by maximizing the lower bound of KL divergence to avoid the calculation of the marginal likelihood function.Therefore,OG-WVBL algorithm can effectively realize DOA estimation and identify the impulsive noise localization.OG-WVBL algorithm also introduces the iterative reweighted strategy to hyperparameters so that the more importance is given to those hyperparameters with non-zero entries over others which can encourage sparsity and achieve the consistent convergence.The experimental results show that OG-WVBL algorithm can automatically identify the number of sources without any prior knowledge.OG-WVBL algorithm can effectively identify the localization and amplitude of impulsive noise,deal with the problem of small snapshots and low SNR,and improve the accuracy of DOA estimation.3)Wideband DOA estimation based on focusing signal subspace:To solve the problem that the focusing transformation of coherent signal-subspace method needs the preliminary angles and the performance depends on the accuracy of initial angle estimation,this paper proposes a wideband DOA estimation based on unitary focusing signal subspace(FSS).FSS algorithm uses signal subspace of the reference frequency and signal subspace of other discrete frequencies to construct a unitary focusing matrix.The focusing matrix does not change SNR of array output and the focusing process is lossless focusing transformation.Besides,in order to deal with the coherent signal and reduce the complexity of FSS algorithm,FSS based on real-value signal(RFSS)is further designed.The real covariance matrix is constructed with unitary transformation technique,and decorrelation is realized by using forward backward spatial smoothing.The experimental results show that FSS algorithm will not be affected by the preliminary estimation error,which improves the focusing performance.RFSS algorithm can effectively estimate coherent and incoherent wideband signal sources.4)Microphone array system based on FPGA and DSP:The narrowband DOA algorithms for impulsive noise are combined with FSS algorithm,which builds the unified model of narrowband and wideband signal.Then,DOA estimation platform based on FPGA and DSP chips is designed and established.The effectiveness of innovations 1,2,3 for DOA estimation is verified for static and dynamic truck under the Gaussian noise environment and impulsive noise environment respectively.Therefore,this paper can be effectively applied to the acoustic localization systems such as battlefield monitoring and video conference,and has important practical value.