The Research On Acoustic Source Localization Technology Based On Microphone Arrays

The research of sound source localization based on microphone arrays is related to acoustic,electronics,array signal processing and many other topics.Direction of arrival(DOA)estimation is one of the main issues in array signal processing for sound source localization.Traditional DOA estimation algorithms are based on the data model of uniform linear arrays.However,not only sparse arrays can evade the interaction between sensors because the inter-element spacing of the sparse arrays can break the limit of half wavelength of the carrier wave,but also sparse arrays can obtain the increased number of degrees of freedom and the extended array aperture with the same number of sensors from uniform linear arrays,thus the corresponding DOA estimation algorithms can achieve superior detection perfomance and estima-tion accuracy.Due to the invertibility of the array manifold for coprime array,the constraint of aliasing ambiguities because of subsampling can be well resolved.This dissertation discusses the high-precision and superresolution DOA esti-mation problem from the view of sparse signal recovery,with emphasis on DOA estimation algorithms based on the coprime array.Exploiting the spatial sparsity of the incident signals,the sparse representation algorithms have witnessed a signif-icant enhancement in the scenarios of reverberations and background noises.The sparse Bayesian learning(SBL)algorithm,developed within the sparse Bayesian framework,computes the posterior distribution of the sparse weight vectors by tak-ing the prior statistical information of the signal into account,and then provides estimations of their covariance along with the mean through evidence maximization or Type-? Maximum likehood.The research of this dissertation is as follows:1.An underdetermined wideband DOA estimation algorithm is proposed for off-grid sources with coprime array via SBL.Exploiting the augmented covariance matrix,the coprime array can achieve a higher number of degrees of freedom and the extended array aperture to resolve more sources than the number of physical sensors.By utilizing the spatial sparsity of the incident signals,the SBL algorithm can guarantee the global optimality using fixed point updates.The SBL scheme for wideband DOA estimation can provide processing advantages especially at low signal-to-noise ratio(SNR)under the acquisition of a small number of samples.Nu-merical simulations are conducted to valide the effectiveness of the underdetermined wideband DOA estimation via SBL based on coprime array.Furthermore,SBL can achieve superior detection performance and estimation accuracy compared to other DOA estimation algorithms without knowing the prior information of the number of incident signals.2.As the sound source is wideband signals,the wideband sparse spectrum fit-ting(W-SpSF)algorithm is proposed to solve the aliasing ambiguities using coprime array.The coprime array can benefit from frequency diversity in handling the wide-band signals.Moreover,the inter-element spacing of coprime array is larger than half wavelength without ambiguous angles.To resolve the basis pursuit denois-ing(BPDN)problem corresponding to wideband nonnegative sparse signal recov-ery with identical sparse support but different basis matrices,W-SpSF algorithm is exploring l2,1-norm sparse regularization and a weighted covariance fitting crite-rion.Generally,traditional wideband DOA estimation algorithms are to partition wideband signals into multi-narrowband components and those signal subspace are aligned(focused)by proper transformation matrices associated with spectral density matrices of the narrowband components.Simulation results illustrate that W-SpSF algorithm can not only obtain high precision DOA estimation but also predict the resolvability of two close signals.3.An underdetermined DOA estimation algorithm called multiple measure-ment sparse Bayesian learning(MSBL)is proposed to meet the requirement of DOA estimation in demanding scenarios of computational efficiency and estimation ac-curacy using coprime array.Vectorizing the covariance matrix,MSBL algorithm is applied to the augmented covariance matrix according to the virtual uniform linear arrays.Exploiting the extended difference coarray from coprime array,MSBL can locate more sources than the number of physical sensors in the time domain.MSBL is a Bayesian compressive sensing method by employing an empirical Bayesian strategy to resolve l0 minimization problem.MSBL can also support the exact re-covery when there are more sources than the number of physical sensors.Numerical simulations are illustrated the superiority of MSBL with coprime array in terms of DOA estimation performance and available degrees of freedom in comparison to other DOA estimation algorithms.