Study On The Robust Microphone Array System

Microphone arrays can be utilized to realize sound source localization,speech enhancement and sound field analysis using proper algorithms,leading to wide applications in acoustic signal processing.Microphone arrays applications are always challenged by the following adverse issues:the band with of the speech that needs to be processed is high,the array size is significantly limited by the hardware system and the room reverberation,near-field scattering and ubiquitous background noise have detrimental effect on array performance.Furthermore,the self-noise and the inconsistency of the microphone elements cannot be neglected.It is of great importance to investigate and improve the robustness of the microphone array systems in practical applications,which is the main topic of this dissertation.Three key technologies based on the microphone arrays,i.e.,beamforming,sound source localization and blind source separation,are studied.Conventional beamforming optimization algorithms usually emphasize only on distinguishing signals from different directions,while they cannot attenuate noise effectively from the direction of the expected signal.A set of linear constraints are introduced in several typical fixed beamforming techniques to improve distance discrimination.The robust broadband optimization method is proposed considering the errors in the microphone array characteristics.The performance of the proposed optimization methods are investigated by both simulations and experiments.Scatterers in the near-field of the microphone array inevitably affect the performance of the array system.A near-field scattering model is established for the commonly used linear microphone arrays,where the scatterer is assumed as a rigid sphere.Besides the far-field plane wave model and the near-field point source model,a reasonable human head model is also utilized as the incident sound source.The time domain simulation is used to analyze the influence of the scattering on the wave propagation.The effects of the scatterer on two typical beamformers,i.e.,the delay-and-sum beamformer and the superdirective beamformer,are investigated.Using microphone arrays and rigid spheres with different parameter characteristics,the influences of the near-field scattering on the performance of the linear microphone array beamforming are systematically analyzed by simulations and experiments.For sound source localization applications,the spherical microphone array is the most effective model for lumped arrays due to its unique localization results in the three-dimensional space.A multiple sound source localization method based on the maximum likelihood method is proposed in the spherical harmonic domain.To improve the computational efficiency,an efficient sequential iterative search of maxima on the cost function in the spherical harmonic domain is utilized.The proposed method avoids the division of the spherical Bessel function,which makes it suitable for both rigid-sphere and open-sphere configurations.Simulation results show that the proposed method has a significant superiority over the commonly used frequency smoothing multiple signal classification method.Experiments in a normal listening room and a reverberation room validate the effectiveness of the proposed method.The blind source separation algorithm based on the frequency domain independent component analysis is also analyzed in this dissertation.The performance of the commonly used permutation alignment methods is investigated,based on which a robust method combining the time difference of arrival and the inter-frequency correlation is proposed.Furthermore,a batch-processing real-time blind source separation system is designed,and its robustness and efficacy is validated by experiments using a two-elements microphone array.