Research On Multi-sound Source Localization Method Based On Acoustic Sensor Network

Based on the National Science Foundation of China(61401203)and the natural science foundation of Jiangsu province(BK20130776),the paper studies the localization of multiple sound sources with two-node distributed acoustic sensor networks.Acoustic sensor network is a new information acquisition and processing technology which integrates sensor technology,embedded computing and distributed information processing technology.It is composed of a large number of micro-sensor nodes with acoustic signals and computing power.Sound source localization is one of the most important applications of acoustic sensor networks,and has been applied in many military,ecological and commercial fields including sniper positioning system,field bird monitoring system and video conference.The sound source localization method based on sensor network is one of the research hotspots in the field of signal processing in recent years.In this paper,a multi-acoustic source localization method based on acoustic sensor network is studied.After introducing the concept of acoustic sensor network,three sound localization strategies based on acoustic sensor network are compared and compared.The DOA cross-location strategy was chosen to research on.Then,the traditional node-level direction finding method is introduced.For DOA estimation of multi-sound sources in reverberation environment with a widely spaced microphone array,a pooled angular spectrum method based on iterative signal-to-ratio(SNR)weighting and diffuse noise model is proposed.Using the property of time-frequency sparsity and short-time orthogonality of speech sources,the approach eliminates false peaks in each time-frequency local angular spectrum based on the frequency-dependent locations of aliasing-induced grate lobes.As a result,the influence of spatial aliasing is considerably suppressed in the final pooled angular spectrum.for spatial ambiguity problem in multi-source DOA cross-localization of acoustic sensor network composed of multiple microphone array nodes,a unambiguous DOA association algorithm is studied,but the estimation of the sound source location will result in the spatial ambiguity problem.In response to these problems,based on minimum variance distortionless response(MVDR)beamforming,the time-frequency spectra containing the corresponding energy distribution features associated with those sources are reconstructed in each node-array.Then,scale invariant feature transform(SIFT)is employed to solve the DOA association problem.Performance evaluation is conducted with field recordings and experimental results prove the effectivity and feasibility of the proposed method.