Research On Optimal Sound Source Localization Algorithm Based On Expected Value Maximization

With the development of related technologies such as sensor networks and computers,sound source localization has been widely used in different fields such as industrial inspection,video conferencing,and intelligent robots.It extracts relevant characteristic parameters from the received acoustic signals for sound source localization.At present,researchers at home and abroad are committed to improving the performance of sound source localization systems.Their research mainly focuses on noise resistance,localization accuracy,real-time performance.Based on wireless acoustic sensor network localization,this thesis mainly analyzes and discusses the following parts:1.In practical application scenarios,environmental factors such as noise and reverberation causes the decrease of the time delay estimation performance,and a sub-band time delay estimation method based on the expectation maximization algorithm is presented.This method is based on the generalized cross correlation algorithm.The signal received by acoustic sensors is divided into sub-bands in the frequency domain,and the GCC of each frequency band is derived.Then the maximum likelihood model is established for all sub-band GCCs,and the iterative formula under this condition is deduced according to the maximum expected value algorithm.Finally,the time delay is estimated from the optimal frequency band.The method uses the iterative steps of the EM algorithm,so it has the excellent convergence just as the traditional EM algorithm.Simulation experiments show that compared with the traditional GCC algorithm,EM-GCC has improved stability and noise immunity.2.In outdoor sound source localization,the signal-to-noise ratio of the signal received by different acoustic sensors is inconsistent due to the complex background noise and wide search area.After analyzing and discussing the Time Difference of Arrival and the Steered Response Power algorithms,the EM-SRP is proposed.During reconstructing the GCC of each frequency band,the method introduces a single-peak Gaussian function to modulate the waveform of the sub-band GCC.The refined GCC waveform can adapt to the grid scale,and reduces the influence of spurious peaks.Both indoor reverberation and outdoor simulation and field experiment verify that the EM-SRP method achieves better results in both noise immunity and anti-reverberation capability.3.This thesis introduces a real time location system based on SRP,including hardware components and software components.The system realizes the whole location process based on the SRP algorithm,and finally we uses the explosion experiment to test the system and the EM-SRP algorithm.