Study On Generalized Cross-Correlation Based Localization Algorithms For Broadband Acoustic Source

Beamforming with microphone arrays is a popular technique to obtain acoustic map for visualizing source localization.It can be performed in the frequency domain and the time domain,but for broadband sound sources,the time domain techniques have a computational efficiency advantage over the frequency domain ones.Among them,the classic generalized cross-correlation(GCC)and its enhanced versions are increasingly widely used.In this paper,many GCC based enhancement algorithms are compared and analyzed,and the newly proposed inverse model method with the appropriate temporal width threshold was illustrated to obtain the best spatial resolution and sidelobe suppression.However,when the author researches this inverse model method,there is no selection rule for the temporal width threshold in the existing literature,so the GCC based inverse model needs to be perfected.To solve this issue,two effective methods are proposed to perfect the inverse model method and ensure its superior sound source localization performance.Firstly,the basic principles of time-domain delay and sum(DAS)and classic GCC are studied.On this basis,through analysis and comparison,it is clear that GCC can suppress the quantization error in DAS,but still suffers from defects such as wide mainlobes and plenty of sidelobes.Therefore,further analysis and comparison of the localization performance and computational efficiency are performed on three typical GCC enhancement algorithms,including the spatial weighting method,the generalized mean method and the inverse model method.Then a comprehensive evaluation indicates that the inverse model method is expected to obtain the best performance.The localization performance of inverse model method is sensitive to the temporal width threshold of the propagation model matrix.The appropriate threshold should vary with focus distance,array geometry and array size,but there is still a lack of uniform and effective methods to determine it.To solve this issue and perfect the technique,two methods are proposed:(1)a selection rule based on the cumulative probability of all differences of time delay estimation between the focus point and the microphone pair is proposed to determine the temporal width threshold,(2)an alternative propagation model matrix that circumvents the threshold is derived and constructed.Finally,simulations and experiments show that both proposed methods are effective under different test configurations,with strong stability and adaptability to focus distance,array geometry and array size.Either proposed method can ensure that the GCC based inverse model method enjoys satisfactory source localization performance,including narrow mainlobes,few spurious sources,and highlighted source positions.In addition,the former proposed method is comparable to the latter in terms of localization performance,but the computational efficiency is lower.