Comparison of time domain noise source localization techniques: Application to impulsive noise of nail guns

Microphone array techniques are an efficient tool to detect acoustic source positions. The standard technique is the delay and sum beamforming. In the time domain, the generalized cross correlation of the microphone signals is used to compute the Spatial Likelihood Functions (SLF) of all microphone pairs and the noise source map is provided by the arithmetic mean of these functions. To improve the former noise source map, which means narrowing the main lobe and removing side and spurious lobes, several techniques have been developed in the past. In this work, the performances of three of these techniques (in terms of source position detection, amplitude estimation and computation time) are compared in the case of both synthetic and real data: (1) energetic and geometric criteria are applied in order to remove the SLF with useless information, (2) the arithmetic mean is replaced by the generalized mean and (3) linear inverse problem is solved with sparsity constraint. In the case of real data, the source to be located and quantified is an impulsive noise radiated by nail guns which is recorded by a spiral arm microphone array.