Off-grid sound source localization based on Compressive Sensing

In a distributed sensor network for sound source localization, it is important to develop a method which can accurately localize the sound source with a small amount of data collected by the sensors. Traditional methods for combining the Compressive Sensing (CS) with the sound source localization are not always accurate and effective in real applications due to the fact that time delay between different sensor pairs is sometimes not integral times of sampling duration. This phenomenon is called off-grid problem in this paper, and the off-grid problem leads to an inaccurate result. In this paper, we proposed a method for accurate sound source localization using CS and Fractional-Delay (FD) Filter to overcome the off-grid problem. Experiments demonstrated that the proposed method reduced the TDOA estimating error rate to under 0.7%. Simulation results demonstrated that the proposed method improved the average sound source localizing error by 70.91%.