Blind 3D localization and separation of multiple vibration and acoustic sources simultaneously active

Abstract

Signal source localization and separation are key tasks for many applications. In this paper, a new deterministic method is proposed for estimating the 3D location and separating multiple acoustic or vibration sources, simultaneously active. The method is based on TDOA measurements obtained via crosscorrelation. Then, with the information from the estimated locations, source separation is achieved. The performance of the method was evaluated, only through simulations, in terms of accuracy and computational load. The obtained results, with SNR=6dB, showed estimation errors for the localization method always bounded below 10 centimeters, obtaining the best results with a lower number of sources and a higher number of receivers. Furthermore, the computational load was substantially reduced as compared to exhaustive search, being the gain more noticeable for a higher number of sources. The corresponding error for the separation of the original source signals was bounded below 25%, for 2 sources and 6 receivers. Thus, there are strong evidences that the here-proposed method is accurate and robust enough, while being efficient in computational terms, so many applications can benefit from its use.