Distributed Source Localization in Acoustic Sensor Networks Using the Coherent-to-Diffuse Power Ratio

摘要

In this paper, acoustic source localization in an acoustic sensor network (ASN) based on estimates of the coherent-to-diffuse power ratio (CDR) is investigated. It is shown that the CDR reflects the distance between an acoustic source and an observing microphone pair in real-world acoustic enclosures. As this relation is dependent on various physical parameters of the given acoustic environment which are unknown in practice, we estimate it by Gaussian process regression. Subsequently, the obtained distance estimates are used by an ASN with known relative node positions for estimating the position of the acoustic source. The algorithm is designed in a completely distributed fashion, i.e., no fusion center is needed. The computational complexity of the proposed algorithm is low and the amount of data to be shared between the sensor nodes is small and can be adjusted to the constraints imposed on the network by using a sparse Gaussian process regression method. A simulation study confirms, employing also real measurement data in a low-reverberant room, the efficacy of the proposed algorithm for a multitude of scenarios.