Time-Synchronization for Coherent Digital Signal Processing in Wireless Acoustic Sensor Networks

点击翻译按钮获取中文摘要

With the advent of smart home technology, acoustic sensor networks (ASN) have made a huge step from niche application to home appliance. Within an ASN, sensor nodes are not per se synchronized since individual oscillators in the nodes inevitably introduce a sampling-rate offset (SRO). Even small SROs in the order of parts per million may be detrimental to typical audio processing involving multiple nodes such as beamforming or source localization. Over a longer time period, small SROs will further accumulate substantial signal drifts that result in entirely asynchronous audio streams. Sampling rate synchronization is, hence, a crucial component for signal processing tasks in ASNs. In this project extension within the Research Unit (RU) "Acoustic Sensor Networks", we consider the dynamics that result from smart home settings as both challenge and opportunity for synchronization. In line with the overarching goals of the RU, we introduce the reliability of estimated SROs to reflect how well a particular acoustical scene lends itself to further signal processing. With several pairwise estimates obtained over time, we then fuse the information based on the reliability to obtain more stable synchronization in the long run. Furthermore, we extend the perspective from synchronizing two sensor nodes to an arbitrary number of nodes within a smart home by applying the fusion across the ASN. The 'Acoustic Map' foreseen in the RU acts as an interface to provide synchronization for other projects within the RU and allows high-level information to be exploited in order to facilitate synchronization efforts.