Distributed Acoustic Signal Processing over Wireless Sensor Networks

在“无线传感器网络分布式声学信号处理”的提案中，我们申请了相应的先前项目的扩展。我们仍然考虑相同的基本场景：声学传感器网络包括多个麦克风（可能还有扬声器），通过无线链路进行通信，通常通过一个或几个网关节点连接到互联网。在这样的网络内，声学应用（例如，扬声器分离）。一个简单的解决方案可能是将所有数据传输到网关节点并在那里处理它。然而，这不一定是最佳解决方案（数据速率瓶颈，大声学数据的大延迟）;它还浪费了在网络内部（预）处理数据的机会，因为它向网关传输。因此，我们研究的场景（灵感来自于微服务或网络功能虚拟化等趋势）可以将声学信号处理链分解为单个块并分布到传感器网络的节点上。在这个扩展项目中，我们继续研究声学处理以及自动分配此类任务的框架。具体来说，我们将研究硬件方面（特别是全双工音频），如何利用此类硬件在MAC协议中进行同步以及声学往返时间的估计。这将使我们能够在静态和动态环境中估计和校准场景几何形状。 考虑到来自合作伙伴项目的声源的实用信息，我们考虑从声学和网络的角度来选择哪些源。最后，一个关键问题将是如何处理动态场景的网络方面，其中动态来自不受控制或受控制（例如，基于机器人的）设备或环境的移动性。我们的工作将最终形成一个联合测试平台，利用所有关键功能并支持不同算法思想的实验。

In the proposal "Distributed Acoustic Signal Processing over Wireless Sensor Networks", we apply for an extension of the corresponding previous project. We still consider the same basic scenario: an acoustic sensor network comprising multiple microphones (and possibly speakers), communicating via wireless links, typically connected to the Internet via one or few gateway nodes. Inside such a network, acoustic applications (e.g., speaker separation) should be executed. A straightforward solution might be to transport all data to a gateway node and process it there. This is, however, not necessarily the best solution (data rate bottlenecks, large latency for big acoustic data); it also squanders the opportunity to (pre-)process data inside the network as it travels towards a gateway. We hence look at scenarios (inspired by trends like microservices or network function virtualization) where the acoustic signal processing chain can be broken down into individual blocks and distributed onto the nodes of the sensor network.In this extension project, we continue our work on both acoustic processing as well as a framework for the automatic distribution of such tasks. Specifically, we will look at hardware aspects (in particular, full-duplex audio), how to leverage such hardware for synchronization in MAC protocols and the estimation of acoustic round-trip times. This will allow us to estimate and calibrate scenario geometries in static and dynamic environments. Taking into account utility information about acoustic sources derived by partner projects, we consider which sources to select from both an acoustic and a networking perspective. Finally, a key question is going to be how to handle the networking aspects of dynamic scenarios, where dynamics comes from either uncontrolled or controlled (e.g., robot-based) mobility of devices or environment. Our work will culminate in a joint testbed that harnesses all key functionality and support experimentation with different algorithmic ideas.