Age- and Deviation-of-Information of Signal-agnostic and Signal-aware Sensor Sampling in Networked Monitoring

我们考虑一个系统，其中传感器被采样，样本通过网络传输到远程监视器的延迟。典型的例子是一般的网络控制和网络信息物理系统。这样的系统的一个重要的性能指标是信息的年龄，量化的新鲜度的传感器数据在远程监视器。一般来说，我们的目标是找到一个抽样政策，最大限度地减少年龄的信息。在时间触发系统中，在某些时间点进行采样，因此采样是信号不可知的。信号感知采样适用于事件触发系统，其中定义的传感器事件，例如，如果传感器信号的变化超过阈值，则触发新样本的传输。从实用的角度来看，事件触发系统是有吸引力的，但研究的程度要小得多。值得注意的是，信息年龄的概念本身是信号不可知的，因此不能表达信号感知采样的优点。在这个项目中，我们将考虑与年龄相关但信号感知的指标，这些指标考虑远程监视器上的信息与当前传感器信号的实际偏差。我们将开发信号感知的采样策略，并评估其偏差的信息性能。为此，我们将通过随机过程对传感器进行建模，并定义触发新样本传输的事件。我们将使用确定性和随机网络演算的方法来推导年龄和信息偏差的界限，这些界限不会被超过或只会以较小的定义概率被超过。我们将包括最新网络技术的服务模型，即，IEEE时间敏感网络/IETF确定性网络和5G超可靠低延迟通信，这与网络网络物理系统中的预期应用最相关。分析工作将伴随着试验台安装的实证研究。我们将考虑的两个示例性应用是网络化机器人，我们使用我们研究所的机器人/TSN测试平台，以及基于已知SUMO/OMNeT++模拟器耦合的车辆通信。

We consider a system where a sensor is sampled and the samples are transmitted via a network subject to latency to a remote monitor. Typical examples are in networked control and networked cyber-physical systems in general. An important performance metric of such systems is the age-of-information, that quantifies the freshness of the sensor data at the remote monitor. Generally, the aim is to find a sampling policy that minimizes the age-of-information. In time-triggered systems, samples are taken at certain points in time, thus the sampling is signal-agnostic. Signal-aware sampling applies to event-triggered systems, where a defined sensor event, e.g., if the change of the sensor signal exceeds a threshold, triggers the transmission of a new sample. Event-triggered systems are attractive from a practical point of view, but have been researched to a much lesser extent. Notably, the notion of age-of-information itself is signal-agnostic and hence cannot express the advantages of signal-aware sampling. In this project, we will consider age-related but signal-aware metrics that consider the actual deviation of the information at the remote monitor from the current sensor signal. We will develop signal-aware sampling policies and evaluate their deviation-of-information performance. For this, we will model sensors by random processes and define events with respect to these processes that trigger the transmission of new samples. We will use methods of the deterministic and stochastic network calculus to derive age- and deviation-of-information bounds, that are not exceeded or only exceeded with a small, defined probability. We will include service models of recent network technologies, i.e., IEEE Time Sensitive Networking/IETF Deterministic Networking and 5G Ultra Reliable Low Latency Communications, that are most relevant for the envisioned applications in networked cyber-physical systems. The analytical works will be accompanied by empirical studies in testbed installations. Two exemplary applications that we will consider are networked robotics, where we use our institute's robotic/TSN testbed, and vehicular communications, based on the known SUMO/OMNeT++ simulator coupling.