I-Corps: Accurate and energy-efficient sensory stream analysis via configurable trigger signature detection

网络物理系统依赖于对真实世界现象的持续感知来指导数据收集、存储、分析、计算、通信、决策和驱动。传感器和传感处理系统通常部署在能源高度受限的环境中，在这种环境中，分析原始数据流所消耗的电力主导着整个平台的能源状况。通常，必须分析数据流的时间和空间触发器签名，以识别感兴趣的事件已经发生，或者相反，排除此类事件的可能性。通过分析到达的感觉流，触发信号可以在软件中灵活地检测到，但这需要大量的能量预算。该项目计划开发一种低功耗可重构硬件/软件基板，该基板专为1)识别输入数据流中的时间和空间触发特征以及2)动态调用软件例程以进行进一步处理和动态重构而定制。硬件基板可以适合于处理多个输入模式和时标，将易于针对大范围的触发签名进行训练和重新配置，并且将消耗最小的能量，从而使其适合于在实际场景中的持续部署。这项建议中描述的工作可能会使连续传感应用能够以较低的成本广泛部署在具有严格能源限制的移动或偏远环境中。有许多影响社会的领域可能适用于这样的系统。移动空间中的许多新的和新兴的终端用户应用程序也依赖于环境感知来触发上下文和位置感知计算或通信。例如，移动电话加速计数据可以用来推断用户的活动类型和水平，以及他或她的特定位置或背景(例如，爬进汽车的驾驶座会生成可识别的加速计签名，该签名可用于在驾驶时禁用文本消息)。灵活地为这些和其他应用部署连续传感的能力有可能改变这些市场，并创建全新的和不可预见的应用领域。

Cyber-physical systems rely on continuous sensing of real-world phenomena to guide data collection, storage, analysis, computation, communication, decision-making, and actuation. The sensors and the sensory processing system are often deployed in highly energy-constrained environments, where the power consumed to analyze the raw data stream dominates the energy profile of the entire platform. Typically, the data stream must be analyzed for temporal and spatial trigger signatures that either identify that an event of interest has occurred or, conversely, rule out the possibility of such an event. Trigger signatures can be detected flexibly in software by analyzing the sensory stream as it arrives, but this requires a significant energy budget. This project plans to develop a low-power reconfigurable hardware/software substrate that is tailored for 1) identifying temporal and spatial trigger signatures in the input data streams and 2) dynamically invoking software routines for further processing and on-the-fly reconfiguration. The hardware substrate may be suitable for processing multiple input modalities and time scales, will be easy to train and reconfigure for a broad range of trigger signatures, and will consume minimal energy, thereby making it suitable for continuous deployment in practical scenarios. The work described in this proposal may enable widespread deployment of continuous sensing applications at lower cost and in mobile or remote environments with strict energy constraints. There are many domains impacting society that may be applicable for such systems. Many new and emerging end-user applications in the mobile space also rely on environmental sensing to trigger context- and location- aware computation or communication. For example, mobile phone accelerometer data can be used to infer the type and level of activity of the user, as well as his or her specific location or context (e.g. climbing into the driver's seat of a car generates an identifiable accelerometer signature which can be used to disable text messaging while driving). The ability to flexibly deploy continuous sensing for these and other applications has the potential to change these markets and create entirely new and unforeseen application domains.