WiFiUS: Collaborative Research: Secure Inference in the Internet of Things

下一代通信网络预计将在2022年前提供互联多达1万亿个传感器、产品、机器和设备的物联网。这项研究解决了这些重要新兴网络中的关键安全问题。由于典型物联网终端的能力有限，预计的实时物联网应用对延迟的严格要求，以及无线信道的广播性，本项目研究了物理层和低复杂度的密码方法来防止窃听者访问物联网数据。物联网终端将经常通过短消息进行通信，这需要超越现有基于无限长消息的理论。由于推理在物联网系统中起着核心作用，因此本研究主要关注基于推理的度量方法。该项目支持在三所参与的大学进行学生培训，在主要会议上组织研讨会，并开发课程材料。基于项目团队成员最近的工作，该团队正在开发基本性能指标的界限和近似值，以便理解和设计基于短包通信的推理网络。为了深入了解大规模推理网络，该团队正在研究推理性能的缩放规则，该定律可以在合法用户数量与窃听者数量的比率发生变化时实现最佳缩放行为。使用非二进制量化、符号翻转、信道状态信息和相关方法的新的实用的低复杂性密码方法正在开发中，预计将提供比现有的物联网数据保护方法更大的优势。补充已开发方法的分布式推理和学习方法也在开发中。物联网中分布式传感器的快速自举方法和低复杂度的估计、决策和分类方法也在不断涌现。

The next generation of communication networks is predicted to provide an Internet of Things (IoT) interconnecting up to 1 trillion sensors, products, machines, and devices by 2022. This research addresses the critical problem of security in these important emerging networks. Due to the limited capability of typical IoT terminals, the strict delay requirements of anticipated real-time IoT applications, and the broadcast nature of wireless channels, this project investigates physical layer and low complexity cryptographic methods to prevent eavesdroppers from gaining access to IoT data. IoT terminals will often communication via short messages, requiring new theory beyond the existing theory based on infinitely long messages. Since inference takes a central role in IoT systems, this research focuses on inference-based metrics. This project enables student training at the three participating universities, organization of workshops at major conferences, and development of course materials.Building on very recent work by the members of the project team, the team is developing bounds and approximations to fundamental performance indices in order to understand and design inferential networks based on short packet communications. To gain insight into large-scale inference networks, the team is investigating scaling laws for inferential performance which achieve optimum scaling behavior as one varies the ratio of the number of legitimate users to the number of eavesdroppers. New practical low-complexity cryptographic methods employing non-binary quantization, symbol flipping, channel state information and related approaches are under development and are expected provide significant advantages over existing methods of protecting IoT data. Distributed inference and learning methods that complement the developed approaches are also under development. Fast bootstrapping methods and low-complexity estimation, decision making and classification methods for distributed sensors in IoT are being derived as well.