Coding for Networked Control Systems over Lossy Links

未来十年将见证具有无线通信功能的设备数量的爆炸式增长，这些设备可以以特别和/或集中的方式感知、测量、驱动、交互和相互通信。这些设备将出现在汽车、家用电器、可穿戴电子产品、传感器网络、监控系统、各种控制器和许多其他设备中。这些新兴设备的统称是物联网（Internet-of-Things）。如此众多的智能设备所带来的一个巨大机遇是，它们可以合作、协调、共同决策，并通过采取联合行动来影响它们所处的环境。实现这一机会的一个主要挑战是，这些多个设备或代理必须以分散的方式进行操作，并且通过不可靠的链路进行通信。虽然通信系统一直必须处理不可靠的链路，但由于需要交换的控制和传感信号的实时性、动态变化的交通模式以及代理可能的移动性，为通信系统设计的传统方案已不再适用。本提案的目标是促进网络系统的设计和分析，这些系统在嘈杂的通信链路上相互作用和通信，并建立在研究者的基础上。构建第一个可有效解码的树码的前期工作，该树码可用于在有损擦除链路上可靠地实现任何分布式协议。如何在有损链路的网络上最好地操作分布式异构代理是一个活跃的研究领域。问题包括代理应该交换什么信息？他们应该如何最好地量化和编码他们的测量或控制信号，以对抗网络中的不确定性（延迟或消息丢失）？如何促进合作，克服竞争？他们如何以分布式的方式做到这一切呢？等等......。这些非常具有挑战性的问题与通信、网络和控制等传统领域重叠，并以新颖的方式将每个领域的元素结合起来。因此，一种能够以整体方式处理它们的理论尚未出现。本提案试图通过研究树编码来开始这样一个理论，树编码是一种新的编码范式，允许在有损链路上可靠地实现交互协议，也可以通过研究因果源编码。树码在近20年前被引入，作为信息理论和控制之间的桥梁，但由于没有有效解码的明确结构，因此没有获得太多的关注。因果源编码领域的探索甚至更少。提出的研究主要有五个重点：(1)为其他类别的信道（如AWGN和BSC）设计有效的可编码和可解码的树码；(2)研究控制系统的性能以及源编码、信道编码和控制器设计之间的相互作用和权衡；(3)研究因果源编码；特别是，因果转换代码以及如何在控制设置中优化它们，(4)在擦除链接上实现分布式协议，以及(5)从控制理论确定指标以指导代码设计。

The coming decade will witness an explosive growth in the number of devices with wireless communication capabilities and that can sense, measure, actuate, interact, and communicate with each other in ad hoc and/or centralized fashions. Such devices will appear in automobiles, household appliances, wearable electronics, sensor networks, surveillance systems, various controllers and many others. A common catch-all phrase for these emerging devices is the Internet-of-Things. One of the great opportunities that such a plethora of intelligent devices presents is that they can cooperate, coordinate, make joint decisions, and influence the environment they are in by taking joint actions. A main challenge in realizing this opportunity is that these multiple devices, or agents, must do so in a decentralized way and while communicating over unreliable links. Although communication systems have always had to deal with unreliable links, because of the real-time nature of the control and sensing signals that need to be exchanged, the dynamically-changing traffic patterns, and the possible mobility of the agents, traditional schemes designed for communication systems are no longer applicable. The goal of this proposal is to facilitate the design and analysis of networked systems that interact and communicate with each other over noisy communication links and builds on the investigators? prior work of constructing the first efficiently decodable tree codes that can be used to reliably implement any distributed protocol over lossy erasure links. How best to operate distributed heterogeneous agents over networks of lossy links is an area of active research. Questions include what information should the agents exchange? How best should they quantize and encode their measurements or control signals to combat the uncertainties (delay or message drops) in the network? How can they facilitate cooperation and overcome the competition? How can they do all this in a distributed fashion? And so on. These very challenging questions overlap with the conventional fields of communications, networking and control and combine elements of each in novel ways. As result, a theory that can deal with them in a holistic manner has yet to emerge. This proposal attempts the beginning of such a theory through the study of tree codes, a new paradigm in coding that allows the reliable implementation of interactive protocols over lossy links, as well as through the study of causal source coding. Tree codes were introduced nearly twenty years ago as a bridge between information theory and control, but did not gain much traction since explicit constructions with efficient decoding did not exist. The area of causal source coding is even less explored. The proposed research has five main thrusts: (1) the design of efficiently encodable and decodable tree codes for other classes of channels, such as AWGN and BSC, (2) the study of the performance of control systems and the interaction and tradeoffs between source coding, channel coding and controller design, (3) the study of causal source coding; in particular, causal transform codes and how they can be optimized in the control setting, (4) implementing distributed protocols over erasure links, and (5) determining metrics from control theory to guide the code design.