Robust Covert Wireless Communications: Fundamental Limits and Algorithms

无线通信网络是民用和军用国家基础设施的重要组成部分。由于这些网络传播的无线电信号很容易被观察到，因此安全和隐私是主要问题。许多安全和隐私研究的重点是保护消息内容不被窃听，但在一些重要的应用中，即使是对手检测到通信信号的存在也是不可取的。在民用环境中，检测到嵌入式医疗设备发出的信号会泄露健康信息，或者检测到各方之间传输的加密信号会导致通信被威权政府关闭。在军事环境中，通信信号的存在可以作为部队活动的代表。因此，不可检测或低概率检测（LPD）通信激发了重要的历史和最近的研究。鉴于算法和计算的最新进展，防止无线电信号被检测是具有挑战性的。基于人工智能（AI）技术的算法进步和基于量子计算的计算进步为对手提供了一套强大的工具来开发信号探测器。这激发了十年前由该研究团队的一部分发起的信息系统研究的一个分支，现在被称为“秘密通信”，该研究正在发展对双方通信能力的基本理解，而不会被细心和有能力的对手发现。该项目将开发用于支持无线电通信模型的秘密通信成果，以保护对现代社会至关重要的无线通信系统中的信息。对于离散时间加性高斯白噪声（AWGN）信道，Bash、Goeckel和Towsley发起了最近的秘密通信工作，他们在2012年展示了发射机Alice可以在n个信道中可靠地向接收方Bob传输O（sqrt(n)）比特，而不会被细心而有能力的对手Willie发现。几乎所有隐蔽通信的工作都是在离散时间模型上进行的，隐含的理解是，结果将适用于真正的连续时间系统。但对于用于隐蔽无线通信的重要新兴架构来说，情况并非如此；相反，必须直接考虑连续时间信道。本项目将考虑隐蔽通信系统的基本特征，因为它们被认为是通过三个研究重点在无线通信中实现的：1。连续时间系统中隐蔽通信的基础：面对连续时间无线系统中对手威利可用的工具，我们专注于性能表征和设计：干扰消除，循环静止检测器和发射机识别。2. 存在环境干扰时的隐蔽通信：隐藏在环境信号后面的隐蔽通信比使用干扰更可取。我们将重点放在隐蔽吞吐量边界作为干扰动力学的函数和实现这些边界的技术上。3. 基于学习的隐蔽通信方法：我们通过同时训练对手来设计和分析基于深度神经网络（DNN）的隐蔽收发器，该对手的性能被用作发射器Alice训练中的正则器。在安全和隐私方面培训多样化的工作人员是该项目的一个重要方面。第三个重点是考虑通信双方和对手的基于人工智能的方法，为马萨诸塞大学工程和计算机科学专业的本科生提供了一个可访问且非常理想的研究领域。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Wireless communication networks form a critical part of the national infrastructure for both civilian and military applications. Because these networks propagate a radio signal that can be readily observed, security and privacy are major concerns. Much of security and privacy research focuses on protecting the content of messages from eavesdroppers, but there are important applications where even the detection of a communication signal's presence by an adversary is undesirable. In a civilian setting, the detection of a signal from an embedded medical device leaks health information, or the detection of encrypted signals transmitted between parties can lead to that communication being shut down by an authoritarian government. In a military setting, the presence of communication signals can be used as a proxy for troop activity. Hence, undetectable or low probability of detection (LPD) communications has motivated significant historical and recent study. Protection against the detection of a radio signal is challenging given recent advances in algorithms and computation. Algorithmic advances through techniques based on artificial intelligence (AI) and computational advances based on quantum computing provide the adversary with a powerful set of tools for developing a signal detector. This has motivated a branch of information systems research initiated by a portion of this research team ten years ago, now termed “covert communications,” that is developing a fundamental understanding of the ability of two parties to communicate without detection by an attentive and capable adversary. This project will develop results in covert communications for the models that underpin radio communications to allow for the protection of messages in the wireless communication systems that are crucial to modern society.For discrete-time additive white Gaussian noise (AWGN) channels, Bash, Goeckel, and Towsley initiated recent covert communications work by demonstrating in 2012 that a transmitter Alice can reliably transmit O(sqrt(n)) bits in n channel uses (and no more) to recipient Bob without detection by an attentive and capable adversary Willie. And nearly all work in covert communications has been performed on discrete-time models, with the implicit understanding that the results would be applicable to the true continuous-time system. But this is not true for important emerging architectures for covert wireless communications; rather, the continuous-time channel must be considered directly. This project will consider the fundamental characterization of covert communication systems as they are considered for implementation in wireless communications through three research thrusts: 1. Foundations of covert communications in continuous-time systems: We focus on performance characterization and design in the face of tools available for the adversary Willie in continuous-time wireless systems: interference cancellation, cyclostationary detectors, and transmitter identification. 2. Covert communications in the presence of environmental interference: Covert communications hidden behind environmental signals is preferable to employing jamming. We focus on covert throughput bounds as a function of the interference dynamics and techniques to achieve those bounds. 3. Learning-based covert communication approaches: We design and analyze deep neural network (DNN)-based covert transceivers by the simultaneous training of an adversary whose performance is used as a regularizer in the training of the transmitter Alice. The training of a diverse workforce in security and privacy is an important aspect of the project. The third thrust that considers AI-based approaches at both the communication parties and the adversary provides an accessible and highly desirable research area for University of Massachusetts undergraduate students in engineering and computer science.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.