CAREER: Advancing Space Optical Communication Systems Via Hybrid Model-Based and Learning-Based Frameworks

空间光通信（SOC）可以提供比射频（RF）通信高几个数量级的数据速率，并有望成为空间通信网络的关键技术。然而，发展 SOC 仍需克服的局限性包括：1）大气通道是动态的且未被充分理解，妨碍了其统计特征； 2）传统的通信系统设计忽略了充分利用实时数据中的相关信息； 3) 实现 SOC（相对于 RF）性能增益所需的系统复杂性将迅速增加，因此 SOC 工程解决方案必须将复杂性作为约束。为了应对这些挑战，本研究开发了一个系统设计框架，该框架结合了基于模型和数据驱动的 SOC 设计范式，系统 1) 建模并预测大气通道的长期动态，2) 主动调整其通信和网络策略以适应环境动态，从而最大限度地提高数据速率、能源效率、频谱效率和链路可靠性方面的端到端系统性能。所提出的方法将展示 SOC 如何成为一个可靠的平台，补充现有技术，以满足未来系统的轻松部署、高数据速率和可承受的复杂性的要求。该项目的潜在好处包括通过光学无人机在贫穷国家部署宽带互联网，从而实现信息和教育的获取；确保飞机之间的连通性，从而提高航空旅行的安全性、可靠性和效率；提高太空探索任务的可靠性，从而增加我们发现的潜力。研究工作将与主要研究者的教育职业目标相结合，即促进本科生研究、鼓励高中生入学 STEM 并通过与该机构现有的多元化招聘和支持计划合作来招募代表性不足的学生。当前的通信系统要么难以大规模部署，要么受到 RF 频谱许可负担的限制。该项目的贡献意义重大，因为它们通过理论和实践框架的结合展示了 SOC 如何成为补充和增强现有技术的可靠平台。研究的第一个目标是推导出 SOC 的性能限制，它描述了精心设计的系统在各种相关设置（例如多址和中继信道）中可以实现的最佳错误概率和信道容量，并考虑了大气损害。为了减轻大气影响，将设计一个清晰的统计通道模型。这项工作将涵盖深空、近地和空间系统网络。虽然通过泊松信道模型可以很好地描述深空通信，但可以通过统计模型捕获大气衰减和指向误差的影响。根据通信场景，还可以合并输入相关或输入无关的高斯噪声。实现这一目标的方法基于应用信息和通信理论的工具以及非参数统计信道学习方法。第二个目标是开发强大的机器学习技术来执行信号分类、估计大气参数、确定输入和输出数据之间的映射并推断概率分布，以便设计能够在不严重依赖信道模型的情况下高效运行的通信系统。将考虑针对点对点和多用户设置的基于块结构深度神经网络 (DNN) 以及基于端到端 DNN 的设计。实现这一目标的方法主要依靠采用无梯度优化技术和基于块结构 DNN 的信道估计、信号分类和检测来设计 SOC 自动编码器。该项目由通信、电路和传感系统 (CCSS) 计划和刺激竞争性研究既定计划 (EPSCoR) 联合资助。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Space optical communication (SOC) can provide orders-of-magnitude higher data rates than its Radio-Frequency (RF) communication counterpart and promises to be a key technology for space communication networks. However, limitations to developing SOC yet to be overcome include: 1) the atmospheric channel is dynamic and not well-understood, preventing its statistical characterization; 2) the traditional communication system design ignores full use of relevant information from real-time data; 3) the complexity of the systems required to achieve the performance gain of SOC (over RF) would increase rapidly, hence SOC engineering solutions must incorporate the complexity as a constraint. To address these challenges, this research develops a systematic design framework which combines model-based and data-driven design paradigms for SOC and where the system 1) models and predicts the long-term dynamics of the atmospheric channel, and 2) proactively adapts its communication and networking strategy to the dynamics of the environment, thereby maximizing end-to-end system performances in terms of data rates, energy efficiency, spectrum efficiency, and link reliability. The proposed approach will demonstrate how SOC can be a reliable platform that complements existing technologies to fulfill the requirements of easy deployment, high data rates, and affordable complexity of future systems. Potential benefits of the project include deploying broadband internet via optical drones in poor countries, thus enabling access to information and education; ensuring connectivity between aircraft, thus improving the safety, reliability, and efficiency of air travel; and enhancing the reliability of space exploratory missions, thus increasing our potential for discovery. The research effort will be integrated with the principal investigator's educational career goal of promoting undergraduate research, encouraging enrollment of high-school students in STEM and recruiting underrepresented students by working with the institution's existing diversity recruitment and support programs.Current communication systems are either difficult to deploy at large scale or limited by the RF spectrum licensing burdens. This project's contributions are significant because they show, via a mix of theoretical and practical frameworks, how SOC can be a reliable platform that complements and enhances existing technologies. The first objective of the research is to derive the performance limits of SOC, which describe the best error probability and channel capacity that a well-designed system can achieve in various relevant settings such as multiple access and relay channels, and accounting for atmospheric impairments. To mitigate the atmospheric effects, a sharp statistical channel model will be devised. The work will encompass deep-space, near-earth and space system networks. While deep space communication is well described via the Poisson channel model, the effects of the atmospheric attenuation and pointing error could be captured via statistical models. Depending on the communication scenarios, an input-dependent or an input-independent Gaussian noise could also be incorporated. The methodology to undertake this objective is based on applying tools from information and communication theories along with a non-parametric statistical channel learning approach. The second objective is to develop powerful machine learning techniques to perform signal classification, estimate parameters of the atmosphere, determine the mapping between input and output data and infer probability distributions in order to design communication systems that can efficiently perform without relying heavily on channel models. Block structure Deep Neural Network (DNN)-based as well as end-to-end DNN-based designs, for point-to-point and multiuser settings, will be considered. The methodology to undertake this objective relies mainly on designing SOC auto-encoders with gradient-free optimization techniques and block structure DNN-based channel estimation, signal classification, and detection.This project is jointly funded by the Communications, Circuits and Sensing Systems (CCSS) Program and the Established Program to Stimulate Competitive Research (EPSCoR).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.