EAGER: ML-enabled early warning of blockage and beam transitions in mobile, hybrid sub-6GHz/mmWave systems

新一代蜂窝通信5G预计将利用数十至数百GHz（毫米波）范围内的频率，以克服4G（低于6 GHz）系统固有的带宽限制。然而，毫米波信号不能传播那么远，更容易被物理物体阻挡，并且比4G信号需要更多的定向通信。事实上，2019年和2020年商用5G毫米波网络的早期部署遇到了重大的覆盖和渗透问题。该项目评估了一种新颖的、潜在的变革性方法的可行性，该方法使用低于6ghz（类似4g）的观测，在毫米波（5G）下获得阻塞和天线波束转换的早期预警。研究了机器学习（ML）实现此任务的适用性。增强了一个真实的基于物理的传播模型来验证所提出的方法。该项目由跨学科的PI团队加强，他们结合了通信理论，信号处理和传播物理方面的专业知识。所提出的方法引入了创新，以推进5G网络和毫米波传播建模的更广泛领域。拟议研究的见解将整合到课程和学生组织的报告中，并为专业人士发布结果。研究生和本科生将在一个多样化、多学科和包容的环境中接受有关充满活力的无线通信主题的培训。与美国国家科学基金会BWAC和PAWR平台的密切合作将通过在中心活动中传播研究成果来提高拟议研究和推广计划的成功。这个高风险、高回报的项目开发了新的数字信号处理（DSP）和ML解决方案，用于解决现实世界毫米波部署中的弹性问题。这些方法适用于同时使用6 GHz以下和毫米波频段的混合通信系统。利用菲涅耳衍射理论和我们精确的基于物理的模型，我们之前已经证明，衍射的sub-6 GHz信号比毫米波信号更早达到指定的接收信号强度（RSS）阈值。后一种特性在本项目中被用于开发一种预警方法，该方法可以在移动通信系统中提前几到几十毫秒（几到几百个槽）预测毫米波信号的阻塞、波束方向和其他快速变化。预警方法为混合移动通信系统提供了足够的时间来适应数据速率，改变天线方向，或者在毫米波信号发生重大变化之前在两个频率或基站之间执行切换。早期预警方法仅依赖于衍射的物理性质，而不依赖于测量的环境或密集光束或用户。它通过不断适应环境特征，例如移动障碍物或基站选址、环境测绘和其他先前提出的方法无法捕获的小型反射器，提高了移动毫米波网络对阻塞和其他快速信号变化的弹性。预警算法使用精确的时空sub-6 GHz/毫米波混合信道模型进行训练和验证，该模型可以提供大量物理现实场景。该项目中物理模型的利用将为在未来使用北卡罗莱纳州PAWR平台和在线数据存储库的混合通道测量中收集“智能数据”提供见解。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。

The emerging generation of cellular communication, 5G, is expected to utilize frequencies in the range of tens to hundreds of GHz (mmWave) to overcome the bandwidth limitations inherent to 4G (sub-6 GHz) systems. However, mmWave signals do not propagate as far, are more susceptible to blocking by physical objects, and require more directional communication than 4G signals. Indeed, early deployments of commercial 5G mmWave networks in 2019 and 2020 have suffered from major coverage and penetration problems. This project evaluates the feasibility of a novel, potentially transformative approach to obtain an early warning of blockages and antenna beam transitions at mmWave (5G) using sub-6GHz (4G-like) observations. Suitability of machine learning (ML) for enabling this task is investigated. A realistic physics-based propagation model is enhanced to validate the proposed approach. The project is strengthened by the interdisciplinary PI team with combined expertise in communication theory, signal processing, and propagation physics. The proposed methods introduce innovations to advance the broader fields of 5G networks and mmWave propagation modeling. The insights of the proposed research will be integrated into courses and presentations to student organizations, and the outcomes published for professionals. A graduate student and undergraduate students will be trained in a diverse, multidisciplinary, and inclusive environment about vibrant wireless communications topics. Close collaboration with NSF BWAC and PAWR platforms will enhance the success of proposed research and outreach plans by dissemination of the research outcomes in the centers' events.This high-risk, high reward project develops novel digital signal processing (DSP) and ML solutions for solving resiliency problems in real-world mmWave deployments. These methods are suitable for hybrid communication systems, where the sub-6 GHz and mmWave bands are employed simultaneously. Using the Fresnel theory of diffraction and our accurate physics-based model, we have previously demonstrated that diffracted sub-6 GHz signals reach a specified received signal strength (RSS) threshold much earlier than mmWave signals. The latter property is exploited in this project to develop an early-warning method that forecasts blockage, beam direction, and other rapid changes in mmWave signals several to tens of milliseconds (several to hundreds of slots) ahead in mobile communications systems. The early-warning approach provides hybrid mobile communications systems with sufficient time to adapt the data rate, change the antenna direction, or perform a handover between the two frequencies or base stations before a significant change of the mmWave signal occurs. The early warning method relies solely on the physical properties of diffraction, not on measured environments or dense beams or users. It improves resilience of mobile mmWave networks to blockages and other rapid signal changes by continuously adapting to the environmental features, e.g., moving obstacles or small reflectors not captured by base-station siting, environmental mapping, and other previously proposed approaches. The early warning algorithm is trained and validated using an accurate spatiotemporal sub-6 GHz/mmWave hybrid channel model, which can provide a large set of physically realistic scenarios. Utilization of the physical model in this project will provide insights for collecting 'smart data' in future hybrid channel measurements using PAWR platform at NC State and online data repositories.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.