CAREER: Enabling Highly-Mobile Large-Scale MIMO Systems with Machine Learning

建立能够支持新兴无线应用的高效通信网络是美国和世界上许多其他国家的国家优先事项。未来的无线通信网络将需要提供比当前网络更高的数据速率，并支持高移动性的低延迟应用，如虚拟/增强现实和自动驾驶汽车。为了满足这些要求，该项目将利用机器学习来设计下一代无线通信系统，该系统可以有效地增加发射机和接收机的天线数量，这对于支持高数据速率至关重要。该项目还将利用其他传感信息，如GPS位置和摄像机图像，以提高无线通信网络的性能。在教育方面，该项目将导致本科生和研究生课程材料的开发，并将通过让本科生参与构建概念验证原型，为提高本科生的基本工程技能提供机会。此外，该项目中的拟议工作预计将通过向工业界转让技术以及公开发布项目期间开发的所有机器学习数据集和算法实现，对多个领域产生影响。该项目的主要目标是在现实动态环境中实现高移动性的大规模MIMO系统。为了实现这一目标，将开发能够有效利用先前系统观察和侧多模态感觉信息的全新机器学习方法。该项目有几个相互关联的目标：（i）开发统计信道预测方法，用于动态大规模MIMO系统中快速而稳健的信道捕获;（ii）表征跨频率和跨空间信道预测可行的基本条件;（iii）利用基于学习的条件信道协方差预测，为高度移动的大规模MIMO系统设计稳健和自适应的多用户预编码;（iv）开发一种新的多模式学习框架，用于视线和非视线场景中的快速毫米波波束预测，依赖于各种传感数据的有效融合，例如位置、6 GHz以下信道和视觉数据;（v）研究设计有效的多模态学习方法和算法，用于使用先进的视觉场景理解和端到端技术进行动态毫米波阻塞预测。该奖项反映了NSF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。

Building efficient communication networks that can enable the emerging wireless applications is a national priority for the United States and many other countries around the world. Future wireless communication networks will be required to provide significantly higher data rates compared to the current networks and to support highly-mobile low-latency applications such as virtual/augmented reality and autonomous vehicles. To respond to these requirements, this project will leverage machine learning to design next-generation wireless communication systems that can efficiently increase the number of antennas at the transmitters and receivers, which is essential for supporting high data rates. This project will also make use of the other sensory information such as GPS positions and camera images to improve the performance of the wireless communication networks. On the educational front, this project will lead to the development of undergraduate and graduate course materials and will provide opportunities for enhancing the basic engineering skills of undergraduate students through engaging them into building proof-of-concept prototypes. Further, the proposed work in this project is expected to have impact on a number of areas through the technology transfer to industry and the public release of all the machine learning datasets and algorithmic implementations developed during the project.The main goal of this project is to enable highly-mobile large-scale MIMO systems in realistic dynamic environments. To achieve this goal, fundamentally new machine learning approaches that are capable of efficiently leveraging the prior system observations and the side multi-modal sensory information will be developed. The project has several inter-related thrusts: (i) Developing statistical channel prediction approaches for fast yet robust channel acquisition in dynamic massive MIMO systems; (ii) characterizing the fundamental conditions under which across-frequency and across-space channel prediction is feasible; (iii) Designing robust and adaptive multi-user precoding for highly-mobile massive MIMO systems leveraging learning-based conditional channel covariance prediction; (iv) Developing a novel multi-modal learning framework for fast mmWave beam prediction in line-of-sight and non-line-of-sight scenarios, relying on the efficient fusion of various sensory data, such as positions, sub-6GHz channels, and visual data; (v) Investigating the design of efficient multi-modal learning approaches and algorithms for dynamic mmWave blockage prediction using advanced visual scene understanding and end-to-end learning.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.