NSF-AoF: CNS Core: Small: Machine Learning Based Physical Layer and Mobility Management Solutions Towards 6G

5G演进和未来6G蜂窝网络的目标是在更高的毫米波和亚太赫兹（sub-THz）频段上运行，因为大信道带宽的可用性可以进一步提高数据速率、延迟、服务质量和可靠性。然而，将这些频段用于移动无线电接入带来了巨大的技术挑战，包括电子设备的质量、成本和能源效率、极端路径损耗和传播特性，以及提供具有移动性支持的室内和室外网络覆盖的总体部署成本。考虑到这些挑战，该项目将研究机器学习算法的实用性，这些算法已经成功地解决了各个领域的复杂问题，用于设计涉及用户设备和基站的物理层技术和网络管理程序，旨在提高移动连接的鲁棒性和可靠性。该项目的预期贡献处于新兴6G标准和现代机器学习工具在高频无线通信中的应用的前沿。该研究将解决三个关键问题：i)推力1将开发机器学习辅助和数据驱动的方法，用于用户设备波束训练和跟踪，使用基于压缩感知的信道探测，以实现低延迟和对相控天线阵列损伤的鲁棒性。此外，它将通过使用深度强化学习来优化基于环境特征和用户轨迹的波束探测策略，加速基站侧的波束训练和跟踪。这种推力的结果将在机动性存在的情况下显著减少光束管理开销；ii)推力2将使用一种新的接收器处理架构，其中信号路径利用时域和频域的卷积神经网络层来补偿宽带正交频分复用（OFDM）接收器中的功率放大器非线性和相位噪声的影响，同时考虑到频率选择的多径信道效应。这种推力的结果将是提高发射机的功率效率，覆盖范围和误码概率。iii) Thrust 3将开发智能切换算法，通过使用分布式深度强化学习利用位置估计和波束电平参考信号功率测量来最大限度地减少用户连接中断，同时考虑用户速率要求并减少基站之间的测量报告和共享。研究方法将依赖于广泛的测量和数据集的生成，使用毫米波测试平台来开发和评估所提出的解决方案。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。

5G evolution and future 6G cellular networks are targeting operations at higher millimeter wave and sub-Tera Hertz (sub-THz) bands due to the availability of large channel bandwidths to further improve data rate, latency, quality-of-service, and reliability. However, the use of these bands for mobile radio access imposes substantial technical challenges, including the quality, cost- and energy-efficiency of the electronics, the extreme path loss and propagation characteristics, and the overall deployment costs to provide indoor and outdoor network coverage with mobility support. Considering these challenges, this project will investigate the utility of machine learning algorithms, that have been successful in solving complex problems in various domains, for designing physical layer technologies and network management procedures, involved in both user equipment and base stations, that aim to improve robustness and reliability of connectivity under mobility. The project’s expected contributions are at the forefront of emerging 6G standard and applications of modern machine learning tools in wireless communications at high frequency bands.The research will address three key thrusts: i) Thrust 1 will develop machine learning assisted and data driven approaches for user equipment beam training and tracking using compressive sensing-based channel probing for low latency and robustness to phased antenna array impairments. In addition, it will accelerate beam training and tracking on the base station side by using deep reinforcement learning for optimizing beam probing strategies based on environment characteristics and user trajectories. The outcome of this thrust will be significant reduction in beam management overhead in the presence of mobility; ii) Thrust 2 will use a novel receiver processing architecture where the signal path exploits convolutional neural network layers in both time and frequency domains to compensate the effects of the power amplifier nonlinearity and phase noise in a wideband orthogonal frequency division multiplexing (OFDM) receiver, while accounting for frequency-selective multipath channel effects. The outcome of this thrust will be improvement in transmitter power efficiency, coverage and bit error probability. iii) Thrust 3 will develop intelligent handover algorithms to minimize disruptions in user connectivity by exploiting position estimates and beam-level reference signal power measurements with distributed deep reinforcement learning while considering user rate requirements and reducing measurement reporting and sharing overheard between base stations. The research methodology will rely on extensive measurements and data set generations using millimeter wave testbeds for the development and evaluation of the proposed solutions.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.