Radar Aided 6G Beam Prediction: Deep Learning Algorithms and Real-World Demonstration
Radar Aided 6G Beam Prediction: Deep Learning Algorithms and Real-World Demonstration
复制标题
DOI：
10.1109/wcnc51071.2022.9771564
复制
发表时间：
2021-11
期刊：
影响因子：
--
通讯作者：
Umut Demirhan;A. Alkhateeb
中科院分区：
文献类型：
--
作者：
Umut Demirhan;A. Alkhateeb
在毫米波（mmWave）和太赫兹（THz）MIMO通信系统中调整窄波束与高波束训练开销相关联，这使得这些系统难以支持高度移动的应用。如果关于发射器/接收器位置和周围环境的足够意识是可用的，则可以潜在地减少或消除该开销。在本文中，开发了利用雷达传感数据的高效深度学习解决方案，以指导毫米波波束预测并显着降低波束训练开销。我们的解决方案集成了雷达信号处理方法来提取学习模型的相关特征，从而优化其复杂性和推理时间。提出的基于机器学习的雷达辅助波束预测解决方案使用大规模真实世界毫米波雷达/通信数据集进行评估，并在现实的车辆通信场景中展示了其功能。除了完全消除雷达/通信校准开销外，所提出的算法能够实现约90%的前5个波束预测精度，同时节省93%的波束训练开销。这突出了解决毫米波/太赫兹通信系统中训练开销挑战的有希望的方向。
Adjusting the narrow beams at millimeter wave (mmWave) and terahertz (THz) MIMO communication systems is associated with high beam training overhead, which makes it hard for these systems to support highly-mobile applications. This overhead can potentially be reduced or eliminated if sufficient awareness about the transmitter/receiver locations and the surrounding environment is available. In this paper, efficient deep learning solutions that leverage radar sensory data are developed to guide the mmWave beam prediction and significantly reduce the beam training overhead. Our solutions integrate radar signal processing approaches to extract the relevant features for the learning models, and hence optimize their complexity and inference time. The proposed machine learning based radar-aided beam prediction solutions are evaluated using a large-scale real-world mmWave radar/communication dataset and their capabilities were demonstrated in a realistic vehicular communication scenario. In addition to completely eliminating the radar/communication calibration overhead, the proposed algorithms are able to achieve around 90% top-5 beam prediction accuracy while saving 93% of the beam training overhead. This highlights a promising direction for addressing the training overhead challenge in mmWave/THz communication systems.