CAREER: Harnessing Interference with Deep Learning: Algorithms and Large-Scale Experiments

蜂窝网络密度的增加和无人机的采用使干扰成为实现高效和高质量通信的重大障碍。现有的蜂窝间干扰解决方案由于信号开销和同步负担过大而不切实际。该项目利用深度学习来管理来自相邻单元的干扰，而不需要干扰信号的信道信息。拟议的算法将依赖于测量的信号和干扰功率或质量，这在现代蜂窝网络中是可用的，特别是在5G中。目标是提高蜂窝网络的频谱效率，实现无人机的广泛采用，并简化通信系统的设计。该项目还包括为费城未被充分代表的高中生提供教育服务。本科生将有机会进行无线通信和人工智能的跨学科研究，并将设计一门新的通信深度学习高级选修课程。这项研究的结果将通过高影响力的期刊、会议和讲习班传播，使学术界和工业界都受益。该项目将寻求在二维和三维蜂窝网络中减少干扰的新方法。所提出的深度学习辅助算法可以通过两种方式改变现实生活中蜂窝网络的设计和运行方式：(i)通过创建新的干扰管理基础，即使在信道信息未知和网络拓扑动态变化的情况下也能工作；（ii）设计端到端通信模型，可以适应干扰，简化当前的逐块设计，并产生显着更高的数据速率。所提出的算法将在小型和大型上进行广泛的测试，结果将在NSF PAWR平台上进行评估，从而可能导致现实世界的实施。该项目有可能显著提高当前和下一代蜂窝网络的效率，并使无人机在蜂窝网络中得到广泛应用。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 increasing density of cellular networks and the adoption of drones have made interference a significant obstacle to achieving efficient and high-quality communication. Existing solutions for inter-cell interference are impractical due to excessive signaling overhead and synchronization burden. This project leverages deep learning to manage interference from neighboring cells without requiring channel information for interference signals. The proposed algorithms will rely on measured signal and interference power or quality, which are available in modern cellular networks, particularly in 5G. The goal is to improve the spectral efficiency of cellular networks, enable widespread drone adoption, and simplify the design of communication systems. The project also includes outreach efforts to educate underrepresented high school students in Philadelphia. Undergraduate students will have opportunities for interdisciplinary research in wireless communications and artificial intelligence, and a new senior elective course on deep learning for communications will be designed. The findings of this research will be disseminated through high-impact journals, conferences, and workshops, benefiting both academic and industrial communities. This project will pursue a new approach for interference mitigation in two-dimensional and three-dimensional cellular networks. The proposed deep learning-aided algorithms could transform the way real-life cellular networks are designed and operated in two ways: (i) by creating new foundation for interference management that work even when channel information is unknown and the network topology dynamically changes; and (ii) designing end-to-end communication models that can adapt to interference, simplify the current blockby-block design, and yield significantly higher data rates. The proposed algorithms will undergo extensive testing on both small and large scales, and the results will be evaluated on the NSF PAWR platforms that could lead to real-world implementation. This project has the potential to significantly increase the efficiency of current and next-generation cellular networks and enable the widespread use of drones in cellular networks.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.