Collaborative Research: CNS Core: Medium: Data Augmentation and Adaptive Learning for Next Generation Wireless Spectrum Systems

深度学习在解决无线网络研究和应用中的许多开放挑战方面显示出巨大的前景。深度学习需要数据，而实现其承诺的关键障碍之一是促进获取足够数量的数据来训练和验证深度学习模型。该项目的主要目标是设计创新的方法，使无线研究人员和从业者能够以更低的成本更有效地获取数据，并更有效地利用现有数据。该项目的发现有望通过使深度学习模型更广泛地适用来推动未来无线研究的突破。通过整合研究和教育，拟议的工作将为三所合作大学的本科生和研究生提供极好的动手练习、研究和教育机会。该项目将利用这三个机构现有的与多样性有关的外联方案，扩大代表性不足群体的参与。来自奥本大学、坦普尔大学和加州州立大学的四名研究人员组成的团队将开展一项连贯的研究议程，其中包括以下四项内容：(1)在生成性对抗网络的帮助下进行频谱数据合成和增强；(2)通过新颖的转移学习算法利用历史和合成的无线网络数据；(3)表征数据集大小和性能之间的关系；(4)整合、验证和应用前三项任务中开发的关于频谱数据库建设、射频频谱异常检测和发射机分类的方法。步骤1-3与应用无关，专注于研究基本概念和技术，这些概念和技术有助于获取足够量的无线数据，能够更有效地利用现有数据，并能够预测需要多少数据才能满足期望的性能。Struts 4是特定于应用的，专注于特定的无线应用，在这些应用中，深度学习已经被应用并展示了巨大的潜力。由该项目开发的数据、软件和教育材料将得到广泛传播。该项目将在与项目相关的问题上吸引行业利益相关者，目的是传播想法并了解行业在将深度学习应用于无线应用时面临的相关挑战。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Deep learning has shown great promise in solving many open challenges in wireless networking research and applications. Deep learning is data hungry, and one of the critical obstacles towards fulfilling its promise is facilitating the acquisition of sufficient amounts of data to train and validate deep learning models. The primary goal of this project is to devise innovative approaches that enable wireless researchers and practitioners to acquire data more efficiently at reduced cost and to utilize existing data more effectively. Findings from this project are expected to fuel future breakthroughs in wireless research by making deep learning models more widely applicable. By integrating research and education, the proposed work will provide excellent hands-on exercises, research, and educational opportunities for undergraduate and graduate students at the three collaborating universities. The project will leverage the existing diversity-related outreach programs at the three institutions to broaden participation from under-represented groups. A team of four investigators with complementary expertise from Auburn University, Temple University, and California State University, Sacramento will carry out a coherent research agenda consisting of the following four thrusts: (1) Spectrum data synthesis and augmentation aided by generative adversarial networks; (2) Exploiting historical and synthetic wireless networking data through novel transfer learning algorithms; (3) Characterizing the relationship between dataset size and performance; (4) Integrate, validate and apply approaches developed in the first three thrusts on spectrum database construction, RF spectrum anomaly detection, and transmitter classification. Thrusts 1-3 are application-agnostic and focused on studying fundamental concepts and techniques that facilitate the acquisition of sufficient amounts of wireless data, enable more effective utilization of existing data, and enable the prediction of how much data is needed to meet desired performance. Thrust 4 is application-specific and focused on specific wireless applications where deep learning has been applied and demonstrated great potential. The data, software and education materials developed from this project will be widely disseminated. The project will engage industry stakeholders on project-related issues, with the aim to disseminate ideas and learn relevant challenges faced by the industry when applying deep learning to wireless applications.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.