Collaborative Research: Expedite CSI Processing with Lightweight AI in Massive MIMO Communication Systems

下一代无线通信将需要支持在通信、计算和功率方面具有不同功能的异构设备，以提供具有各种性能需求（如高数据速率、低功耗和低延迟）的应用程序。大规模多输入多输出（MIMO）已被广泛认为是未来无线通信网络中实现高容量和高频谱效率的重要技术。为了充分发挥大规模MIMO通信系统所要求的潜在性能增益，在发射机，特别是在基站侧获得及时准确的信道状态信息（CSI）至关重要。该项目的主要目标是探索一种在大规模MIMO通信系统中以数量级加速CSI处理的系统方法。该项目将为提高下一代无线通信的数据速率和能源效率、频谱效率奠定基础。与该项目相关的研究工作可以对无线通信系统的轻量级人工智能（AI）设计产生重大影响，这将进一步改善许多应用领域，包括超5G无线网络、自主机器对机器通信、车载网络和物联网。该项目的成果可以促进我们社会向智能无线网络时代的过渡，无线通信系统可以提供无缝支持，以匹配大量网络设备的多种不同无线应用，并支持具有高计算需求和高服务质量需求的多种业务。此外，首席研究员致力将研究与教育结合，在三所参与大学的现行电气及电脑工程课程中引入无线通讯系统的新兴计算及轻量级人工智能。该项目还将为学生提供学习、开发和应用先进无线通信的机会，这些是他们无法从传统的理学士或硕士课程中获得的。由于复杂的传统模型和人工智能模型开发以及不同环境下性能的不一致，在大规模MIMO系统中满足相干时间要求对于CSI处理来说是非常困难的。本课题将对CSI重构过程中解压缩特征的优化、多速率压缩重构的AI结构简化、自主CSI重构性能评价和AI模型更新等方面的新算法进行理论分析和性能评价。优化的特性和简化的AI结构可以显著降低每秒浮点运算（FLOPs）的复杂性。因此，在大规模MIMO通信系统中，AI实现可以在不损失及时CSI处理重建精度的情况下加速1到2个数量级。系统的方法可以很容易地扩展，以促进许多其他应用程序，这些应用程序遇到类似的挑战，并在减少延迟和计算需求方面提出类似的需求。此外，该研究项目可以极大地促进对人工智能支持的大规模MIMO系统的理解，以获得更好的频谱和功率效率，并将从根本上有助于设计需要高度自治的高效机器对机器通信。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。

Next generation wireless communications will need to support heterogeneous devices with different capabilities on communications, computations, and power to deliver applications with various performance demands such as high data rate, low power consumption, and low latency. Massive multiple-input multiple output (MIMO) has been widely considered a compelling technology for achieving high capacity and high spectrum efficiency in the future wireless communication networks. To fully unleash the potential performance gains claimed by massive MIMO communication systems, it is of vital importance to have timely and accurate channel state information (CSI) at the transmitters, especially at the base station side. The main goal of this project is to explore a systematic approach that accelerates the CSI processing by orders of magnitude in massive MIMO communication systems. The project will lay a foundation to enhancing data rate and energy efficiency, spectral efficiency in the next-generation wireless communications. The research efforts associated with the project can have a significant impact on the lightweight artificial intelligence (AI) design for wireless communication systems, which will further improve many application domains, including beyond 5G wireless networks, autonomous machine-to-machine communications, vehicular networks, and Internet-of-Things. The outcomes of the project can foster the transition of our society into the intelligent wireless networking age, where wireless communication systems can provide seamless support to match many different wireless applications for massive network devices and support many services with high computation demands and quality of service needs. Moreover, the Principal Investigators are committed to integrating research and education by introducing emerging computing and lightweight AI in wireless communication systems into the current electrical and computer engineering curricula in the three participating universities. The project will also provide opportunities for students to learn, develop and apply advanced wireless communications, which they would not receive from a traditional B.S. or M.S. curriculum.Meeting the coherence time requirement in massive MIMO systems can be extremely difficult for CSI processing due to the complex traditional model as well as AI model development and inconsistent performance across environments. In this research project, theoretical analysis and performance evaluations will be obtained for novel algorithms designed for 1) optimization on the decompressed feature in the CSI reconstruction process, 2) simplifying the AI structures for multi-rate compression and reconstruction, and 3) autonomous CSI reconstruction performance evaluation and AI model update. The optimized features and simplified AI structures can significantly reduce the complexity in terms of floating point operations per second (FLOPs). Thus, the AI implementation can be accelerated by 1 to 2 orders of magnitude without losing reconstruction accuracy for timely CSI processing in massive MIMO communication systems. The systematic methodologies can be readily extended to facilitate many other applications that encounter the similar challenges and present similar needs on reducing latency and computation needs. Furthermore, this research project can greatly promote the understanding in AI-supported massive MIMO systems for better spectrum and power efficiency and will contribute fundamentally to the design of highly efficient machine-to-machine communications that require high level of autonomy.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.