SpecEES: Collaborative Research: Enabling Spectrum and Energy-Efficient Dynamic Spectrum Access Wireless Networks using Neuromorphic Computing

在过去的二十年中，由于对无线连接的需求不断增长，射频(RF)频谱的使用大幅增加。支持当前无线数据需求的现有网络技术预计将在未来十年大幅增加其容量，这就需要频谱和能源高效的通信策略。有两种流行的方法来有效利用射频频谱：一种是认知无线电网络，它允许移动用户共享主要分配给其他服务的频谱，例如电视广播、全球定位系统(GPS)、雷达、天气预报等，前提是移动用户对现有服务施加的干扰有限。另一种方法是通过扩展带宽、大规模多输入多输出(MIMO)系统和增密异质网络(HetNet)来增强移动宽带网络。然而，这两种方法都有局限性，对频谱效率和能源效率有不同的影响。此外，当前的硬件平台在支持高计算复杂性和低功耗方面显示出巨大的挑战。该项目介绍了一种新的网络体系结构及其使用神经形态计算设备的特定于应用的硬件优化。新的无线网络架构允许移动用户执行时空频谱感知，并主动搜索动态频谱接入(DSA)机会，以实现短距离和本地通信。同时，模仿生物神经过程的神经形态计算设备将被设计成以极低的功耗解决新的动态频谱接入方法的高计算复杂性。通过这种方式，我们将能够实现我们国家的下一代无线通信和网络，在动态的频谱环境中实现智能、频谱效率和能源效率。开发的概念和技术还将有助于实现国家宽带计划，该计划的目标是大幅提高射频频谱的利用效率。该项目有一个广泛的教育和推广计划，包括设计有关节能通信、模拟神经元电路和无线网络计算智能的新课程组件，两个合作机构联合培训研究生和本科生研究人员，以及通过研讨会和多样性计划接触电信行业和代表性不足的学生。短距离和本地通信对频谱和能源效率非常有利。该项目的研究目标是1)设计支持DSA的HetNet，以实现短距离/本地频谱接入，以提高频谱和能量效率；2)利用神经形态计算体系结构，以极高的能量效率高效地解决相关的资源分配问题。为了实现这一目标，该项目被组织成四个相互关联的研究推动力。推力1专注于使用MIMO收发信机进行时空频谱感知。推力2研究支持DSA的HetNet的协作通信和资源分配。推力3研究基于神经形态计算的支持DSA的HetNet的硬件设计。推力4号开发并评估了硬件-软件试验台。提出的从集中式基站控制方法到分散式方法的范式转变将彻底改变未来的无线网络设计，个人用户将在频谱接入中发挥更大的作用，并通过利用神经形态计算设备来彻底改变网络拓扑。本项目开发的软硬件协同设计方法可以很容易地应用于其他相关领域：计算机通信网络、网络安全、能源采集通信等。

During the last two decades, the use of Radio Frequency (RF) spectrum has increased tremendously due to the ever growing demand for wireless connectivity. The existing network technologies that support the current wireless data demand are expected to increase their capacity significantly in the next decade, calling for spectrum and energy efficient communication strategies. There are two popular approaches to efficiently utilize the RF spectrum: One is the cognitive radio networks which allow mobile users to share the spectrum that has been primarily allocated to other services such as television broadcasting, global position system (GPS), radar, weather forecasting, etc., provided that the mobile users impose limited interference to existing services. Another approach is to enhance the mobile broadband networks via expanded bandwidth, massive Multiple-Input Multiple-Output (MIMO) systems, and densified heterogeneous networks (HetNets). However, both approaches have limitations and have different impacts on spectrum efficiency and energy efficiency. In addition, current hardware platforms exhibit formidable challenges in supporting high computational complexity and low power consumption. This project introduces a novel network architecture and its application-specific hardware optimization using neuromorphic computing devices. The new wireless network architecture allows mobile users to perform spatio-temporal spectrum sensing and actively search for dynamic spectrum access (DSA) opportunities to enable short-range and local communications. Meanwhile, neuromorphic computing devices that mimic bio-neurological processes will be designed to tackle the high computational complexity of the new dynamic spectrum access approach with extremely low power consumption. In this way, we will be able to enable our nation's next-generation wireless communications and networking that are intelligent, spectrum-efficient, and energy-efficient in a dynamic spectrum environment. The developed concepts and technologies will also help achieve National Broadband Plan which targets at significant improvements in the efficiency of RF spectrum utilization. The project has an extensive education and outreach plan which includes designing new course components on energy-efficient communications, analog neuron circuits, and computational intelligence for wireless networks, joint training of graduate and undergraduate researchers between the two collaborative institutions, and outreach to telecommunication industry and underrepresented students through seminars and diversity programs.Short-range and local communications are extremely beneficial for spectrum and energy efficiency. The research objective of the project is to 1) design DSA-enabled HetNets to enable short-range/local spectrum access to improve the spectrum and energy efficiency, and 2) leverage neuromorphic computing architecture to efficiently solve the associated resource allocation problems with extremely high energy efficiency. To achieve the goal, the project is organized into four interconnected research thrusts. Thrust 1 focuses on spatio-temporal spectrum sensing with MIMO transceivers. Thrust 2 investigates cooperative communications and resource allocation for DSA-enabled HetNets. Thrust 3 studies neuromorphic computing based hardware design for DSA-enabled HetNets. Thrust 4 develops and evaluates the hardware-software test-bed. The proposed paradigm shift from centralized base-station-controlled approach to the decentralized approach will revolutionize the future wireless network design, where the individual users will play stronger role in spectrum access and drastically change the network topology by utilizing neuromorphic computing devices. The hardware-software co-design methodologies developed in this project can be readily applied to other related fields: computer communication networks, cyber security, and energy-harvesting communications, etc.