EAGER: SC2: Intelligent spectrum collaboration via a dynamically reconfigurable radio architecture

目前有关射频频谱使用的工程实践和监管方法过于陈旧，无法满足对射频频谱不断增长的需求。解决这一频谱稀缺问题的一个有希望的新解决方案是为无线电网络配备人工智能，使它们能够学习和预测射频环境，并通过与其他无线电网络交互实现社交，从而更多地协作使用射频频谱。该项目将开发一个软件定义的无线电系统，该系统可以智能地感知和适应他人对无线电频谱的使用，并与其他无线电网络协作共享公共射频频谱。开发的系统将以其灵活性为特征，能够快速灵活地适应其他人如何使用射频频谱的变化。它的特点还在于如何使用机器学习技术来提取有关RF频谱使用方式的最相关信息，并根据这些信息调整通信策略。该项目将开发一个动态可重构的系统架构，以最有效地利用所有可用的计算资源，以支持所有无线电和机器学习功能。这种高度灵活的软件定义结构通过调整物理和介质访问控制层对频谱的使用，并通过精心设计的网络协议协调这种利用，充分利用了对射频环境的了解。开发了一种机器学习系统，用于识别通信场景演变的关键信息，并自主学习模型的状态。强化学习将用于根据系统状态生成适当的自适应通信策略。

Current engineering practices and regulatory approaches on the use of the radio frequency (RF) spectrum are too antiquated to meet the eversurging demand on the RF spectrum. A promising new solution to tackle this spectrum scarcity problem is to equip radio networks with artificial intelligence so that they can learn and predict the RF environment, as well as be social by interacting with other radio networks, leading to more collaborative use of the RF spectrum. This project will develop a software-defined radio system that can intelligently sense and adapt to others' use of the radio spectrum and collaborate with other radio networks in sharing the common RF spectrum. The developed system will be characterized by its flexibility to quickly and agile adaptability to changes in how others are using the RF spectrum. It will be also be characterized by how it uses machine-learning techniques to both extract the most relevant information about how the RF spectrum is being used and to adapt the communication strategies based on this information.A dynamically reconfigurable system architecture will be developed in this project to make most efficient use of all the available computational resources in order to support all radio and ML functionalities. This highly flexible software-defined structure takes advantage of the learned knowledge about the RF environment by adapting the physical and medium access control layers use of spectrum and coordinating this utilization through carefully designed network protocols. A machine learning system is developed to identify the key information about the evolution of the communication scenario, and autonomously learn the state of the model. Reinforcement learning will be used to generate appropriate adaptive communication strategies based on the system state.