EAGER: SC2: Load Prediction and Collision Coordination for Collaboration Channel

点击翻译按钮获取中文摘要

The demand for wireless bandwidth has grown exponentially in the past decade, motivating the need for novel spectrum access techniques. In this project, a hierarchical testing and implementation approach is used to generate and evaluate innovative spectral access techniques. To do so, the team is equipped with a broad range of backgrounds from FPGA development and machine learning to 3GPP standardization and extensive in-field experimentation. There are four particular areas of focus for the team: (i) network management including control structure and topology discovery, (ii) network discovery including modulation recognition and network recognition, (iii) spectrum access including channel selection, access mechanism design, and decision metrics, and (iv) link design including waveform selection, channel coding, and channel estimation. A central concept hinges on the role of the collaboration channel, on which the role of periodic and uniform levels of information exchange is studied related to channel availability across networks from no, partial, and full knowledge scenarios. Using machine learning and on-the-fly training with observations of network decisions and resulting performance, the focus then shifts to designing for robustness with respect to greater levels of latency and heterogeneity that result from non-standard protocols and algorithms across a diverse set of radios.The scope of the project includes the following three aspects. First, an agile research approach is employed using a hierarchy of implementation complexity to evaluate multiple design ideas, progressing through simulation tools, software-defined platforms, and FPGA hardware implementations and weighting the time spent according to the measured success of each design idea. Second, the role of the collaboration channel is extensively studied and evaluated in terms of spectral efficiency and coordination across teams by attempting to predict spectral availability of all users across the relevant spectrum in a homogeneous network context and select the channel availability update rate and size to attempt to maximize performance of these two metrics. Third, these findings develop the basis for extension to networks with high levels of latency and heterogeneity across the network stack where flexibility of information exchange across the collaboration channel is far more critical, forcing on-the-fly training that are built into the nodes to observe and reinforce machine-based decision making.