Collaborative Research: SWIFT-SAT: RFI Detection Across Six Orders of Magnitude in Intensity: A Unifying Framework with Weakly Supervised Machine Learning

随着无线电发射机数量的增加，卫星星座与地面天文学的共存是一个越来越大的挑战。对天文学家来说，一个极其重要的宇宙学信号是21厘米的“自旋翻转”跃迁，表明宇宙中存在中性氢。这个信号以1420mhz的频率发射，但由于宇宙红移，从非常遥远的星系接收到的信号频率较低。在人为产生的无线通信射频传输干扰存在的情况下，检测这种微弱信号可能很困难。该研究项目将使用机器学习算法来更好地检测和减轻这种干扰，这将使在早期宇宙中检测中性氢成为可能。本科生将参与该计划的各个方面，为他们提供频谱管理、空间态势感知和机器学习算法等关键问题的实践经验。来自卫星星座的无线电频率干扰（RFI）对射电天文学观测实验构成了严重威胁，这些实验试图探测到跨越宇宙时间的21厘米中性氢信号。这些高灵敏度的实验必须集成1000多个小时才能检测到红移21厘米的信号；在这些极端的灵敏度下，即使是非常微弱的射频信号也会成为一种重要的污染物。目前，没有一种单一的RFI检测技术可以有效地识别非常明亮和非常微弱的RFI（其信号强度可能相差多达六个数量级）。该研究团队将开发一种弱监督机器学习框架，该框架使用现有的RFI检测技术来创建一种自一致的标记策略，适用于所有事件，从明亮的发射机到立方体卫星地面信号的微弱反射。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。

The coexistence of satellite constellations with ground-based astronomy is a growing challenge with the increase in the number of radio transmitters. One cosmological signal of extreme importance to astronomers is the 21 cm “spin flip” transition, indicating the presence of neutral hydrogen in the cosmos. This signal is emitted at 1420 MHz but received at a range of lower frequencies from very distant galaxies due to cosmological redshift. Detecting this weak signal can be difficult in the presence of interference from human-generated radio-frequency transmissions for wireless communications. This research project will use machine learning algorithms to better detect and mitigate such interference, which will enable detection of neutral hydrogen in the very early universe. Undergraduate students will participate in all aspects of this program, providing them with hands-on experience in key issues of spectrum management, space situational awareness, and machine learning algorithms. Radio frequency interference (RFI) from satellite constellations poses a critical threat to observational radio astronomy experiments seeking to detect the 21 cm signal of neutral hydrogen across cosmic time. These highly sensitive experiments must integrate over a thousand hours to detect the redshifted 21 cm signal; even very faint RFI becomes a significant contaminant at these extreme sensitivities. Currently, no single RFI detection technique can effectively identify both very bright and very faint RFI (which can differ by as much as six orders of magnitude in signal strength). This research team will develop a weakly supervised machine learning framework that uses existing RFI detection techniques to create a self-consistent flagging strategy suitable for all events, from bright transmitters down to faint reflections of terrestrial signals off CubeSats.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.