A team of three Warren B. Nelms Institute students recently won the NSF SpectrumX Data & Algorithm Competition (SpX-DAC) Overall Runner-Up Award. PhD students Xin Yao, Xi Wang, and Yanyu Hu, advised by Dr. Mingyue Ji, participated in the competition.
Spectrum is the range of radio frequencies used for wireless communication. Cellular networks, Wi-Fi, satellites, and IoT devices all use spectrum. Since all connected devices share this limited resource, good spectrum monitoring can help avoid interference and enable more reliable communication. It can support more efficient and connected wireless systems, contributing to better use of the available spectrum.
SpX-DAC aims to inspire students to turn raw spectrum data into intelligent detection and decision-making tools that can shape the future of spectrum sharing by giving them the opportunity to work on real spectrum field data. The competition had 37 team registrations from 25 leading institutions around the country, involving 83 students. The award was announced during the NSF SpectrumX Spring 2026 Center Meeting on June 1 and 2, 2026 at the University of Notre Dame.
The goal of the competition was to detect whether an active wireless transmission was present in a spectrum recording. The signals were similar to 5G signals in the lower 7 GHz band. The data were collected during NSF SpectrumX’s July 2025 VLA experiment. Teams were evaluated on detection accuracy, implementation efficiency, algorithmic novelty, and visualization quality.
“Our algorithm received a one-second raw signal recording. It then gave a simple answer: either a transmission was present or no transmission was present. We developed a hybrid method that combined signal processing and machine learning. Some signals could be detected using a simple spectrum-based rule. Other signals were more difficult, so we used an improved convolutional neural network. This allowed us to use a simple method when possible and a more flexible method for difficult cases,” said Hu.
Working with spectrum data is closely related to the research in Professor Ji’s lab, which focuses on wireless spectrum, signal processing, spectrum sharing, and machine learning. The team used the University of Florida’s HiPerGator computing system to design and train the algorithm they used.
“The competition gave us a chance to work with real spectrum data. It also helped us better understand the difficulty of detecting very weak signals. We learned that signal-processing knowledge and machine-learning methods can work well together,” said Hu.
The biggest challenge was the limited amount of training data– only 200 labeled training samples and 40 local test samples were given. The team needed to create a model that could learn from limited data and still generalize to new signal conditions. To improve the model, the team created additional weak-signal examples during training. They also combined several trained models instead of relying on only one model. This made the final method more stable.
The competition also gave the researchers new research ideas. In the future, they hope to pursue detecting signals close to the noise floor, handling differences between training data and real-world data, and developing efficient models for IoT and edge devices. Understanding spectrum is an important part of developing IoT technologies for more secure and connected communities.
