Collaborative Research: Generalizable and Adaptive RF Signal Generative Models for Next-Generation Wireless Communication and Sensing

About This Grant

This project develops an innovative radio frequency (RF) signal propagation modeling framework to advance next-generation wireless technologies, including Wi-Fi 7 and sixth-generation cellular networks. These technologies enable critical applications in smart cities, precision agriculture, and smart healthcare by supporting efficient wireless communication and sensing tasks such as human activity recognition and environmental monitoring. A key challenge in data-driven wireless systems is the labor-intensive process of collecting large-scale RF signal datasets for training deep learning models. This project addresses that challenge by generating high-fidelity synthetic RF datasets using advanced propagation modeling. These synthetic datasets support improved network planning, resource allocation, and sensing accuracy, ultimately leading to more efficient and scalable wireless infrastructures. The outcomes of this research contribute to societal benefits such as economic development, cost-effective network deployment, and enhanced connectivity in dynamic and infrastructure-limited environments. The project also integrates educational activities at the University of California, Merced and the University of California, Los Angeles, incorporating wireless communications and generative artificial intelligence into undergraduate and graduate curricula. In addition, the research team trains graduate students and postdoctoral scholars to support workforce development in RF modeling and next-generation wireless systems. The research investigates the use of Neural Radiance Fields for RF signal propagation modeling, with the goal of synthesizing received signals at arbitrary transmitter and receiver positions in complex three-dimensional environments. The scientific problem centers on overcoming key limitations of existing models, including high data requirements, computational inefficiencies, and poor adaptability to dynamic scenes and spatial variations. To address these issues, the research team develops a scalable approach that combines Gaussian-distribution-based representations, Graph Neural Network-guided scene modeling, and accelerated neural ray tracing to reduce data needs, training duration, and inference latency. Temporal adaptability is introduced through the use of deformation fields that capture dynamic environmental changes, while spatial generalization allows for applications across varying receiver positions and new environments without extensive retraining. The approach integrates techniques such as multi-head deformation decoders, neural-driven ray tracing, and contextual scene embeddings. The resulting models are evaluated using both simulation and experimental testbeds at UC Merced and UCLA. Evaluation metrics include fidelity of signal reconstruction, computational efficiency, and task performance in key applications such as device localization, activity recognition, network design, and resource allocation for diverse wireless technologies. 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.