NSF Org:	CNS Division Of Computer and Network Systems
Recipient:	RECTOR & VISITORS OF THE UNIVERSITY OF VIRGINIA
Initial Amendment Date:	August 26, 2024
Latest Amendment Date:	August 26, 2024
Award Number:	2403124
Award Instrument:	Continuing Grant
Program Manager:	Alhussein Abouzeid aabouzei@nsf.gov  (703)292-0000 CNS  Division Of Computer and Network Systems CSE  Direct For Computer & Info Scie & Enginr
Start Date:	October 1, 2024
End Date:	September 30, 2028 (Estimated)
Total Intended Award Amount:	$900,000.00
Total Awarded Amount to Date:	$600,000.00
Funds Obligated to Date:	FY 2024 = $600,000.00
History of Investigator:	Kun Qian (Principal Investigator) xts4uh@virginia.edu Xinyu Zhang (Co-Principal Investigator)
Recipient Sponsored Research Office:	University of Virginia Main Campus 1001 EMMET ST N CHARLOTTESVILLE VA  US  22903-4833 (434)924-4270
Sponsor Congressional District:	05
Primary Place of Performance:	University of Virginia Main Campus 1001 EMMET ST N CHARLOTTESVILLE VA  US  22903-4833
Primary Place of Performance Congressional District:	05
Unique Entity Identifier (UEI):	JJG6HU8PA4S5
Parent UEI:
NSF Program(s):	Networking Technology and Syst
Primary Program Source:	01002425DB NSF RESEARCH & RELATED ACTIVIT 01002728DB NSF RESEARCH & RELATED ACTIVIT
Program Reference Code(s):	7924
Program Element Code(s):	736300
Award Agency Code:	4900
Fund Agency Code:	4900
Assistance Listing Number(s):	47.070

ABSTRACT

The forthcoming 6G cellular network infrastructure is anticipated to reclaim part of the 7-24 GHz upper mid-band (FR3) spectrum to strike a balance between the high capacity of millimeter-wave and wide coverage of the sub-6 GHz bands. However, the prohibitive channel probing overhead and spectrum sharing with various incumbent upper mid-band users (e.g., radar and satellite transceivers) remain major challenges. The objective of this project is to harness innovations in generative AI to overcome these limitations of 6G radios. The PI team explores the design of a generative channel modeling framework to tame the channel probing overhead and enable efficient interference management and spectrum coexistence between directional transmitters in the 6G upper mid-band. The project outcome will likely inform the beyond 5G/6G standardization and adoption through collaboration with industry partners. The project impact is further extended by creating a new curriculum focusing on ?generative AI for 5G and beyond,? engaging underrepresented researchers, and disseminating open-source experimental hardware and software for 6G FR3 experimentation.

The proposed research innovates a generative channel cartography framework that can predict the spatial channel distribution (i.e., length and angles of dominant propagation paths) across unseen locations. The framework enables unique 6G FR3 use cases, including efficient directional interference management and spectrum coexistence. Classical channel estimation methods heavily rely on real-time probing and largely ignore the channel's intrinsic structures originating from the sparsity of the multipath environment. This project advances generative models to encapsulate the channel prior, enabling directional beam/interference prediction conditioned on few-shot channel samples across locations. The resulting models can be broadly used to facilitate other types of networks, such as massive MIMO and mmWave networks facing the prohibitive channel probing overhead. This framework is applied to resolve the 6G FR3 networking problems through 3 thrusts: (1) Designing a cross-band generative diffusion model to predict FR3 interference maps using co-located FR1 measurements to maximize the spatial reuse and potentially reduce the interference management overhead. (2) Developing a generative NeRF which embeds prior knowledge of the physical environment into generative models, to predict the sparse multipath channels and efficiently estimate 6G-to-incumbent interference without deploying extensive spectrum sensors. (3) Developing a large generative model pre-trained on massive ray-tracing simulations to reconstruct the spatiotemporal channel dynamics with few-shot prompts. At runtime, it performs in-context learning from few-shot channel samples to enable dynamic beam and interference management for FR3 links.

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.

Please report errors in award information by writing to: awardsearch@nsf.gov.