Lead PI: Brian L. Mark, George Mason University
Abstract:
Integrated Sensing and Communication (ISAC) is emerging as a foundational capability for future 6G wireless networks. ISAC enables wireless infrastructure to support both high-speed communication and environmental sensing, reducing the need for dedicated sensing systems and unlocking new applications such as airspace monitoring, smart transportation, and autonomous systems.
This project extends a learning-based resource management framework, called QoS-aware State-Augmented Learnable (QaSAL), to jointly manage communication and sensing functions within a radio unit. The goal is to dynamically allocate wireless resources in a way that maximizes overall system performance while ensuring reliable communication quality and managing operational risk. QaSAL was previously developed to handle resource sharing between 5G and Wi-Fi systems; here, it is adapted to the more complex demands of integrated sensing and communication.
A key focus of this work is incorporating risk awareness into the learning process and enabling coordinated control across the physical and medium-access layers of the wireless system. We will identify the most important system parameters for ISAC operation, with particular emphasis on unmanned aerial vehicle (UAV) detection as a representative sensing application, while simultaneously supporting 5G and future 6G communication services. The resulting framework aims to improve the efficiency and reliability of ISAC systems, providing strong performance while maintaining guarantees on service quality and risk.

Lead PI: Michael Wu, University of Arizona
Abstract:
Large language models (LLMs) are transforming many areas of artificial intelligence, but their rapidly growing size makes them difficult to deploy on resource-constrained edge devices. Cloud-based solutions can provide the necessary computing power, but they introduce challenges related to privacy, latency, and network reliability. This project explores a new approach for enabling distributed and collaborative LLMs across edge devices in next-generation wireless networks. Instead of running a large model in a centralized cloud, the proposed system divides a pretrained LLM into smaller modules that can be deployed across multiple edge devices. Each module can operate independently to support local applications, while the modules can also be dynamically reassembled to reconstruct a larger model for more complex tasks. The project aims to develop scalable architectures that balance performance, communication efficiency, and resource constraints in distributed AI systems. The results will contribute to enabling intelligent applications, such as digital twins, network automation, and multimodal reasoning directly on next-generation wireless edge infrastructures.

Lead PI: Jacek Kibilda, Virginia Tech
Abstract:
Wireless geofencing refers to the external ability to control the availability of wireless services within a well-defined perimeter, for example, an indoor conference room. Conventional methods for wireless geofencing require either control over the transmitter configuration, which requires direct access to the network, or the generation of jamming signals, which may not be permitted in licensed spectrum. In this project, we propose using passive reconfigurable intelligent surfaces (RISs) to control the coverage of wireless services within a well-defined perimeter. Passive RISs, which are composed of a large number of reflective elements that can be coherently configured to steer, shape or cancel incident wireless signals, can be used to recycle ambient cellular signals towards exploiting initial access and association protocols. In this project, we propose a RIS controller that is able to identify opportunities for altering coverage to end-user devices in well-defined perimeter, we verify its efficacy in a large-scale simulation setup, and propose an efficient RIS placement strategy and evaluation method for the controller’s coverage impact.

Lead PI: Marwan Krunz, University of Arizona
Abstract:
The main objective of this project is twofold: (1) Design and evaluate a radically innovative energy-efficient hardware/software framework for real-time inference using machine learning (ML) pipelines, and (2) customize this framework to enable real-time RF signal classification in next-generation wireless systems. By integrating processing elements within memory chips, the energy consumption of a DNN can be significantly reduced, and more computations can be done faster. The hardware-accelerated DNN designs provided by this project will facilitate rapid identification of wireless transmissions (e.g. radar, 5G, LTE, Wi-Fi, microwave, satellite, and others) in a shared-spectrum scenario, enabling better use of the spectrum and facilitating accurate detection of adversarial and rogue signals. Specifically, the project extends our previous efforts by considering multi-stage signal classifiers for protocol identification over the shared UNII and CBRS bands, as well as modulation classification for commercial protocols, including Wi-Fi 6, LTE, and 5G NR. Acceleration through novel hardware architectures as well as software are used to achieve real-time inference. For modulation classification, we consider lightweight transformer architecture, suitable for wireless datasets. Mixed-precision multi-format number representation are used to achieve fast and energy-efficient multi-stage classification.

Lead PI: Joao Santos, Virginia Tech
Abstract:
Mobile networks are transitioning toward AI-based management, where independent AI agents enhance the capabilities of the RAN. A key example of this evolution is O-RAN, where third-party xApps can independently control RAN parameters. However, this flexibility introduces new security risks, as xApps may have contrasting objectives that result in conflicting interactions, potentially leading to performance degradation and network instability. Despite recent efforts on conflict detection in O-RAN, there is still a limited understanding of the impact of conflicts on network performance and causal relationships between parameters and KPIs, limiting our ability to attribute performance degradations to specific control actions. In this project, we aim to address these limitations by leveraging a digital twin of an O-RAN network to quantify the impact of conflicts on the network performance and explore cause-and-effect dependencies between parameters and KPIs to assess how individual control actions contribute to system-wide network performance.

Lead PI: Jacek Kibilda, Virginia Tech
Abstract:
Integrated Sensing and Communication (ISAC) is a new function of mobile networks that is anticipated to bring new capabilities such as high accuracy positioning, tracking, environmental monitoring, activity recognition, and many others, that are anticipated to enable new vertical services and solutions. Mobile networks already implement a variety of reference signals that can jointly be used for sensing and connectivity functions. However, the reference signals afford limited flexibility due to their primary designation in service of the control plane and communication specific tasks. This poses a significant challenge in offering support for differentiated services with performance guarantees, potentially hampering the adoption of ISAC by verticals. In this project, we aim to develop a slicing framework to jointly support allocation of data and reference signal opportunities while catering to the performance needs of sensing and connectivity services.

Lead PI: Marwan Krunz, University of Arizona
Abstract:
In the era of digital-first branding, intelligent content discovery, and real-time search optimization,businesses must navigate a complex topic universe spanning brand marketing, SEO, and multi-domaincontent intelligence. The evolution of semantic search, algorithmic shifts, and dynamic user behaviordemands a move from static keyword-based strategies to adaptive, multi-modal, agentic intelligencesystems. To address this, we propose Agentic Orchestrations for Topic Intelligence Management System(AOTIMS)—a multi-agent, multi-modal AI framework that autonomously curates, analyzes, and optimizesbrand- and search-relevant topics. AOTIMS integrates Retrieval-Augmented Generation (RAG), vectordatabases, reinforcement learning, and multi-modal agentic reasoning, ensuring scalability, adaptability,and intelligent governance of topic ecosystems.

Lead PI: Marwan Krunz, University of Arizona
Abstract:
Effective beam management is essential for maintaining robust connectivity in dynamic environments. Traditional beam management strategies rely on exhaustive beam sweeping and measurement-based feedback, which introduce substantial overhead, latency, and computational complexity. Managing beams efficiently is even more challenging in multi-user and multi-cell scenarios due to increased interference, mobility, and dynamic channel conditions. This project proposes a machine learning (ML) driven approach for beam prediction and optimization, aiming to reduce the overhead associated with conventional beam training methods. Novel ML models for predicting the optimal beam for communication are proposed using a reduced beam search space. For this year, the project is focused on temporal-domain beam management, whereby past RSRP measurements for a subset of probed beams (donated as Set B beams) are used as input to a ML model that predicts the beam qualities for all possible beams. The prediction window covers one or multiple future time instances. A time-domain UNet (Td-UNet) model will be developed for ML prediction. The model extends the previously developed spatial-domain UNet (SD-UNet) by incorporating a conbolutional LSTM (convLSTM) network and a convolutional block attention module (CBAM). The project also leverages multi-armed bandits, a special type of reinforcement learning, to identify the optimal Set B beams for probing. Measurements and validation using the Keysight Channel Studio RaySim, along with the WirelessPro simulator will be conducted.

Lead PI: Marwan Krunz, University of Arizona
Abstract:
The design and optimization of next-generation (NextG) wireless systems requires comprehensive understanding of the radio frequency (RF) environment in which such systems are expected to operate. Not only will such systems be expected to coexist over and harmoniously share heavily congested mid-band spectrum (e.g., 1–7 GHz), but they will also likely operate over millimeter-wave (mmWave), including current 5G Bands (e.g., 24, 28, 37, 39, and 43 GHz) as well as sub-THz bands above 100 GHz. The RF propagation environment in these high-bands is not well characterized, despite a few sporadic efforts that focused on path-loss modeling. Network planning, spectrum-access coordination, and protocol adaptation can be greatly facilitated using RF maps, which provide a 2D/3D description of the received signal. However, obtaining RF maps through site surveys is quite labor-intensive and sometimes not feasible. To address this challenge, in this project, we explore generative adversarial network (GAN) based approaches for synthesizing RF maps at mmWave and sub-THz frequencies. Specifically, for this year, we are focusing on synthesizing coverage maps for outdoor environments and applying such maps for optimal placement and configuration of reflective intelligent surfaces (RIS) systems, a technology that is expected to be used in NextG systems. We are also designing generative models for synthesis of 3D RF maps (to be used in beam tracking for drone communications) and generative models for multi-transmitter environments under destructive interference.

Lead PI: Daniel J. Jakubisin, Virginia Tech
Abstract:
A driving use case for 6G wireless technology is integrated sensing and communications (ISAC). ISAC has the potential to unlock new network applications and capabilities, opening the door to rich analytics from the network, driving subscriber and revenue growth in the next generation. A key challenge of ISAC is to efficiently accomplish both tasks within limited spectrum resources. We draw from the iterative receiver concept to jointly perform demodulation, sensing, and detection tasks using probabilistic information—with the goal of high-resolution sensing. Through this work, we are developing receiver structures and algorithms which will advance 6G ISAC. In year 2 of the effort, we focus our exploration on design improvements that enhance computational and energy efficiency of the receiver algorithm and communications system.

Lead PI: Marwan Krunz, University of Arizona
Abstract:
5G NR enhances the security of legacy networks but also introduces new vulnerabilities, particularly in thebeamforming and beam tracking processes. In 5G NR, Channel State Information – Reference Signal (CSIRS) and Sounding Reference Signals (SRS) are used for downlink and uplink channel estimation,respectively. These signals play a crucial role in beamforming and modulation and coding scheme (MCS)selection. This project focuses on studying adversarial attacks on the digital MIMO beamforming andprecoding process. Specifically, we will investigate tampering and jamming attacks that distort CSIestimation used for MIMO precoding, as well as attacks on the reinforcement learning (RL) process usedfor beam tracking and MCS selection. Countermeasures and defense mechanisms will also be investigated.

Lead PI: Marwan Krunz, University of Arizona
Abstract:
The continuing demand for wireless capacity necessitates vacating additional spectrum for mobile services. Although spectrum is abundantly available at high frequencies (mmWave and sub-THz frequencies bands), operating over these frequencies still poses many technological and economic challenges. A more attractive alternative for NextG systems is to explore spectrum in the mid-band (1 GHz to 7.125 GHz) and FR3 (7.125 to 24. GHz). These bands, however, are already allocated to non-commercial wireless systems, include federal radar systems (stationary, mobile, and airborne). Spectrum sharing and coexistence between commercial wireless systems (e.g., 5G and beyond) and radar over mid-band has recently been a topic of significant interest. The first attempt at enabling such sharing focused on the CBRS band (3.55-3.7 GHz), where spectrum is being sharing according to a 3-tier hierarchy. More recent proposals that target spectrum sharing with radar include the Emerging Mid-Band Radar Spectrum Sharing (EMBRSS) for the 3.1–3.45 GHz band, AMBIT (3.45–3.55 GHz), and others. This project focuses on the security challenges associated with spectrum sharing of NextG systems and radar in mid-bands and FR3. In particular, facilitating coexistence with radar is typically done by utilizing a sensing approach, e.g., the Environmental Sensing Capability (ESC) in the case of CBRS. If a radar signal is detected, non-radar transmissions are prohibiting from operating over the shared channel. Because the ESC does not authenticate radar transmitters, a rogue device can attempt to mimic a radar signal, prompting the ESC to block the channel. Recently, some researchers utilized generative AI techniques to demonstrate the feasibility of producing “fake radar” signals, but the focus so far has been on automotive radar.

Lead PI: Kai Zeng, George Mason University
Abstract:
Integrated Sensing and Communications (ISAC) is considered one of the key innovations in NextG mobile networks. It is expected to enable a wide variety of vertical applications, ranging from drone detection for national security and critical infrastructure protection to vehicle traffic monitoring for smart transportation. Despite its significant socioeconomic benefits, ISAC functions introduce new attack surfaces to NextG infrastructure, thus calling for novel countermeasures. One noticeable security challenge brought by ISAC for NextG is the requirement for securing analog waveforms. Existing data layer security mechanisms adopted in mobile networks become insufficient to defend against analog/wave-domain attacks that can compromise confidentiality and privacy of sensing information while not being detectable or defendable at the bit level or data layer.
To address the new ISAC security challenges, this project proposes to design an OFDM resource grid (including demodulation reference signals and data) obfuscation mechanism that enables sensing privacy protection without sacrificing communication performance. This mechanism can be integrated with the existing cellular network frame structure at the baseband and transparent to data receivers.
The broader impacts of this project include: 1) Informing 3GPP and influencing ISAC standardization; 2) Measurement data sharing and public release of the data; 3) Experimental learning opportunities for undergraduate students through senior design projects and summer internships.

Lead PI: Jacek Kibilda, Virginia Tech
Co PI: William Headley, Virginia Tech
Abstract:
Waveform synthesis refers to the process of generating communication signals that satisfy specific input requirements in a given environmental context. Traditionally accomplished using analytical methods offers limited flexibility in terms of the possible design space. With the advancements in Generative Artificial Intelligence (GenAI), new methods can search from a significantly increased number of possible solutions. The challenge is that those approaches often lack appropriate conditioning mechanisms that would steer the generative process towards operator intended outcomes. In this project, we study the problem of generating communication signals that encode bit-level information for transmission over insecure wireless medium based on prompts that express the operator’s intent. In particular, we develop and validate novel methods for training and waveform synthesis based on dual diffusion process that translate input textual information to arbitrary waveform designs. We evaluate the proposed methods in both benign and adversarial interference scenarios.