ISART™ 2026: Sketching a Spectrum Management Blueprint
Tuesday, August 11, 2026 | |
Opening Remarks: Conference Overview | Just as Manhattan’s vertical expansion was unlocked by technological innovation, ITS recognizes that society now faces a similar capacity limit in radio spectrum use, requiring a technological breakthrough in spectral efficiency. ISART 2026 aims to push the boundaries of spectrum coexistence by discussing research in advanced sensing networks, AI-driven geolocation, and new coexistence techniques, while rigorously validating models and metrics to ensure robust, actionable insights. Research will focus on understanding and mapping spectrum usage, maximizing efficiency, and developing tools that enable transparent, data-driven management. By tackling these generational challenges, ISART will inform policymakers and decisionmakers with evidence-based recommendations, guiding the next era of spectrum utilization and regulatory frameworks. The story arc is clear: from recognizing the limits, to innovating new solutions, to empowering leaders with the knowledge needed for transformative spectrum management. |
Panel: Defining the Data Problem for AI/ML in Spectrum | More and more resources are focused on building and applying AI/ML models in RF. Despite the tremendous acceleration towards adopting the technology, these models are dependent on the availability of substantial amounts of high-quality data. This panel explores this problem through a discussion with representatives from industry, academia, and government: what are the most promising sources of data, what data is missing, and what are the barriers to sharing? |
Technical Talk: Lessons from NASCTN SEA Data Collection | The National Advanced Spectrum and Communication Test Network’s Citizens Broadband Radio Service Sharing Ecosystem Assessment was a multiyear project to give CBRS stakeholders longitudinal insights into a first of its kind spectrum sharing band. Within the framework of the NASCTN CBRS SEA program, multiple sensors were located in the Norfolk, VA , San Diego, CA and Boulder, CO areas. These sensors recorded spectral activity from 3530–3710 MHz continuously for a period exceeding two years. The ability to accurately characterize the spectrum environment is often hindered by the massive volume of raw I/Q data generated by wideband sensing. We present a streamlined approach to signal characterization by transitioning from raw data analysis to the utilization of curated "data products." By extracting key statistical features and reducing the overall data footprint, we demonstrate that signal identification and classification can be achieved with a significant reduction in computational overhead without sacrificing model accuracy. We will discuss the statistical mapping between raw signals and these reduced data products, illustrating how this approach optimizes the training and deployment of simple classifiers. The result is a more scalable architecture capable of rapid signal identification and enhanced spectrum situational awareness. |
Technical Talk: Data Collection, Standardization, and Public Access | In this tech talk, representatives from MITRE will discuss their experiences in building a platform for sharing RF signature data. They will share their lessons learned as part of this project – including the challenges in sharing and publishing PB-scale big data, how the data was produced, data standardization, and how the data is being made available to the public. Comparable north star solutions will also be discussed, like Stanford ImageNet. |
Panel: Evaluation Frameworks for ML Versus Classical Solutions | The integration of Machine Learning (ML) into wireless communication systems promises to revolutionize physical layer design, network optimization, and radio frequency (RF) front-end processing. However, as ML transitions from theoretical research to standardization and deployment, engineering teams must weigh empirical, data-driven models against deterministic, physics-based classical algorithms. This panel explores rigorous evaluation frameworks required to quantify when and where ML provides a demonstrable advantage over classical solutions in core radio domains—such as channel estimation, digital pre-distortion (DPD), beamforming, and propagation modeling. Panelists will discuss critical engineering trade-offs, including the breakdown of conventional sufficient statistics in highly non-linear or unmodeled environments, the operational complexity shift between training and inference, and the systemic energy costs of compute-heavy AI models versus low-latency, deterministic hardware. |
Wednesday, August 12, 2026 | |
Fireside Chat: Autonomous Spectrum Management: Vision and Reality | This fireside chat invites NTIA and FCC leaders to examine how emerging technologies—especially AI and ML—are reshaping the traditional regulatory paradigm for spectrum management, including decision-making loops, automation, and interference resolution. The session explores how today’s processes, built for predictable and repeatable software behavior, must evolve to address dynamic, adaptive systems whose updates and decision patterns may be difficult to fully understand. The discussion will highlight real world implications, such as spectrum auctions conducted amid ongoing technological innovation, and the risks of building technical solutions that may later prove incompatible with policy or regularity constraints. |
Technical Talks: Intelligent Operations | These sessions bring together a range of perspectives on the evolution and future of cognitive radio and intelligent spectrum operations, with an emphasis on implications for real-world deployment. Across three complementary talks, presenters will explore historical shifts in wireless system design, the growing role of adaptability and intelligence in spectrum use, and the technological, regulatory, and market factors shaping autonomous operation. Discussion will connect conceptual foundations with practical challenges, including implementation, emerging standards, and opportunities for commercialization in dynamic, data-driven spectrum management. Collectively, the talks offer attendees a forward-looking view of the field and actionable insight into where innovation and industry engagement are likely to have the greatest impact. |
Data Sharing Workshop | This interactive workshop brings together MNOs, hardware vendors, researchers, and policymakers to confront data scarcity and data‑sharing barriers limiting ML for telecommunications. Participants begin with a “data reality check” poll, then move into field‑specific breakouts to identify unsolved ML problems, essential RF data needs, and minimum metadata and trust requirements. Each group integrates perspectives and works toward shared recommendations. The workshop concludes with whole‑group consensus building to outline practical steps for enabling trustworthy, cross‑sector data sharing to advance ML and AI for spectrum. |
Thursday, August 13, 2026 | |
Panel: Research and Requirements for Hardware Solutions | Achieving dynamic spectrum coordination and utilizing insights provided by AI/ML techniques within the increasingly congested and contested wireless spectrum requires RF hardware capable of realizing the operational agility necessary to quickly, accurately, and reliably reconfigure an RF system for the current environment and goals. Although significant progress has been made in recent years with the introduction of software defined radios, RF system-on-chip platforms, and improved beamforming and MIMO technologies, significant work remains to be done, including: increasing the agility of RF front-end components such as amplifiers, filters, and antennas, refining the real-time control mechanisms that link these components to the overall system, and synchronizing multiple systems and validating the overall performance and behavior of a complete network of adaptive systems. This panel discusses what hardware capabilities are available today that address the needs of reconfigurable RF systems and advanced AI/ML compute, what shortcomings existing solutions have, and what should be the focus for additional developments towards improved spectrum agility and adaptive and reconfigurable RF technologies. |
Technical Talk: Physics-Informed Machine Learning: Methods, Frameworks, and Applications in Computational Electromagnetics | Classical numerical solvers — finite difference, finite element, and method of moments — have long been the backbone of computational science, yet they remain computationally prohibitive for large-scale, multi-query, and real-time applications. Physics-informed machine learning (PIML) addresses this bottleneck by fundamentally rethinking how physical knowledge is incorporated into learning systems. Rather than treating governing equations as external validation, PIML embeds them as intrinsic structure — through physics-constrained loss functions, operator-theoretic learning, symmetry-preserving architectures, and hybrid solver-network coupling. This tight integration of data and physics enables models to generalize from limited observations, remain consistent with conservation laws, and extrapolate reliably beyond the training distribution. PIML has demonstrated transformative impact across computational fluid dynamics, acoustic and elastic wave propagation, heat and mass transfer, and shape optimization — establishing a unified framework for surrogate modeling of complex PDE-governed systems across science and engineering. The ultimate objective is high-accuracy Radar Cross Section (RCS) computation directly from full-wave field solutions — positioning PI-DeepONet as a real-time surrogate for FDTD and MoM, with broad implications for radar system design, target identification, and stealth engineering. |
Technical Talk: Emerging Hardware Architectures for AI-enabled Spectrum Systems | As spectrum environments become increasingly dynamic and autonomous, advances in artificial intelligence must be matched by corresponding innovations in the underlying RF hardware. Future spectrum systems will require architectures capable of supporting real-time sensing, edge AI/ML inference, adaptive waveform generation, low-latency decision making, and operation across increasingly congested and contested environments. This talk will examine emerging hardware architectures that enable these capabilities, focusing on advances in RF front ends, phased arrays, heterogeneous computing platforms, and tightly integrated hardware-software co-design. The discussion will highlight practical design considerations, current research directions, and the challenges associated with transitioning AI-enabled spectrum technologies from the laboratory to operational systems. |
Technical Talk: One Aperture, One Solution: Wideband Active Antenna Performance Across 600 MHz to 7 GHz | Battelle has developed an active antenna technology that uses a denser element configuration than the standard λ/2 spacing. In comparison to traditional approaches, this architecture allows for several benefits, including significantly wider bandwidth, improved noise figure and greater spectral efficiency while still being designed for manufacturing at scale. Battelle has demonstrated the ability for a single aperture to be used from 600 MHz to 7 GHz with multiple waveforms. The current version of the active antenna, which implements the O-RAN 7.2 split, includes 64 elements providing the ability to optimize 5G energy efficiency, coverage and capacity. Multiple antenna units can be combined to create larger arrays which are fundamental to 6G implementations. This presentation will discuss the test setups for characterizing the mid-band performance of the radio unit. |
Panel: AI-RAN in Practice: Testbeds, Hardware Enablement, and the Path to 6G | The development of AI-native Radio Access Networks (AI-RAN) has moved from theory to an active research and engineering challenge, one that demands hardware testbeds capable of emulating commercial-scale networks: gathering data, training models, running inference, and operating in real time on live traffic. The discussion should give the audience a grounded picture of how AI-RAN hardware is progressing, from GPU-accelerated baseband and neural receiver validation to the MLOps and digital twin infrastructure needed to manage AI models at scale. For anyone working on or following AI-RAN over the next few years, this panel offers the ground truth: what today's testbeds have proven, what remains out of reach, and where the hardware must go next. |
Keynote: Reflections on Three Eras of Wireless Technology: Efficiency, Structural Margins, and Network Resilience | Inspired by Dale Hatfield's previous ISART presentations in 2000, 2002, and 2004, this session considers earlier ideas that were technically or institutionally premature when first discussed decades ago but which may now be more realistic because of advances in computing, sensing, software-defined radios, databases, AI-enabled control, and machine-to-machine coordination. The discussion asks what new risks or governance challenges come with technological advancements and what role academia, government, and industry can play in helping the community understand both. The speaker's distinguished career informs and illustrates the past and future challenges of spectrum management. |