How RF Computing Is Redefining the Boundaries of Intelligence: Five Talks to Watch at CNCC

When RF Signals Become Computational Objects

RF computing is evolving from mere signal processing into a foundational paradigm for intelligent systems. According to the CNCC 2026 forum "RF Computing: Exploring the Boundaries of Intelligence," the field is defined as "transforming RF signals within the RF space to accomplish corresponding information processing and conversion." But with the deep penetration of artificial intelligence, its scope is no longer confined to the physical layer.

Building on its first two editions—which focused on RF signals as information carriers, integrated sensing and intelligence, and spatial computing reconstruction—the forum now brings large models, embodied intelligence, and edge computing into the discussion. The topics span a remarkably wide range: from theoretical frameworks to information metamaterials, intelligent reflecting surfaces, and passive IoT, and further to the low-altitude economy, digital twins, and smart cities. This breadth is precisely what shows that RF computing is becoming a cross-cutting thread running through communication, sensing, and computation.

Intelligent Metasurfaces: Weaving Communication, Sensing, and Compute Power into One

Professor Cheng Qiang of Southeast University delivers the talk "New Advances in Integrated Sensing, Communication, and Computing with Intelligent Metasurfaces," which targets a clear engineering pain point: millimeter-wave blockage, spectrum resource competition, and the imbalance between sensing and communication performance. The intelligent metasurface approach uses programmable electromagnetic wave manipulation to reconstruct the wireless propagation environment, enabling high-speed communication, high-precision target sensing, and edge-collaborative compute scheduling to occur simultaneously.

The talk will review schemes including space-time coding metasurfaces, adaptive channel regulation, beamforming, RF computing, and AI-driven phase adaptive optimization, and attempt to integrate them into a "communication-sensing-computing" coordinated optimization framework. The track record of Cheng Qiang's team also shows this path is no empty talk: related research has been recognized with the State Natural Science Award (Second Class), the Top Ten Scientific Advances in China, and the IEEE Marconi Prize.

Reconfigurable Intelligent Surface (RIS) is a planar array composed of a large number of subwavelength elements, each of which can independently regulate the phase, amplitude, and even polarization of incident electromagnetic waves. By digitally encoding and controlling the state of each element, an RIS can act like a "mirror" to change the reflection direction and wavefront shape of wireless signals, thereby reconstructing the propagation environment without adding new transmitters. Unlike traditional relays, an RIS does not amplify signals and has extremely low power consumption, typically operating in a passive or semi-passive manner. Space-time coding metasurfaces further introduce the time dimension, switching element states at high speed according to specific sequences to generate controllable harmonic components in the frequency domain, integrating beam scanning and information modulation. These characteristics make intelligent metasurfaces a key enabling technology for integrated sensing, communication, and computing, and one of the current hot directions in 6G research.

Progress and Open Problems in RF Computing

Professor He Yuan of Tsinghua University delivers the talk "RF Computing: Progress and Reflections," which is more foundational. His definition emphasizes that RF computing treats RF signals "as computational objects," completing information transmission, processing, and conversion through transformations within the RF space. The value of this perspective is that it offers another solution path for several bottlenecks in the Internet of Things—data does not necessarily have to be moved into the digital domain before computation; computation can occur in the RF domain itself.

The word "Reflections" in the talk title is worth noting. For RF computing to become a mature discipline, several key scientific questions remain to be answered—for example, where the boundaries of programmability in RF-domain transformation lie, and how to trade off computational precision against energy consumption. The answers to these questions may determine whether RF computing remains a clever trick for specific scenarios or grows into a general-purpose technological foundation.

Embodied Intelligence Needs a Network That Integrates Sensing, Computing, and Communication

Professor Gong Wei of the University of Science and Technology of China delivers the talk "Networked Spatial Embodied Intelligence," shifting the discussion of embodied intelligence from single-agent capabilities to multi-agent network collaboration. His judgment is that intelligence is evolving from "standalone intelligence" to system intelligence that spans nodes, modalities, and spaces.

The problem with existing technical systems is the separate design of sensing, computing, and communication, which brings bottlenecks such as high connection energy consumption, fragile communication in complex environments, difficulty in unified representation of multimodal information, and excessive data movement and computation overhead. The path proposed in the talk includes highly reliable and low-power communication, multimodal spatiotemporal sensing, on-device intelligent computing, and heterogeneous integration—in particular, reconfiguring RF transceiver architectures to break through the energy-efficiency and anti-interference constraints of large-scale connectivity. The more critical conceptual shift is that sensing data must move from "presented for humans" to "understood and used for machine decision-making."

Embodied intelligence refers to the ability of an agent to interact with the environment through a physical body and to learn and complete tasks in a perception-action loop, emphasizing that intelligence cannot be separated from the coupling of body and environment. When embodied intelligence expands from a single robot to a multi-agent network, the system needs to exchange sensing data and coordinate decisions among nodes in real time, which imposes stringent requirements on communication latency, reliability, and energy efficiency. Heterogeneous integration in this context refers to integrating devices of different processes and functions (such as RF front-ends, sensors, and computing units) at the packaging or chip level to reduce data movement overhead. On-device intelligent computing moves part of the inference tasks down to terminals or edge nodes, reducing raw data uploads. Together, these technologies support the shift toward "sensing data for machine understanding"—data no longer serves human visualization but is fed directly to downstream models and decision modules.

Passive IoT Moves Toward Real Cellular Networks

The two talks on passive IoT address different aspects of 6G scenarios. Professor Yu Jihong of Beijing Institute of Technology focuses on "Backscatter Communication Technology for 6G Passive IoT," pointing out that ubiquitous IoT intelligent applications impose stringent challenges on power consumption, cost, and deployment convenience, while spectrum reuse, commercial compatibility, and long-distance transmission are bottlenecks that must be confronted.

The team of Gong Wei at Huazhong University of Science and Technology (note: the original material attributes Gong Wei to USTC; here it is handled as attributed in the source) introduced two specific efforts for 6G mobile passive IoT: the energy-efficient paging mechanism PScatter, which reduces continuous listening overhead by coordinating sleep and wake-up between terminals and receivers; and the load-independent synchronization system SyncSSB, which leverages 5G frame structure features unaffected by traffic load to adapt to different resource configurations and downlink signals. The significance of these efforts lies in advancing passive IoT from validation in specific scenarios toward long-term reliable operation in real cellular networks.

Backscatter communication is the core mechanism of passive IoT: the tag does not actively transmit RF signals but instead switches the impedance matching state of its own antenna to modulate the reflection of an existing RF carrier in the environment, "piggybacking" information on the reflected wave to the receiver. This approach eliminates high-power RF oscillators and power amplifiers, reducing tag power consumption to the microwatt level, allowing it to be powered by ambient or RF energy harvesting. PScatter targets the energy waste caused by passive terminals continuously monitoring the paging channel, reducing overhead by coordinating sleep/wake cycles between terminals and base stations; SyncSSB leverages the characteristic that synchronization signal blocks (SSB) in the 5G frame structure are unaffected by traffic load to achieve load-independent time synchronization, thereby adapting to different network configurations. The common goal of both is to enable passive tags to operate reliably for long periods in real cellular networks.

Signals of Cross-Disciplinary Innovation

The composition of the forum's speakers itself conveys a message: the State Key Laboratory of Millimeter Waves, the Institute of Trustworthy Networks and Systems in the School of Software, on-device intelligence and system security, satellite IoT and backscatter communication—RF computing brings together people from microwaves, IoT, mobile computing, and system security at one table.

In terms of technological maturity, intelligent metasurfaces and backscatter communication already have relatively clear engineering paths, while RF-domain intelligent computing and networked embodied intelligence remain in the stages of concept building and prototype validation. The distance between them is precisely the true meaning of the forum theme "Exploring the Boundaries of Intelligence": RF is no longer just a pipeline carrying information; it is becoming a physical medium that can be programmed, can compute, and can carry intelligence.

One-line Summary

RF computing is evolving from signal processing into a cross-disciplinary technology system that integrates intelligent metasurfaces, embodied intelligence, and passive IoT.

Paragraph Summary

Based on the CNCC 2026 forum "RF Computing: Exploring the Boundaries of Intelligence," this article reviews the expanding scope of RF computing and key technological advances. RF computing originally referred to transforming signals within the RF space to accomplish information processing; it now incorporates large models, embodied intelligence, and edge computing. Intelligent metasurfaces achieve communication-sensing-computing coordination through programmable electromagnetic manipulation, with a relatively clear engineering path; networked embodied intelligence requires reconfiguring RF transceiver architectures and promotes a shift in sensing data from human-oriented to machine-decision-oriented; PScatter and SyncSSB advance passive IoT toward real cellular networks. RF-domain intelligent computing and networked embodied intelligence remain early-stage, with lower engineering maturity than intelligent metasurfaces and backscatter communication.