Overview

Integrated sensing and communication (ISAC) is widely recognized as a key enabling technology for FutureG wireless systems, where communication infrastructure is expected to provide not only high-rate connectivity but also environmental awareness. While an active line of research focuses on designing new waveforms specifically optimized for ISAC, our work takes a different perspective: we argue that OFDM, the dominant waveform in modern cellular systems, is already highly competent for sensing in FutureG networks. By leveraging the time-frequency structure, wide bandwidth, and multi-antenna capability of OFDM-based communication signals, sensing functions such as ranging, localization, and environmental perception can be realized without fundamentally changing the underlying cellular waveform.

Our approach is based on the principle that communication remains the primary service, while sensing is a secondary function enabled by the same transmitted signals. Therefore, the sensing operation should not degrade, interrupt, or impose additional constraints on communication performance. Instead of introducing dedicated sensing waveforms or occupying new spectral resources, we reuse OFDM signals emitted for communication to extract sensing information from their reflections. This design avoids additional spectrum utilization and simplifies integration into FutureG systems from both technical and regulatory perspectives, since the sensing capability is built on top of existing communication transmissions rather than requiring new spectrum allocation or major changes to the cellular air interface.

OFDM-Based Monostatic ISAC on mmWave Band

Monostatic and bistatic ISAC are two basic architectures for integrated sensing and communication. In bistatic ISAC, the transmitter and receiver are spatially separated, enabling wider coverage and spatial diversity but introducing synchronization errors, geometric uncertainty, and more complex channel estimation. In contrast, monostatic ISAC co-locates the transmitter and receiver on the same device, simplifying synchronization and enabling more accurate range, Doppler, and micro-Doppler sensing. However, it requires full-duplex operation and effective self-interference cancellation.

Currently, no commercially available monostatic ISAC platform supports simultaneous communication and sensing within a single device. To address this gap, we built a COTS-based prototype using an AMD/Xilinx RFSoC 4x2 FPGA board and a Sivers 60 GHz mmWave transceiver with Tx/Rx phased-array antennas. The RFSoC implements an OFDM pipeline compatible with 5G and Wi-Fi, while the mmWave transceiver handles radio transmission and reception. The transmission, reception, and beam-steering pipelines are jointly calibrated and synchronized to support monostatic sensing.

• High-Fidelity 3D Scene Imaging with mmWave Communication Signals (CVPR’26)
  Robust 3D environmental perception is critical for applications such as autonomous driving and robotic navigation. However, optical sensors, including cameras and LiDAR, often degrade or fail under adverse conditions such as smoke, fog, and poor lighting. Although specialized radar systems can operate in these challenging environments, their reliance on bespoke hardware and licensed spectrum limits their scalability and cost-effectiveness. This paper introduces Rascene, an integrated sensing and communication (ISAC) framework that leverages ubiquitous mmWave OFDM communication signals for 3D scene imaging. To overcome the sparsity and multipath ambiguity of individual radio frames, Rascene performs multi-frame, spatially adaptive fusion with confidence-weighted forward projection, enabling the recovery of geometric consensus across arbitrary poses. Experimental results demonstrate that Rascene reconstructs 3D scenes with high precision, offering a new pathway toward low-cost, scalable, and robust 3D perception.

• RF-based 3D Human Pose Estimation via OFDM Monostatic ISAC
  Human pose estimation under adverse lighting and occlusion conditions remains challenging for camera- and LiDAR-based systems. Although mmWave radars provides robustness to illumination and occlusion, they rely on dedicated hardware and licensed spectrum, limiting scalability. In this paper, we present RadPose, a learning-based monostatic integrated sensing and communication (ISAC) framework that enables commodity mmWave communication devices to perform 3D human pose estimation. Unlike specialized radars, RadPose leverages monostatic full-duplex operation to extract channel impulse responses (CIRs) directly from standard OFDM communication signals, generating 3D radio point clouds without additional sensing hardware. To address the sparsity and noise of RF observations, RadPose employs a cross-frame transformer-based pose inference network with semantic-guided spatial attention and target-driven temporal attention for robust spatio-temporal aggregation. It further adopts a joint-aware inverse kinematics refinement module based on graph convolution to enforce skeletal topology, bone-length consistency, and anatomical plausibility. Extensive experiments demonstrate that RadPose achieves competitive accuracy compared to state-of-the-art mmWave radar-based approaches while maintaining robust performance in unseen and occluded scenarios.

• Facial Expression Reconstruction using 5G mmWave Signal
  5G mmWave signals can be used to estimate human facial expressions. The left (blue) image shows the facial expression reconstructed using a depth camera, which serves as the ground-truth label. The right (orange) image shows the facial expression reconstructed solely from 5G mmWave signals using deep learning models.

OFDM-Based Monostatic ISAC on Sub-6GHz Band

• Fine-grained Human Detection Using Single WiFi Device (MobiCom’24)
  Sub-6 GHz radio sensing offers several compelling advantages, including resilience to poor lighting conditions, privacy preservation, and the ability to penetrate walls. However, in indoor environments, the sub-6 GHz ISM spectrum is heavily occupied by WiFi devices, leaving limited available spectrum for dedicated sensing. In this paper, we introduce SiWiS, a new approach for integrating radio sensing capabilities into individual WiFi devices to support fine-grained human activity detection. SiWiS consists of two main components: (i) a new hardware component that can be easily installed on an off-the-shelf WiFi device, and (ii) a dual-branch deep neural network (DNN) optimized for concurrent human mask segmentation and pose estimation. We build a prototype of SiWiS and install it on a commercial WiFi router for evaluation. Extensive experimental results demonstrate that SiWiS achieves significant performance improvements over WiFi channel state information (CSI)-based sensing methods. More importantly, zero-shot experiments confirm that SiWiS can be directly transferred to unseen real-world environments.

Radio signals from a commercial Wi-Fi router are used to estimate the skeleton of a walking person using an AI model. The overlaid green skeleton is generated by a camera using an off-the-shelf computer vision algorithm and serves as the ground-truth label for training the AI model. The overlaid red skeleton represents the skeleton estimated solely from the Wi-Fi signals.

OFDM-Based Bistatic ISAC in 5G

• 5G for Human Respiration Detection (INFOCOM’26 ISAC-WKSP)
  Cellular networks offer a unique opportunity to enable device-free, wide-area health monitoring by exploiting the sensitivity of radio-frequency (RF) propagation to human physiological activities. We present the first experimental study of human sleep monitoring using realistic 5G signals collected from commercial cellular infrastructure. We investigate a practical scenario in which a smartphone is placed near a bed, while a 5G base station periodically configures uplink sounding reference signal (SRS) transmissions to obtain fine-grained channel state information (CSI). Leveraging uplink CSI measurements, we design a lightweight signal processing pipeline for respiration rate estimation and a CNN-based model for sleep body movement classification. Through extensive experiments conducted on an indoor private 5G network, our system achieves over 91.2% accuracy in respiration rate estimation and 85.5% accuracy in sleep movement classification.

• 5G for Human Movement Detection
  In 5G O-RAN systems, CSI provides a fine-grained description of the radio channel conditions between a user equipment (UE) and the base station. CSI can be obtained through the uplink sounding reference signals (SRS), which are periodically transmitted by the UE to allow the base station to measure channel characteristics such as amplitude, phase, and frequency response across multiple antennas and subcarriers. This rich channel feedback is essential for key 5G functionalities including beamforming, link adaptation, and interference management, enabling more efficient and reliable wireless communication.
  
  Beyond communication optimization, CSI offers unique potential for sensing and detection. Since wireless signals propagate through the surrounding environment and are affected by movement, orientation, and human activities, CSI inherently captures subtle changes in the physical world. This means that variations in uplink CSI patterns can be used to detect the presence of devices, monitor user mobility, and even infer fine-grained activities such as gestures, walking, or breathing. Leveraging CSI for device and user activity detection opens the door to a new class of applications that combine communication and sensing in an integrated 5G framework.
  
  The integration of AI amplifies this potential by enabling data-driven modeling of the highly complex and dynamic relationship between CSI measurements and real-world activities. It can automatically extract features from high-dimensional CSI data, learn robust patterns, and perform accurate classification or prediction without requiring handcrafted signal models. In practice, this can enable real-world applications such as secure device authentication, occupancy detection for smart buildings, non-intrusive health monitoring, and enhanced situational awareness in wireless networks. By combining CSI with AI, 5G systems are evolving from pure communication infrastructures into intelligent sensing platforms.
  
  The below video shows the measured channel state information (CSI) at a commercial base station from a smartphone when the user is standing and walking.

• 5G for Drone Detection and Classification
  When a drone communicates through a 5G cellular link, its uplink sounding reference signals expose fine-grained CSI at the base station. As the drone moves, changes in distance, angle, speed, and surrounding multipath produce distinctive temporal patterns across antennas, subcarriers, and time slots. By continuously analyzing these CSI measurements, the base station can infer the drone's trajectory and classify its flight behavior without requiring additional radar hardware or dedicated sensing spectrum. This capability enables cellular infrastructure to support low-cost aerial monitoring, situational awareness, and security applications while maintaining the original communication service.

OFDM-Based Bistatic ISAC in Wi-Fi

• Wi-Fi CSI-based Handwriting Detection (JIoT’22)
  Wireless Internet-of-Things (IoT) applications have penetrated nearly every aspect of society and are becoming increasingly important in smart homes, smart cities, and smart healthcare systems. However, many WiFi-based IoT devices, such as light switches, door/window open-alert sensors, and Google Home devices, lack input interfaces such as keypads or touchscreens due to constraints on physical size, power consumption, and manufacturing cost. This makes it inconvenient and burdensome for end users to authenticate these IoT devices for wireless Internet access. In this paper, we present AuthIoT, a learning-based authentication scheme for wireless IoT devices without input interfaces. The key component of AuthIoT is a channel state information (CSI)-based character classification algorithm implemented at a WiFi access point (AP), which recognizes a passcode from an IoT device when an end user holds the device and writes the passcode over the air. AuthIoT has two salient features: (i) it is transferable across environments, and (ii) it operates in realistic scenarios where the AP is equipped with a nonlinear antenna array. We built a prototype of AuthIoT and evaluated its performance on two testbeds: an Intel 5300 WiFi card with three linear antennas and a USRP N310 with four nonlinear, square-shaped antennas. Experimental results show that AuthIoT achieves recognition accuracies of 84% and 83% on the two testbeds, respectively.

Selected Publications

• Rascene: High-Fidelity 3D Scene Imaging with mmWave Communication Signals
  Kunzhe Song*, Geo Jie Zhou*, Xiaoming Liu, Huacheng Zeng
  IEEE Computer Vision and Pattern Recognition (CVPR), 2026. [Acceptance rate: 25%]
  [PDF] [Poster] [arXiv] [YouTube Video]

• Spectrum Shortage for Radio Sensing? Leveraging Ambient 5G Signals for Human Activity Detection
  Kunzhe Song*, M. Zingraff*, and Huacheng Zeng
  IEEE INFOCOM 2026. [Acceptance rate: 18.9%]
  [PDF] [arXiv]

• Integrating Health Sensing into Cellular Networks: Human Sleep Monitoring Using 5G Signals
  Ruxin Lin*, Peihao Yan*, Jie Lu*, Qijun Wang* and Huacheng Zeng
  IEEE INFOCOM WKSHPS: ISAC-FutureG 2026.
  [PDF] [arXiv]

• SiWiS: Fine-Grained Human Detection Using Single Wi-Fi Device
  Kunzhe Song*, Qijun Wang*, Shichen Zhang*, and Huacheng Zeng
  ACM MobiCom, 2024.
  [PDF]

• AuthIoT: A Transferable Wireless Authentication Scheme for IoT Devices Without Input Interface
  Shichen Zhang*, Pedram Kheirkhah Sangdeh*, Hossein Pirayesh*, Huacheng Zeng, Qiben Yan, and K. Zeng
  IEEE Internet of Things Journal, 2022.
  [PDF]