MobiCom 2025 Papers

Tear Film Break-Up Time (TBUT) is a critical clinical parameter in the management of dry eye disease (DED). However, traditional TBUT assessments rely on costly and time-consuming clinical procedures, while existing home-based solutions fail to provide precise TBUT values. In this work, we present Blinic, a contactless system leveraging commercial millimeter-wave (mmWave) radar to predict precise TBUT values and assess DED severity grades at home. Blinic incorporates detailed blink kinematics that are closely linked to TBUT. To address the challenge of predicting TBUT directly from radar data, we propose a teacher-student learning framework. The teacher model, trained on electronic health records (EHRs) including image-based diagnostic tests, transfers medical insights to the student model, which uses radar-captured blink dynamics. This knowledge transfer is further enhanced by a fine-tuned large language model, DryEye-LLM, which is based on clinical diagnostic reports and employs unsupervised domain adaptation to align EHRs with radar data. To ensure accurate blink motion capture, Blinic employs an antenna-coded MIMO mmWave radar design. Additionally, a query-based multitask learning module simultaneously predicts TBUT and DED severity grades, addressing potential conflicts in feature representation. Evaluated on 192 participants in collaboration with an eye clinic, Blinic demonstrates achieving a mean absolute error of 2.73 seconds for TBUT with an average accuracy of 90.54% for DED grading in real-world settings, providing a practical solution for home-based DED management.

Miniature, battery-free implants promise transformative bio-electronic therapies by enabling minimally invasive implantation procedures, reducing risk, and extending device lifetime. Among all wireless power and data transfer (WPDT) modalities, magnetoelectrics (ME) has emerged as a particularly promising solution for millimeter-scale implants, boasting lower tissue attenuation and higher power transfer efficiency over conventional inductive and ultrasonic methods. However, as an acoustic resonator, ME devices face an inherent tradeoff between Q-factor and bandwidth, limiting their ability to simultaneously achieve high-speed communication and efficient wireless power transfer (WPT). To fundamentally circumvent the challenge, this paper presents dual-resonance ME WPDT that exploits the unique multimode resonances of ME transducers to realize WPT and communication at distinct frequencies. Based on this dual-resonance principle, we demonstrate reconfigurable active and passive schemes for different biomedical applications, with a proof-of-concept system including a miniature implant and an external transceiver. The active scheme achieves 60 kbps at operational distances of 6 cm with 2.5 mW implant power, while the passive backscatter offers 20 kbps continuous streaming at 4 cm with negligible power, demonstrating the first reported non-interrupted ME WPDT system and more than twice the data rate of previous ME backscatter methods. Both schemes further support on-off keying (OOK) and binary phase shift keying (BPSK) modulations, providing additional flexibility to tailor communication needs between power efficiency and robustness. The complete prototype system was validated through comprehensive in-vitro experiments and in-vivo EMG streaming in a rodent model.

The rapid advancements in autonomous driving have introduced increasingly complex, real-time GPU-bound tasks critical for reliable vehicle operation. However, the proprietary nature of these autonomous systems and closed-source GPU drivers hinder fine-grained control over GPU executions, often resulting in missed deadlines that compromise vehicle performance. To address this, we present UrgenGo, a non-intrusive, urgency-aware GPU scheduling system that operates without access to application source code. UrgenGo implicitly prioritizes GPU executions through transparent kernel launch manipulation, employing task-level stream binding, delayed kernel launching, and batched kernel launch synchronization. We conducted extensive real-world evaluations in collaboration with a self-driving startup, developing 11 GPU-bound task chains for a realistic autonomous navigation application and implementing our system on a self-driving bus. Our results show a significant 61% reduction in the overall deadline miss ratio, compared to the state-of-the-art GPU scheduler that requires source code modifications.

Millimeter-wave (mmWave) has emerged as a promising solution for contact-free sensing due to its high resolution. Although promising, it faces several critical issues, including occlusion from the surrounding environment, unstable orientation-dependent sensing performance, and significant interference when multiple targets are in close proximity. These fundamental issues hinder the widespread adoption of mmWave sensing in the real world. In this paper, we propose SeRadar, the first systematic framework that leverages all useful secondary reflections to significantly enhance reliability and bring mmWave sensing one step closer to real-world adoption. Unlike primary reflections commonly used in wireless sensing, secondary reflections—typically much weaker due to being reflected multiple times—are generally ignored in existing literature. However, we observe that secondary reflections are common in various scenarios and carry valuable information about target movements, which could also contribute to sensing. To effectively utilize secondary reflections for sensing, SeRadar addresses several challenges associated with secondary reflections. Specifically, it boosts weak secondary reflections to improve their sensing capability, identifies useful ones from a large number of secondary reflections captured in the environment, and mitigates primary-secondary interference in multi-target scenarios. We evaluate the performance of SeRadar in various environments, including offices, apartments, and vehicle cabins. Extensive experiments demonstrate SeRadar can enhance accuracy and reliability in diverse sensing scenarios.

This paper presents the first nationwide deployment of human activity recognition (HAR) technology in the on-demand food delivery industry. We successfully adapted the state-of-the-art LIMU-BERT foundation model to the delivery platform. Spanning three phases over two years, the deployment progresses from a feasibility study in Yangzhou City to nationwide adoption involving 500,000 couriers across 367 cities in China. The adoption enables a series of downstream applications, and large-scale tests demonstrate its significant operational and economic benefits, showcasing the transformative potential of HAR technology in real-world applications. Additionally, we share lessons learned from this deployment and open-source our LIMU-BERT pretrained with millions of hours of sensor data.

As Frequency-Modulated Continuous Wave (FMCW) radar systems become increasingly prevalent across various sensing applications, detecting their presence is crucial to mitigate interference and address potential security risks. Existing methods for spectrum sensing or detecting unintended Radio Frequency (RF) transmissions rely on expensive specialized hardware because these radars typically operate in the GHz frequency range and utilize large bandwidths. To overcome these limitations, we present ChirpEye, a simple but effective tag design that is capable of identifying FMCW radar waveforms without requiring prior knowledge of radar parameters such as chirp slope, operating frequency, or bandwidth. In addition, ChirpEye can identify the direction of incident signal, hinting at the potential location of the radar. The key innovation of ChirpEye lies in its novel tag design, which uses multiple antennas and delay lines to process GHz-level radar signals with only kHz sampling rates. The tag structure generates unique baseband frequencies that are proportional to FMCW waveform parameters. We also propose a new super-resolution algorithm, called Spectra-MUSIC, which can accurately estimate these beat frequencies from noisy data. Our extensive evaluations demonstrate that ChirpEye achieves 99% accuracy in detecting FMCW radars at distances up to 15 meters with less than 5% median error in estimating the radar chirp slope and less than 15 degrees median error in estimating the direction of the radar.

The evolution of 5G and future-G technologies has led to the development of an open, distributed, and multi-vendor architecture known as Open Radio Access Network (O-RAN). While O-RAN offers greater flexibility and stimulates innovation, it also introduces potential issues such as bugs, inconsistencies, and vulnerabilities. In this paper, we present a novel model-based fuzzing system, ARCANE, to address these challenges. Unlike existing fuzzing techniques that struggle with the complexity and dynamism of O-RAN, ARCANE distinguishes itself by employing a smart, model-based approach. It first analyzes O-RAN specifications using a Large Language Model (LLM) and incorporates a unique method that integrates passive model learning to refine the model. With this refined model, ARCANE designs an O-RAN-aware, model-based fuzzing scheme that maintains high efficiency by adhering to O-RAN's specific semantics and syntax. We implement ARCANE and evaluate it on the OAI5G. It finds 9 root bugs from three categories, all having serious security implications. ARCANE has shown superior efficiency and coverage compared to state-of-the-arts.

Taste perception plays an essential role in promoting human health and maintaining nutritional balance. Current taste perception techniques usually require expensive equipment and delicate storage conditions, which restrict their use primarily to laboratory settings. In this paper, we show that terahertz (THz) signals generate unique fingerprint spectra when interacting with different taste molecules in aqueous solutions. Building on this finding, we propose Molecular-level Taste Sensing (MoleSen), a contact-free gustatory sensing method aimed at achieving wireless human-like perception. Specifically, MoleSen emits terahertz signals towards the aqueous solution, captures the reflected signals, and then determines the type and concentration of the tastes by analyzing the unique fingerprint spectra of the reflected signals influenced by the taste molecules. In MoleSen, we design a bio-inspired deep learning model (DTB, Digital Taste Bud) to identify the subtle taste features diluted by water molecules. Additionally, we incorporate domain adaptive learning to address the issue of feature distribution shifts when multiple tastes are mixed. Through extensive experiments involving over 247,000 samples, we demonstrate that MoleSen can accurately differentiate the five basic tastes—sour, bitter, salty, sweet, and umami—with an accuracy of 98.5% for taste type determination and 96.9% for concentration detection. Moreover, MoleSen outperforms the human's taste sensitivity and achieves a highly accurate perception even for mixed tastes.

Traditional 3D content representations include dense point clouds that consume large amounts of data and hence network bandwidth, while newer representations such as neural radiance fields suffer from poor frame rates due to their nonstandard volumetric rendering pipeline. 3D Gaussian splats (3DGS) can be seen as a generalization of point clouds that meet the best of both worlds, with high visual quality and efficient rendering for real-time frame rates. However, delivering 3DGS scenes from a hosting server to client devices is still challenging due to high network data consumption (e.g., 1.5 GB for a single scene). The goal of this work is to create an efficient 3D content delivery framework that allows users to view high quality 3D scenes with 3DGS as the underlying data representation. The main contributions of the paper are: (1) Creating new layered 3DGS scenes for efficient delivery, (2) Scheduling algorithms to choose what splats to download at what time, and (3) Trace-driven experiments from users wearing virtual reality headsets to evaluate the visual quality and latency. Our system for Layered 3D Gaussian Splats delivery (L3GS) demonstrates high visual quality, achieving 16.9% higher average SSIM compared to baselines, and also works with other compressed 3DGS representations. The code is available at https://github.com/mavenslab/layered_3d_gaussian_splats.

Bluetooth-based location services have experienced significant growth over the past decades. RSSI-based techniques using beacons only provide meters-level accuracy. Angular-based approaches rely on customized antenna arrays, introducing high costs and limited usability. In 2020, Bluetooth Special Interest Group (Bluetooth SIG) released version 5.1, integrating Angle of Arrival (AoA) estimation to enable direction finding capabilities, which has the potential to improve localization across various fields, including logistics and industry. However, more than 4.1 billion devices (68% of the total) still do not support the direction finding feature. To address this issue and ensure backward compatibility, we proposed Bridge, a solution that leverages an additional trigger node (referred to as Trigger) to make the direction finding feature compatible with all Bluetooth devices without requiring modifications to existing hardware or firmware. The Trigger mimics communication behaviors with both locators and targets simultaneously by sending a nesting packet. Subsequently, processes and algorithms are delicately designed to estimate AoA. Bridge also supports large-scale deployment through dynamic packet flow switching, enabling it to handle concurrent targets and manage handover with a consistent operation pattern. We implemented and evaluated Bridge in real-world scenarios. The system achieved an average localization error of 33.4cm while extending the direction-finding feature to 10 target devices of different Bluetooth versions, indicating the effectiveness of Bridge.

The Wi-Fi-enabled ultra-low power communication system exhibits high asymmetry between uplink and downlink speeds. The uplink can reach up to 1 Mbps, while the downlink throughput is around 100 Kbps. In this paper, we present Wook, a novel high throughput downlink system to empower Commercial Off-The-Shelf (COTS) Wi-Fi devices to transmit high-speed OOK messages. The key innovation underpinning Wook is its ability to achieve sub-symbol level modulation, allowing a single OFDM symbol to carry multiple OOK bits. This is done by profoundly exploring the Wi-Fi PHY layer and identifying optimal input payload to achieve fine-grained Wi-Fi waveform manipulation. We fabricate a PCB prototype and employ the COTS Wi-Fi router to implement the entire system. Experimental results show that with a simulated IC power consumption 76.6μW, Wook achieves a data rate of up to 1.1 Mbps, an 8.9X improvement over state-of-the-art systems. Moreover, even at a communication distance of 95 m, Wook maintains a throughput of 82.9 Kbps.

This paper presents CartoRadar, a novel RF-based SLAM system that delivers high-fidelity 3D mapping with centimeter-level accuracy. CartoRadar builds on top of the advancements in learning-based RF imaging. However, learning-based systems often exhibit variation in prediction accuracy during inference. To address this challenge and enable robust RF sensing, CartoRadar introduces a novel, training-free uncertainty quantification method tailored to RF signals. Additionally, CartoRadar features an efficient SLAM algorithm that incorporates this uncertainty into the mapping process. We deploy CartoRadar on a mobile robot and conduct extensive evaluations across 14 floors in 5 buildings. Results show that CartoRadar achieves a trajectory error of 14.1 cm, outperforming camera-based baselines by 72.1%. For mapping, CartoRadar achieves an accuracy of 7.4 cm and a completion of 8.1 cm, improving over vision methods by 46.2% and 67.6%, respectively. Code, datasets, and demo videos are available on our website.

Bluetooth audio, as a common application in our daily life, faces a major challenge due to its limited transmission range in meeting users' demands. Traditional solutions, such as using high-power Bluetooth transmitters, require hardware upgrades that are neither cost-effective nor energy-efficient. This work introduces WiLE-Audio, a novel approach that extends Low Energy Audio (LE Audio) coverage through physical-layer cross-technology communication (CTC) from WiFi to Bluetooth Low Energy (BLE). We first present a novel symbol mapping technique from WiFi DQPSK to BLE GFSK symbols, which enables all-channel and reliable CTC to support Bluetooth channel hopping. Then, to implement CTC to commodity WiFi Network Interface Card (NIC), we present a real-time reverse scrambling method that dynamically calculates the payload of WiFi packets at the NIC driver. Finally, to align with the strict time window requirements of the BLE receiver, we design a precise timing strategy and a priority scheduling mechanism at the WiFi transmitter, effectively mitigating timing offsets due to Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) and queue management. These systematic innovations allow WiLE-Audio to be easily implemented into existing commercial WiFi and BLE devices with only a simple software upgrade on the WiFi side. Furthermore, using existing WiFi infrastructures, WiLE-Audio enables the relaying of Bluetooth Low Energy Audio (LE-Audio) and the roaming of BLE receiver at any WiFi-covered location. We implement WiLE-Audio on commercial WiFi and BLE devices, and conduct extensive evaluations across various scenarios. Experimental results demonstrate that WiLE-Audio extends the transmission distance of LE Audio by 2× in single-hop mode and at least 2.6× in two-hop roaming mode. This work provides a cost-effective and scalable solution for enhancing Bluetooth audio coverage, providing a promising prospect for whole-house or even whole-building LE Audio listening.

Infrastructure-assisted autonomous driving has emerged as a pivotal technology to overcome the challenges posed by occlusions and limited fields of view for individual vehicles. Vehicles can fuse perception information from the infrastructure with their own in real-time, thereby enhancing their perception ability. However, our real-world experiments demonstrate that such an approach could introduce artifacts such as ghost objects, resulting in unsafe and unreliable planning outcomes. Besides, the system integration complexity and communication overhead are typically considerable, posing challenges to practical deployment. Therefore, we propose VI-Planning, an innovative infrastructure-assisted system that effectively optimizes autonomous vehicle planning in real time. The core idea of VI-Planning is to leverage the scene-level future occupancy grid maps constructed by the infrastructure as future drivable area references to directly optimize planning outcomes of autonomous vehicles. Since VI-Planning operates only at the autonomous vehicle's final output stage, without modifying the vehicle's underlying system architecture, it can be plug-and-play for most autonomous driving systems, whether they are modular or end-to-end architectures. Moreover, VI-Planning employs a novel bitwise encoding mechanism to efficiently compress these maps, enabling practical transmission. We implement VI-Planning end-to-end on a real-world testbed. The results of closed-loop and open-loop experiments indicate that VI-Planning can achieve real-time planning optimization (62.54 ms on average) and 817 × data transmission efficiency compared to the state-of-the-art baseline. A video demo of VI-Planning on our real-world testbed is available at: https://youtu.be/DXl5BhDEvFQ.

As a crucial tool for assessing health, spirometry provides valuable insights into pulmonary functions. Recent advancements have enabled more convenient measurements by shifting spirometry solutions from cumbersome clinical devices to portable devices. However, the forced maneuvers and burdensome procedures, which necessitate repeated maximal forced breathing, often lead to dizziness and discomfort, rendering them unsuitable for vulnerable populations. In this paper, we present ESPIRO (Earphone-enabled Speech sPIROmetry) system to furnish user-friendly pulmonary function monitoring for diverse populations. Basically, ESPIRO records normal speech using microphone-embedded earphones and characterizes pulmonary function-related glottal flow during speech production. ESPIRO advances existing spirometry solutions in i) leveraging phonetics to associate pulmonary function with glottal flow in normal speech, thereby eliminating the need for forced breathing; ii) identifying effective speech features according to physiological basis, ensuring reliable spirometry measurements; and iii) effectively addressing ambient noise, making it suitable for various real-world settings. Extensive experiments with 38 subjects on 18 commodity earphones confirm that ESPIRO accurately estimates pulmonary function indices in practice.