Reliable failure detection and causal reasoning are critical in robotic manipulation, as their absence risks robot damage and endangers human safety.Although recent Vision–Language Models (VLMs) are employed to attempt failure detection and causality reasoning, they typically make retrospective assessment only after task completion, and their reasoning accuracy is often limited.To address these issues, we introduce RoboFailRing, which enables timely failure detection during task execution and enhances the reasoning accuracy of VLMs.It achieves rapid failure detection by retrieving a pre-constructed failure memory and returning a similarity-based decision.In addition, by providing grounded failure report to VLMs, it improves the accuracy of their reasoning about the failure causes and repair strategies.We evaluate RoboFailRing on two large-scale simulated datasets comprising over 6,000 failure trajectories and covering 81 distinct manipulation tasks.The results show that the average success rate of out-of-distribution failure detection reaches 80%, while the mean detection time is cut to roughly 50% of the baseline.Moreover, evaluations on real-world systems show an average 35% gain in VLM failure-reasoning accuracy.We make our code publicly available at: https://github.com/DynamicPoet/RoboFailRing.

As the gap between network and CPU speeds rapidly increases, the CPU-centric network stack proves inadequate due to excessive CPU and memory overheads. Though hardware-offloaded network stacks alleviate these issues, they suffer from limited flexibility in both control and data planes. It seems promising to offload network stacks to Smart-NICs to provide high flexibility. However, naive offloading leads to low throughput due to the inherent architectural limitations of widespread off-path SmartNICs. Even simple operations on staged network traffic would overwhelm the limited SmartNIC memory bandwidth. To this end, we design SmartNS, a SmartNIC-centric network stack with software transport programmability and line-rate packet processing capabilities. To tackle the limitations of SmartNIC-induced challenges, we propose a header-only offloading TX path and an unlimited-working-set in-cache processing RX path to minimize memory traffic to fit the wimpy SmartNIC memory bandwidth. To fully utilize the SmartNIC computing resources, we propose a programmable offloading engine to enable cloud providers to offload customized tasks along with the network stack processing. We prototype SmartNS using the widespread Nvidia BlueField-3 SmartNIC, and implement RoCEv2 and Solar transport protocols by leveraging SmartNS's software programmability. SmartNS achieves 2.2× higher throughput than the microkernel-based baseline in block storage disaggregation and 1.3× higher throughput than the hardware-offloaded baseline in KVCache transfer.

Eavesdropping poses a critical threat to the confidentiality and integrity of voice communications. In recent years, techniques have advanced beyond traditional microphone-based methods toward more intelligent approaches, such as leveraging millimeter-wave sensing to detect the subtle vibrations induced by speakers and reconstruct voice information without direct audio capture. Despite their technical feasibility, these methods remain constrained by limited working ranges—typically only several meters—rendering them impractical for real-world stealthy eavesdropping. In this work, we propose MoiréEar, the first long-range passive eavesdropping system based on moiré patterns. The key idea is to exploit the amplification capability of moiré patterns, which amplify the minute vibrations induced by acoustic signals by hundreds of times, enabling long-range eavesdropping. To make the proposed method even more practical and stealthy, we develop new theoretical foundations that relax the strict requirements for generating moiré patterns. Specifically, our approach enables the use of irregular stripe structures (e.g., commonly seen barcodes) instead of standard moiré gratings to generate moiré patterns. We implement our design using a low-cost photodiode instead of cameras, achieving real-time eavesdropping with lightweight signal processing. Comprehensive experiments show that the system can extract intelligible audio at a distance of up to 90 m, outperforming the state of the art by an order of magnitude in range. We believe this new eavesdropping modality can inspire a wide range of IoT applications.

Millimeter-wave (mmWave) technology offers significant potential for high-bandwidth, low-latency communication industrial applications. However, mmWave faces several challenges, such as limited range, susceptibility to blockage, and slow beam alignment. For this, mmWave access points (APs) are often deployed at high densities, which leads to increased costs. In this paper, we introduce mmProjector, the first cost-effective reflector, which is non-reconfigurable with a static structure but can serve mobile industrial robots. The key idea is to reshape the reflected waves only along the robots’ movement trajectories using mmProjector and make the phase of reflected wave constructively added along the trajectories, improving reflection gain with limited incident waves. To achieve this, we develop an analytical model grounded in electromagnetic theory and propose sub-optimal algorithms for the deployment and shape design of reflectors. We prototype mmProjector and deploy it in a real-world airplane assembly factory. Experiments demonstrate that mmProjector operates effectively at 60 GHz, delivering up to 1.8 Gbps of bandwidth in non-line-of-sight (NLOS) scenarios, while achieving up to 75% reduction in costs.

Deep joint source-channel coding (DeepJSCC) has emerged as a promising paradigm for efficient and robust information transmission. However, its intrinsic characteristics also pose new security challenges, notably an increased vulnerability to eavesdropping attacks. Existing studies on defending against eavesdropping attacks in DeepJSCC, while demonstrating certain effectiveness, often incur considerable computational overhead or introduce performance trade-offs that may adversely affect legitimate users. In this paper, we present DeepGuard, to the best of our knowledge, the first physical-layer defense framework for DeepJSCC against eavesdropping attacks, validated through over-the-air experiments using software-defined radios (SDRs). Considering that existing eavesdropping attacks against DeepJSCC are limited to simulation under ideal channels, we take a step further by identifying and implementing four representative types of attacks under various configurations in orthogonal frequency-division multiplexing systems. These attacks are evaluated over-the-air under diverse scenarios, allowing us to comprehensively characterize the real-world threat landscape. To mitigate these threats, DeepGuard introduces a novel preamble perturbation mechanism that modifies the preamble shared only between legitimate transceivers. To realize it, we first conduct a theoretical analysis of the perturbation’s impact on the signals intercepted by the eavesdropper. Building upon this, we develop an end-to-end perturbation optimization algorithm that significantly degrades eavesdropping performance while preserving reliable communication for legitimate users. We prototype DeepGuard using SDRs and conduct extensive over-the-air experiments in practical scenarios. Extensive experiments demonstrate that DeepGuard effectively mitigates eavesdropping threats while preserving reliable communication for legitimate users. In particular, DeepGuard can reduce the eavesdropper’s reconstruction performance by as much as 29 dB in PSNR and decrease classification accuracy by up to 91% compared with the performance achieved by the legitimate user.

Different from traditional object detectors such as YOLO, Detection Transformers (DETR) have reshaped the landscape of object detection by replacing heuristic-driven components like Non-Maximal Suppression with a fully end-to-end framework based on one-to-one Hungarian matching. While the majority of research has focused on improving the slow training convergence of DETR, this work investigates their security from an adversarial perspective. We unveil a critical vulnerability stemming directly from DETR’s core design: the deterministic one-to-one mapping between object queries and ground-truth objects can be exploited. This allows an adversary to craft perturbations that selectively manipulate specific target objects – causing them to vanish or be misclassified – while preserving the detection integrity of all other objects in the scene. Our initial analysis reveals that conventional gradient-based attacks are ill-suited for this task, as they induce unintended interference on non-target instances, a phenomenon we term as the “spillover effect”. To overcome this, we re-formulate the attack optimization by incorporating a novel penalty term that explicitly decouples the adversarial influence on target and non-target objects. Furthermore, we provide theoretical analysis to derive perturbation bounds under which the optimal matching assignments remain invariant, offering deeper insights into the model’s stability. Extensive experiments on standard benchmarks demonstrate that our proposed attack significantly improves the success rate and convergence speed while inducing far fewer feature-level artifacts, making the attack both more effective and stealthier.

Wind speed monitoring is crucial for environmental management and forecasting. However, current solutions often struggle with high power consumption, especially at the end device, which typically has a sensor and wireless radios with limited battery capacity. To this end, we present WindScatter, an ultra-low-power, long-range, and large-scale wind speed monitoring system. WindScatter adopts the Integrated Sensing and Communication (ISAC) paradigm to enable low-power operation. It reuses the sensed data for communication by leveraging a TMR (Tunnel Magneto-Resistance) switch sensor to measure the wind speed information and control the backscatter communication simultaneously, thus avoiding the need for analog-to-digital conversion and a microcontroller for communication control. Our hardware-software co-design enables accurate measurements and stable concurrent transmission. We implement WindScatter and conduct extensive experiments and case studies to evaluate its performance. Results show that WindScatter supports measurements of all wind speed levels on the Extended Beaufort scale, from 1.5 m/s to 60 m/s, with an average error rate of 0.78%. WindScatter can sense and transmit wind speed data at a distance of 800 m with a power consumption of 136.5$\mu$W. Compared with commodity devices, WindScatter achieves comparable measurement range and accuracy while reducing cost by$91.5\times$and power consumption by$8,791\times$.

Electrode coating measurement is a crucial task in practical lithium-ion battery systems, where the thickness and refractive index of the electrode coating directly reflect the battery's quality, energy density, capacity, and lifespan. In this paper, we propose the design, implementation, and evaluation of BatTera, the practical system for an accurate, high-resolution, non-destructive, and safe electrode coating measurement, with the ability to simultaneously measure coating thickness and refractive index. BatTera's contributions are twofold. Firstly, we build a comprehensive mathematical model that characterizes the arrival time of echo signals from both sides of the electrode coating by thoroughly analyzing the electrode structure based on “coating-foil-coating”. This model serves as a theoretical foundation guiding the measurement of coating thickness and refractive index. Secondly, we propose a series of effective signal-processing algorithms to address the practical challenges of double-side coating misalignment and deformation interference, thus adaptive improving the signal-to-noise ratio of Terahertz signals and pushing BatTera one big step closer to real adoptions. We implement BatTera based on the commercial Terahertz device QT-TO1000 and conduct extensive experiments using five types of cathode electrode samples in three different sizes, collected from one of the world's largest new energy battery manufacturers. The results show that BatTera achieves high measurement accuracy with a mean average error of 6.106$\upmu$m for thickness and 0.230 for refractive index.

Millimeter-wave (mmWave) communication offers a promising solution for high-throughput, low-latency unmanned aerial vehicle (UAV) networks. However, maintaining strong received signal strength (RSS) remains a challenge due to UAV mobility. While existing studies have largely focused on beam alignment, they often overlook another critical issue: polarization mismatch caused by UAV orientation changes. This problem is particularly severe in cost-sensitive commercial off-the-shelf (COTS) mmWave devices, which typically employ linearly polarized (LP) antenna arrays. Our measurements reveal that even with perfect beam alignment, UAV orientation can still cause significant signal degradation due to polarization mismatch. To address this challenge, we propose PolarFix, a practical metasurface solution that enables real-time polarization matching without requiring any modifications to existing transceiver hardware. Specifically, we design a linear-to-circular polarization (L2C) metasurface that transforms linearly polarized (LP) waves into circularly polarized signals, allowing LP antennas to maintain consistent signal power despite changes in UAV orientation.

Carrier interference is a fundamental research challenge in backscatter systems. Existing solutions leverage frequency shifting or full duplex designs to mitigate carrier interference in the analog or digital domain. However, these solutions introduce extra spectrum usage, protocol overhead, and power consumption, all of which are undesirable in backscatter systems. In this paper, we revisit polarization mismatch and propose Lotus, a low-cost analog design to combat carrier interference for backscatter systems. Lotus comprises a novel antenna design and a backscatter tag design. At runtime, Lotus antenna replaces the receiver default antenna to cancel out the carrier interference, without any protocol or hardware overhead. Lotus tag mitigates the power loss caused by polarization mismatch and remains compatible with all existing backscatter radios. Experimental results show that Lotus achieves comparable cancellation gain (i.e., 42 dB) with the state-of-the-art frequency-shifting baseline. Meanwhile, Lotus outperforms the baseline by 2× and 6.5× in spectrum and power efficiency, respectively. Additionally, Lotus achieves comparable performance to the frequency shifting baseline regarding backscatter range and throughput across three backscatter technologies, including Wi-Fi, Bluetooth, and LoRa.

In recent years, there has been a growing focus on developing Low Power Wide Area Network (LPWAN) protocols, especially within the LoRa research community. However, the research community for NB-IoT, another crucial LPWAN technology, has not experienced comparable expansion due to the absence of a functional and adaptable software-defined radio (SDR) implementation. To address this gap, we present SoftNB, the first fully functional physical layer SDR implementation for NB-IoT. SoftNB conforms to the latest 3GPP standards and features an efficient and effective signal processing pipeline to mitigate time, frequency, and phase offsets during transmission and reception. Additionally, SoftNB is compatible with various SDR platforms, including USRP, HackRF One, and RTL-SDR Dongle. Extensive evaluations of SoftNB demonstrate its superior performance. Compared to the state-of-the-art baseline, SoftNB achieves an 8× reduction in Block Error Rate (BLER) when the number of repetitions is set to 4 at a distance of 450 meters.

Most of the federated learning techniques are limited to homogeneous model fusion. With the rapid growth of smart applications on resource-constrained edge devices, it becomes a barrier to accommodate their heterogeneous computing power and memory in the real world. Federated Distillation is a promising alternative to enable aggregation from heterogeneous models. However, the effectiveness of knowledge transfer still remains elusive under the shadow of distinct representation power from heterogeneous models. In this paper, we approach from an adversarial perspective to characterize the decision boundaries during distillation. By leveraging K-step PGD attacks, we successfully model the dynamics of the closest boundary points and establish a quantitative connection between the predictive uncertainty and boundary margin. Based on these findings, we further propose a new loss function to make the distillation attend to samples close to the decision boundaries, thus learning from more informed logit distributions. The extensive experiments over CIFAR-10/100 and Tiny-ImageNet demonstrate about 0.5-3.5% improvement of accuracy under different IID and non-IID settings, with only a small increment of computational overhead.

Object detection is a fundamental enabler for many real-time downstream applications such as autonomous driving, augmented reality and supply chain management. However, the algorithmic backbone of neural networks is brittle to imperceptible perturbations in the system inputs, which were generally known as misclassifying attacks. By targeting the real-time processing capability, a new class of latency attacks has been reported recently. They exploit new attack surfaces in object detectors by creating a computational bottleneck in the post-processing module, which leads to cascading failure and puts the real-time downstream tasks at risk. In this work, we take an initial attempt to defend against this attack via background-attentive adversarial training that is also cognizant of the underlying hardware capabilities. We first draw system-level connections between latency attacks and hardware capacity across heterogeneous GPU devices. Based on the particular adversarial behaviors, we utilize objectness loss as a proxy and build background attention into the adversarial training pipeline, and achieve a favorable balance between clean and robust accuracy. The extensive experiments demonstrate the effectiveness of the defense in restoring real-time processing capability from 13 FPS to 43 FPS on Jetson Orin NX, with a better trade-off between the clean and robust accuracy. The source code is available at: https://github.com/Hill-Wu1998/underload.

The emerging multi-modal applications exemplified by multi-DNN inference have renewed interests for mobile intelligence. The goal is to utilize heterogeneous processors on-board to maximize throughput and resource utilization. Among a variety of options, building model-parallel pipelines across different processors is a promising way. However, the existing efforts either focus on optimizing homogeneous DNN executions or simply ignore co-execution slowdown on the shared memory bus. Based on extensive empirical studies and insights with various degrees of resource contention, in this work, we introduce Hetero2Pipe, a two-step pipeline planner based on dynamic programming, and contention-mitigated pipeline bubble minimization to make the problem tractable within manageable search space. The extensive evaluation across three commercial SoCs demonstrates 2-8× speedup compared to the state-of-the-art schemes.

Large language models (LLMs) have emerged as a cornerstone for advancing AI technologies. It revolutionizes the way we interact with devices, websites, and information, and paves the way for the development of highly intuitive and capable virtual assistants. Training of today's LLMs happens in cloud data centers due to the requirement of enormous data and a significant amount of computing power. Despite extensive research in mobile edge computing, fine-tuning pre-trained LLMs using resource-constrained devices like commodity smartphones remains highly under-explored. In this paper, we propose Confidant, a practical collaborative training framework that allows modern LLMs to be fine-tuned across multiple off-the-shelf mobile devices. To this end, Confidant partitions an LLM into several sub-models, allowing each of them to fit in the memory of a mobile device. Multiple mobile devices then collaborate to train the LLM by employing a novel pipeline parallel training approach. In specific, Confidant encompasses a memory-aware dynamic model partitioning and intra-device multi-processor scheduler to minimize the training time across heterogeneous platforms. To ensure resilient distributed training, a hybrid fault tolerance mechanism is devised to proactively manage potential device and network failures. We fully implemented Confidant in C++/Python, and built a cross-framework adapter, enabling collaborative training on a variety of mobile platforms. Experimental results show that Confidant excels in achieving computation-, memory-efficient, and robust customization of LLMs - it manages to train state-of-the-art billion-sized LLMs including BERT, GPT-2, Phi2, and LLaMA3, and fine-tunes Phi2-2.7B on Alpaca in just 40.1 hours using three consumer-grade mobile devices.

Despite large language models (LLMs) being an essential part of our lives, training of LLMs still needs to be done in cloud data centers due to the large requirements of data and computing power, leaving fine-tuning pre-trained LLMs on resource-constrained mobile devices remains highly under-explored. In this demo, we present Confidant, a practical collaborative training system that allows modern LLMs to be fine-tuned across multiple off-the-shelf mobile devices. Confidant partitions an LLM into several sub-models, deploying each of them to a mobile device. Multiple mobile devices then collaborate to train the LLM by employing a novel pipeline parallel training approach. Specifically, Confidant encompasses a memory-aware dynamic model partitioning and intra-device multi-processor scheduler to minimize the training time across heterogeneous platforms. A hybrid fault tolerance mechanism is also devised to proactively manage potential device and network failures. By building a cross-framework adapter and fully implementing Confidant on smartphones and laptops, we present the demo of collaborative training on a variety of mobile platforms.

Millimeter-wave (mmWave) sensing technology holds significant value in human-centric applications, yet the high costs associated with data acquisition and annotation limit its widespread adoption in our daily lives. Concurrently, the rapid evolution of large language models (LLMs) has opened up opportunities for addressing complex human needs. This paper presents mmExpert, an innovative mmWave understanding framework consisting of a data generation flywheel that leverages LLMs to automate the generation of synthetic mmWave radar datasets for specific application scenarios, thereby training models capable of zero-shot generalization in real-world environments. Extensive experiments demonstrate that the data synthesized by mmExpert significantly enhances the performance of downstream models and facilitates the successful deployment of large language models for mmWave understanding.

3D reconstruction plays a critical role in applications such as augmented reality (AR) and robotic systems. However, implicit neural representations (INRs), widely used in modern 3D reconstruction systems, demand substantial communication and computational resources, making the reconstruction process excessively slow and costly. In this paper, we introduce EAR-Mapping, a novel edge-assisted online 3D reconstruction framework designed for latency-sensitive mobile applications. EAR-Mapping incorporates an innovative sampling mechanism that seamlessly integrates explicit and implicit methods, enabling selective processing of camera data to maximize reconstruction performance. In addition, we utilize a value-based representation module to maximize computation resource efficiency. Finally, we design a framework that minimizes communication overhead through ROI-based data transmission. Our prototype implementation on a mobile-edge testbed demonstrates that EAR-Mapping achieves up to a 1.2x reduction in reconstruction latency and a 3.5x reduction in bandwidth usage, offering a significant advancement in the efficiency of 3D reconstruction for mobile-edge systems.

Vehicle edge computing can effectively ensure the quality of experience for user vehicles (UVs), but road side units (RSUs) with limited resources may not be able to handle intensive tasks under high traffic conditions. In this case, worker vehicles (WVs) with idle resources can share resources to alleviate the pressure on RSUs. However, selfish WVs may be reluctant to share idle computation resources without any rewards. In addition, the optimization problems in previous research are relatively simple and cannot be applied to complex scenarios. To address the above challenges, we propose an incentive-driven partial offloading framework aiming to maximize social welfare. In particular, the computing service provider (CSP) managing RSUs first determines resource prices and offloading rates with UVs, while also determining contract terms with WVs. Then, it generates the optimal task scheduling strategy and notifies the UVs to offload tasks to the corresponding WVs. Considering that maximizing social welfare is a mixed-integer nonlinear programming (MINLP) problem, we design the hybrid proximal policy optimization (HPPO)-based task offloading and resource allocation algorithm (HORA) with a hybrid action space to directly solve the original problem. Finally, extensive simulation results show that HORA outperforms other baseline methods across various scenarios, and the contract terms meet the constraints of individual rationality (IR) and incentive compatibility (IC).

Millimeter-wave (mmWave) communication has great potential for high rates and low latency but suffers from severe non-line-of-sight (NLOS) blockage and high cost of beam alignment. These problems exacerbate for mobile industrial robots. Most existing methods use reconfigurable intelligent surface (RIS) to change the channel environment, whose practical applications are hindered by high cost. This paper introduces low-cost mmFlower, consisting of an array of 3D-printing reflector units whose orientation can be adjusted mechanically. We first construct a theoretical reflection model, bridging the mechanical parameters and communication metrics. mmFlower works in two modes according to whether it needs real-time reconfiguration. On the one hand, based on our uniform-RSS (received signal strength) projection algorithm, mmFlower remains static and reshapes mmWave into arbitrary projection trajectory along the robot movement. It maximizes mmWave allocation to the more frequently visited areas and realizes seamless mmWave coverage. On the other hand, mmFlower can dynamically track mmWave pencil beams on robots, providing higher reflection gain. Extensive evaluations in an airplane assembly factory show the superiority of mmFlower on reflection gain (supporting up to 1.8 Gbps in NLOS), flexibility to different trajectories of robots, robustness to deployment deviation, etc.

Whilesmalldrone video streaming systems create unprecedented video content, they also place a power burden exceeding 20% on the drone's battery, limiting flight endurance. We present${\sf Mighty}$, a hardware-software solution to minimize the power consumption of a drone's video streaming system by offloading power overheads associated with both video compression and transmission to a ground controller.${\sf Mighty}$innovates a high performance co-design among:(1)a ring oscillator-based, ultra-low power backscatter radio;(2)a spectrally-efficient, non-linear, low-power physical layer modulation and multi-chain radio architecture; and(3)a lightweight video compression codec-bypassing software design. Our co-design exploits synergies among these components, resulting in joint throughput and range performance that pushes the known envelope. We prototype${\sf Mighty}$on PCB board and conduct extensive field studies both indoors and outdoors. The power efficiency of${\sf Mighty}$is about 16.6 nJ/bit. A head-to-head comparison with aDJI Mini2drone's default video streaming system shows that${\sf Mighty}$achieves similar throughput at a drone-to-controller distance of up to 150 meters, with 34–55× improvement of power efficiency than WiFi-based video streaming solutions.

Frequency switching is a fundamental capability for wireless communication systems. However, this capability is significantly constrained in backscatter systems. The difficulty is to generate tunable high-frequency modulation signals on a backscatter tag at an acceptable power budget. In this paper, we present Twaltz, a new design paradigm for backscatter communication that enables frequency switching across large frequency bands. By exploiting a low-power semiconductor device, i.e., tunnel diode, and carefully addressing its physical features, Twaltz generates oscillation signals up to 1.2 GHz while maintaining micro-watt level power consumption. Twaltz further facilitates on-tag oscillation signal stabilization and programmable oscillation frequency tuning. We prototype Twaltz on a PCB board, demonstrating its efficiency in cross-band communication for LoRa backscatter, and verifying its performance in concurrent transmission, channel hopping, and data transmission.

With the rapid development of Connected and Autonomous Vehicles, In-Vehicle Network attacks have garnered heightened research scrutiny due to vehicles’ increasing connectivities to the external environment. The common characteristic among these attacks is to tamper with targeted powertrain-related signals in the Powertrain Controller Area Network (PT-CAN) and further physically threaten vehicles’ safety. These powertrain-related signals are encoded within CAN messages grounded by the syntax specification, which is proprietary to Original Equipment Manufacturers and publicly unavailable. Thus, to undertake comprehensive security analysis and strategies, reverse engineering PT-CAN to the semantic level is urgently needed. However, the existing methods rely on interactions (injecting challenge signals/actions) with the targeted vehicle, and certain manual efforts are required. To fill this gap, we proposePicaCAN, a novel framework to extract signals from CAN messages and reverse engineer their physical semantics based on physically induced causality. Once access to the CAN traffic,PicaCANoffers the researcher an eye on the vehicle’s powertrain system, decoding binaries flows into powertrain-related signals automatically. We experimentally evaluatePicaCANon PT-CAN of three automobiles containing two power types. The experimental results show thatPicaCANcould successfully extract physical signals representing all targeted semantics (pedals, engine speed, etc.) from two Internal Combustion Engine Vehicles and one Hybrid Electric Vehicle under EV mode.

The virtual switch (vSwitch) serves as a fundamental element in cloud network, critical for high-performance and strongly isolated inter-VM forwarding in local and external networks. Similar to other multicore systems, a vSwitch with multiple cores also faces the issue of core load imbalance. As a major cloud provider, we pinpoint four cases of core load imbalance within the vSwitch in our cloud, stemming from unequal traffic distribution across virtual queues and RSS buckets, as well as from traffic patterns like heavy hitters and micro-bursts. To tackle the different load imbalance cases, we present vSwitchLB, a vSwitch load balance framework. Specifically, we introduce a load imbalance detection module, accompanied by dedicated techniques designed to address each specific type of imbalance. Our preliminary evaluation shows that vSwitchLB can accurately classify different load imbalances encountered in the vSwitch on our cloud and then prevent any single core of vSwitch from being flooded and overwhelmed.

The design of Low Power Wide Area Network (LPWAN) protocols has attracted increasing attention in recent years, particularly within the LoRa research community. However, NB-IoT, another critical LPWAN technology, has not seen similar growth in its research community due to the lack of a functional and flexible software-defined radio (SDR) implementation. To address this gap, we present SoftNB, the first fully functional physical layer SDR implementation for NB-IoT. SoftNB conforms to the latest 3GPP standards and features an efficient and effective signal processing pipeline to mitigate time, frequency, and phase offsets during transmission and reception. Additionally, SoftNB is compatible with various SDR platforms, including USRP, HackRF One, and RTL-SDR Dongle. Extensive evaluations of SoftNB demonstrate its superior performance. Compared to the state-of-the-art baseline, SoftNB achieves an$8\times$reduction in Block Error Rate (BLER) when the number of repetitions is set to 4 at a distance of 450 meters.

This paper presents the design, implementation, and evaluation of MagWear, a novel biomagnetism-based system that can accurately and inclusively monitor the heart rate and respiration rate of mobile users with diverse skin tones. MagWear's contributions are twofold. Firstly, we build a mathematical model that characterizes the magnetic coupling effect of blood flow under the influence of an external magnetic field. This model uncovers the variations in accuracy when monitoring vital signs among individuals. Secondly, leveraging insights derived from this mathematical model, we present a softwarehardware co-design that effectively handles the impact of human diversity on the performance of vital sign monitoring, pushing this generic solution one big step closer to real adoptions. We have implemented a prototype of MagWear on a two-layer PCB board and followed IRB protocols to conduct system evaluations. Our extensive experiments involving 30 volunteers demonstrate that MagWear achieves high monitoring accuracy with a mean percentage error (MPE) of 1.55% for heart rate and 1.79% for respiration rate. The head-to-head comparison with Apple Watch 8 further demonstrates MagWear's consistently high performance in different user conditions.

As society progresses, liquid identification plays an increasingly important role in human life. But for now, minority of existing liquid identification solutions on the market can meet daily requirements of being ubiquitous, cost-effective and non-intrusive enough. In this work, we propose ASLiquid, the first liquid counterfeit identification system with commercial off-the-shelf earphones. Our core insight is that earphones can effectively induce acoustic resonance in container, and this phenomenon is observed highly associated with the changes in liquid density and solute compositions. Deploying ASLiquid introduces three main challenges: hardware heterogeneity among different earphones, diversity of user operations, and data complexity due to variations in liquid volume and device placement. To address these issues, we first propose to eliminate the existence of hardware noise and frequency response diversity for an earphone-irrelevant solution. Afterwards, we design a user operation adaptation algorithm to extract valuable feature data during each measurement period. To alleviate problems in data complexity, we propose a spectrum projection algorithm that can effectively generate CFR data of unknown liquid volumes and a VAE based anomaly detection model for counterfeit identification. We evaluate our system with six different earphones and under various conditions. Experimental results reveal that ASLiquid can achieve F1 scores of 95%-99.25% for seven frequently occurring liquid counterfeit tasks, even in specialized attacks on liquids with 1% difference in mass fraction and different types of solutions but with the same density.

It is usually infeasible to fit and train an entire large deep neural network (DNN) model using a single edge device due to the limited resources. To facilitate intelligent applications across edge devices, researchers have proposed partitioning a large model into several sub-models, and deploying each of them to a different edge device to collaboratively train a DNN model. However, the communication overhead caused by the large amount of data transmitted from one device to another during training, as well as the sub-optimal partition point due to the inaccurate latency prediction of computation at each edge device can significantly slow down training. In this paper, we propose AccEPT, an acceleration scheme for accelerating the edge collaborative pipeline-parallel training. In particular, we propose a light-weight adaptive latency predictor to accurately estimate the computation latency of each layer at different devices, which also adapts to unseen devices through continuous learning. Therefore, the proposed latency predictor leads to better model partitioning which balances the computation loads across participating devices. Moreover, we propose a bit-level computation-efficient data compression scheme to compress the data to be transmitted between devices during training. Our numerical results demonstrate that our proposed acceleration approach is able to significantly speed up edge pipeline parallel training up to 3 times faster in the considered experimental settings

This paper presents the design, implementation, and evaluation of MagWear, a novel biomagnetism-based system that can accurately and inclusively monitor the heart rate, respiration rate, and blood pressure of users. MagWear's contributions are twofold. First, we build a mathematical model that characterizes the magnetic coupling effect of blood flow under the influence of an external magnetic field. This model uncovers the variations in accuracy when monitoring vital signs among individuals. Second, leveraging insights derived from this mathematical model, we present a software-hardware co-design that effectively handles the impact of human diversity on the performance of vital sign monitoring, pushing this generic solution one big step closer to real adoptions. Following IRB protocols, our extensive experiments involving 30 volunteers demonstrate that MagWear achieves high monitoring accuracy with a mean percentage error (MPE) of 1.55% for heart rate (HR), 1.79% for respiration rate (RR), 3.35% for systolic blood pressure (SBP), and 3.89% for diastolic blood pressure (DBP). MagWear can also be extended to detect anemia and blood oxygen saturation, which is also our ongoing work.

In this paper, we study a new domain adaptation setting on camera networks, namely Multi-View Domain Adaptive Object Detection (MVDA-OD), in which labeled source data is unavailable in the target adaptation process and target data is captured from multiple overlapping cameras. In such a challenging context, existing methods including adversarial training and self-training fall short due to multi-domain data shift and the lack of source data. To tackle this problem, we propose a novel training framework consisting of two stages. First, we pre-train the backbone using self-supervised learning, in which a multi-view association is developed to construct an effective pretext task. Second, we fine-tune the detection head using robust self-training, where a tracking-based single-view augmentation is introduced to achieve weak-hard consistency learning. By doing so, an object detection model can take advantage of informative samples generated by multi-view association and single-view augmentation to learn discriminative backbones as well as robust detection classifiers. Experiments on two real-world multi-camera datasets demonstrate significant advantages of our approach over the state-of-the-art domain adaptive object detection methods.

We organized an online fingerprint-based indoor localization competition in 2021. It attracted 1,170 teams worldwide. The teams were provided with a 60 GB dataset including WiFi, BLE, IMU, and geomagnetic field strength data collected from 204 buildings to build their localization algorithms, which were then evaluated against a separate test dataset. The competition received 28,009 submissions. The top team achieved an average accuracy of 1.50m. This paper reports the lessons we learned from analyzing the submissions, as well as our experiences in organizing the competition, through both qualitatively studying the teams' algorithms and quantitatively characterizing the competition results.

Today, face editing is widely used to refine/alter photos in both professional and recreational settings. Yet it is also used to modify (and repost) existing online photos for cyberbullying. Our work considers an important open question: 'How can we support the collaborative use of face editing on social platforms while protecting against unacceptable edits and reposts by others?' This is challenging because, as our user study shows, users vary widely in their definition of what edits are (un)acceptable. Any global filter policy deployed by social platforms is unlikely to address the needs of all users, but hinders social interactions enabled by photo editing. Instead, we argue that face edit protection policies should be implemented by social platforms based on individual user preferences. When posting an original photo online, a user can choose to specify the types of face edits (dis)allowed on the photo. Social platforms use these per-photo edit policies to moderate future photo uploads, i.e., edited photos containing modifications that violate the original photo's policy are either blocked or shelved for user approval. Realizing this personalized protection, however, faces two immediate challenges: (1) how to accurately recognize specific modifications, if any, contained in a photo; and (2) how to associate an edited photo with its original photo (and thus the edit policy). We show that these challenges can be addressed by combining highly efficient hashing based image search and scalable semantic image comparison, and build a prototype protector (Alethia) covering nine edit types. Evaluations using IRB-approved user studies and data-driven experiments (on 839K face photos) show that Alethia accurately recognizes edited photos that violate user policies and induces a feeling of protection to study participants. This demonstrates the initial feasibility of personalized face edit protection. We also discuss current limitations and future directions to push the concept forward.

Vehicle tracking, a core application to smart city video analytics, is becoming more widely deployed than ever before thanks to the increasing number of traffic cameras and recent advances in computer vision and machine-learning. Due to the constraints of bandwidth, latency, and privacy concerns, tracking tasks are more preferable to run on edge devices sitting close to the cameras. However, edge devices are provisioned with a fixed amount of computing budget, making them incompetent to adapt to time-varying and imbalanced tracking workloads caused by traffic dynamics. In coping with this challenge, we propose WatchDog, a real-time vehicle tracking system that fully utilizes edge nodes across the road network. WatchDog leverages computer vision tasks with different resource-accuracy tradeoffs, and decomposes and schedules tracking tasks judiciously across edge devices based on the current workload to maximize the number of tasks while ensuring a provable response time-bound at each edge device. Extensive evaluations have been conducted using real-world city-wide vehicle trajectory datasets, achieving exceptional tracking performance with a real-time guarantee.

As various computing devices, such as smartphones, IoT devices, smart speakers etc, becomes omnipresent in our daily lives, interest in ubiquitous computing interfaces is increasing. In response to this, various studies have introduced on-surface input techniques that leverage the surface of surrounding objects as touch interfaces. However, most of them struggle to support ubiquitous interaction due to their dependency on specific hardware or environments. In this work, we propose${{\sf S \text{-}UbiTap}}$, an input method that turns any flat solid surface into a touch input space by listening to sound (i.e., with microphones already present in the commodity devices). More specifically, we develop a novel touch localization technique that leverages the physical phenomenon, calleddispersion, which is the characteristic of sound as it travels through solid surfaces, and address the challenges that limit existing acoustic-based solutions in terms of portability, accuracy, usability, robustness, scalability, and responsiveness. Our extensive experiments with a prototype of${{\sf S \text{-}UbiTap}}$show that we can support sub-centimeter accuracy on various types of surfaces with minor user calibration effort. In addition, the accuracy is maintained even when the size of the touch input space increases. In our experience with real-world users,${{\sf S \text{-}UbiTap}}$significantly improves usability and robustness, thus enabling the emergence of more exciting applications.

Edge computing is being widely used for video analytics. To alleviate the inherent tension between accuracy and cost, various video analytics pipelines have been proposed to optimize the usage of GPU on edge nodes. Nonetheless, we find that GPU compute resources provisioned for edge nodes are commonly under-utilized due to video content variations, subsampling and filtering at different places of a video analytics pipeline. As opposed to model and pipeline optimization, in this work, we study the problem of opportunistic data enhancement using the non-deterministic and fragmented idle GPU resources. In specific, we propose a task-specific discrimination and enhancement module, and a model-aware adversarial training mechanism, providing a way to exploit idle resources to identify and transform pipeline-specific, low-quality images in an accurate and efficient manner. A multi-exit enhancement model structure and a resource-aware scheduler is further developed to make online enhancement decisions and fine-grained inference execution under latency and GPU resource constraints. Experiments across multiple video analytics pipelines and datasets reveal that our system boosts DNN object detection accuracy by 7.27 -- 11.34% by judiciously allocating 15.81 -- 37.67% idle resources on frames that tend to yield greater marginal benefits from enhancement.

Object detection is a fundamental building block of video analytics applications. While Neural Networks (NNs)-based object detection models have shown excellent accuracy on benchmark datasets, they are not well positioned for high-resolution images inference on resource-constrained edge devices. Common approaches, including down-sampling inputs and scaling up neural networks, fall short of adapting to video content changes and various latency requirements. This paper presents Remix, a flexible framework for high-resolution object detection on edge devices. Remix takes as input a latency budget, and come up with an image partition and model execution plan which runs off-the-shelf neural networks on non-uniformly partitioned image blocks. As a result, it maximizes the overall detection accuracy by allocating various amount of compute power onto different areas of an image. We evaluate Remix on public dataset as well as real-world videos collected by ourselves. Experimental results show that Remix can either improve the detection accuracy by 18%-120% for a given latency budget, or achieve up to 8.1× inference speedup with accuracy on par with the state-of-the-art NNs.

Information barrier coverage has been widely adopted to prevent unauthorized invasion of important areas in Internet of Things. As sensors are typically placed outdoors, they are susceptible to getting faulty. Previous works assumed that faulty sensors are easy to recognize, e.g., they may stop functioning or output apparently deviant sensory data. In practice, there exist multiple types of fault that sensors may have during operation. It is, thereby, difficult to recognize faulty sensors as well as their invalid output and attain accurate intrusion detection. We, in this paper, propose a novel fault-tolerant intrusion detection algorithm (TrusDet) based on trust management to address this challenging issue. TrusDet comprises of three steps: i) sensor-level detection, ii) sink-level decision by collective voting, and iii) trust management and fault determination. In the Step i) and ii), TrusDet divides the surveillance area into a set of fine-grained subareas and exploits temporal and spatial correlation of sensory output among sensors in different subareas to yield a more accurate and robust performance of information barrier coverage. In the Step iii), TrusDet builds a trust management based framework to determine the confidence level of sensors being faulty. We implement TrusDet on HC-SR501 infrared sensors, and design hardware and software to build a practical detection system. Extensive experimental results and simulation results validate the information coverage model and demonstrate that TrusDet has a very low false alarm rate.

Cameras are deployed at scale with the purpose of searching and tracking objects of interest (e.g., a suspected person) through the camera network on live videos. Such cross-camera analytics is data and compute intensive, whose costs grow with the number of cameras and time. We present Spatula, a cost-efficient system that enables scaling cross-camera analytics on edge compute boxes to large camera networks by leveraging the spatial and temporal cross-camera correlations. While such correlations have been used in computer vision community, Spatula uses them to drastically reduce the communication and computation costs by pruning search space of a query identity (e.g., ignoring frames not correlated with the query identity’s current position). Spatula provides the first system substrate on which cross-camera analytics applications can be built to efficiently harness the cross-camera correlations that are abundant in large camera deployments. Spatula reduces compute load by $8.3\times$ on an 8-camera dataset, and by $23\times-86\times$ on two datasets with hundreds of cameras (simulated from real vehicle/pedestrian traces). We have also implemented Spatula on a testbed of 5 AWS DeepLens cameras.

Given a wide spectrum of demands for indoor location-based service, great research effort has been devoted to developing indoor navigation systems. Nevertheless, due to high engineering complexity and expensive infrastructure and labor cost, scalable indoor navigation is still an unsolved problem. In this paper, we present SWiN, a Self-evolving WiFi-based Indoor Navigation system. SWiN provides plug-and-play and light-weight indoor navigation in a sharing manner. To alleviate the impact of the environmental change and device diversity, SWiN extracts both the static and dynamic properties of WiFi signals including scanned AP list, variations of signal strength, and AP's relative strength order. SWiN exploits the leader-follower structure, navigating following users by tracking their motion patterns to provide real-time navigation guidance. In specific, during navigation, SWiN utilizes a light-weight synchronization algorithm to synchronize multi-dimensional WiFi measurements between leader and follower traces. Furthermore, a trace updating mechanism is developed to guarantee the long-term utility of SWiN by extracting useful information in followers' traces. Consolidating these techniques, we implement SWiN on commodity smartphones, and evaluate its performance in a five-story office building and a newly opened two-story shopping mall with test areas over 8000 m2and 6000 m2, respectively. Our experimental results show that 95 percent of the tracking offsets during navigation are less than 2 m and 3.2 m in these two environments.

Despite years of research attention, localization-based indoor navigation has not found wide-spread practical use, largely due to the high burden on deployment and bootstrapping. Lightweight peer-to-peer navigation systems that use a leader-follower model have recently been proposed to alleviate these burdens. However, typical peer-to-peer navigation suffers from poor scalability and flexibility as navigation is only possible over pre-collected leader paths. In this paper, we present FollowUs, an easily-deployable (bootstrap-free) and scalable indoor navigation system. In addition to robust navigation through real-time trace-following, FollowUs integrates cloud services to process and combine traces at large scale. Optionally, it can also leverage floor plans to further enhance navigation efficiency. We design and implement FollowUs, including mobile app and cloud services. Experimental results from a company-internal beta release show that 91% of FollowUs' spatial errors on reaching destinations to be 3m or less, and 95% of navigation instructions are shown to users within a 4-step error margin during navigation.

No abstract available.

The automotive industry is increasingly employing software- based solutions to provide value-added features on vehicles, especially with the coming era of electric vehicles and autonomous driving. The ever-increasing cyber components of vehicles (i.e., computation, communication, and control), however, incur new risks of anomalies, as demonstrated by the millions of vehicles recalled by different manufactures. To mitigate these risks, we design B-Diag, a battery-based diagnostics system that guards vehicles against anomalies with a cyber-physical approach, and implement B-Diag as an add-on module of commodity vehicles attached to automotive batteries, thus providing vehicles an additional layer of protection. B-Diag is inspired by the fact that the automotive battery operates in strong dependency with many physical components of the vehicle, which is observable as correlations between battery voltage and the vehicle's corresponding operational parameters, e.g., a faster revolutions-per-minute (RPM) of the engine, in general, leads to a higher battery voltage. B-Diag exploits such physically-induced correlations to diagnose vehicles by cross-validating the vehicle information with battery voltage, based on a set of data-driven norm models constructed online. Such a design of B-Diag is steered by a dataset collected with a prototype system when driving a 2018 Subaru Crosstrek in real-life over 3 months, covering a total mileage of about 1, 400 miles. Besides the Crosstrek, we have also evaluated B-Diag with driving traces of a 2008 Honda Fit, a 2018 Volvo XC60, and a 2017 Volkswagen Passat, showing B-Diag detects vehicle anomalies with >86% (up to 99%) averaged detection rate.

The world is witnessing an unprecedented increase in camera deployment. The USA and UK, for instance, have one camera for every 8 people. Video analytics from these cameras are becoming more and more pervasive, exerting important functions on a wide range of verticals including manufacturing, transportation, and retails. While vision techniques have seen considerable advancement, they have come at the expense of compute and network cost.

As an innovative mobility strategy, public bike-sharing has grown dramatically worldwide. Though it provides convenient, low-cost, and environmental-friendly transportation, the unique features of bike-sharing systems give rise to problems for both users and operators. The primary issue is the uneven distribution of bikes caused by ever-changing usage and (available) supply. This imbalance necessitates efficient bike rebalancing strategies, which depends highly on bike mobility modeling and prediction. In this paper, a trace-driven simulation-based prediction approach is proposed by simultaneously taking user mobility demand and real-time status of stations into consideration. We extensively evaluate the performance of our design with the dataset from one of the world's largest public bike-sharing systems located in Hangzhou, China, which owns more than 2800 stations. The evaluation results show an 85 percentile relative error of 0.6 for checkout and 0.4 for checkin prediction. The preliminary results on how the predictions can be used for bike rebalancing are also provided. We believe that this new mobility modeling and prediction approach can improve the bike-sharing system operation algorithm design and pave the way for rapid deployment and adoption of bike-sharing systems across the globe.

The rapid development of bike-sharing systems has brought people enormous convenience during the past decade. On the other hand, high transport flexibility gives rise to problems for both users and operators. For users, dynamic distribution of shared bikes caused by uneven user demand often leads to the check in or check out service unavailable at some stations. For operators, unbalanced bike usage comes with more bike broken and growing maintenance cost. In this paper, we consider enhancing user experiences and rebalance bicycle utilization by directing users to different stations with a higher success rate of rental and return. For the first time, we devise a trip advisor that recommends bike check-in and check-out stations with joint consideration of service quality and bicycle utilization. To ensure service quality, we firstly predict the user demand of each station to obtain the success rate of rental and return in the future. Experiments indicate that the precision of our method is as much as 0.826, which has raised by 25.9 percent as compared with that of the historical average method. To rebalance bike usage, from historical data, we identify that biased bike usage is rooted from circumscribed bicycle circulation among few active stations. Therefore, with defined station activeness, we optimize the bike circulation by leading users to shift bikes between highly active stations and inactive ones. We extensively evaluate the performance of our design through real-world datasets. Evaluation results show that the percentage of frequently used bikes decreases by 33.6 percent on usage number and 28.6 percent on usage time.

With the omnipresence of computing devices in our daily lives, interests in ubiquitous computing interfaces have grown. In response to this, various studies have introduced on-surface input techniques which use the surfaces of surrounding objects as a touch interface. However, these methods are yet struggling to support ubiquitous interaction due to their dependency on specific hardware or environments. In this paper, we propose UbiTap, an input method that turns solid surfaces into a touch input space, through the use of sound (i.e., with microphones already present in the commodity devices). More specifically, we develop a novel touch localization technique which leverages the physical phenomenon, referred to as dispersion, a characteristic of sound as it travels through solid surfaces, so as to address challenges which limit existing acoustic-based solutions in terms of portability, accuracy, usability, robustness, and responsiveness. Our extensive experiments with a prototype of UbiTap show that we can support sub-centimeter accuracy on various surfaces with minor user calibration effort. In our experience with real-world users, UbiTap significantly improves usability and robustness, thus enabling the emergence of more exciting applications.

Rapid development of bike-sharing systems has brought people enormous convenience during the past decade. On the other hand, high transport flexibility comes with dynamic distribution of shared bikes, leading to an unbalanced bike usage and growing maintenance cost. In this paper, we consider to rebalance bicycle utilization by means of directing users to different stations. For the first time, we devise a trip advisor that recommends bike check-in and check-out stations with joint consideration of service quality and bicycle utilization. From historical data, we firstly identify that biased bike usage is rooted from circumscribed bicycle circulation among few active stations. Therefore, with defined station activeness, we optimize the bike circulation by leading users to shift bikes between highly active stations and inactive ones. We extensively evaluate the performance of our design through real-world datasets. Evaluation results show that the percentage of frequent used bikes decreases by 33.6% on usage number and 28.6% on usage time.

Bootstrapping efforts and scalability issues hinder large-scale deployment of indoor navigation systems. We present FollowUs, an easily-deployable (bootstrap-free) and scalable indoor navigation system. In addition to robust navigation through real-time trace-following, FollowUs integrates cloud services to process and combine traces at large scale. It can also leverage optional floor plans to further enhance navigation performance. We designed and implemented FollowUs, including a mobile app and cloud services on Azure, and validate its real-world usability.

In this demo, we propose I-Navi, an Indoor Navigation system which leverages the gradient WiFi signal. To be more adaptive to time-variant RSSI and enrich information dimension, I-Navi exploits a three-step backward gradient binary method. Meanwhile, we adopt a lightweight online dynamic time warping (DTW) algorithm to achieve real-time navigation. We fully implemented I-Navi on smartphones and conducted extensive experiments in a five-story campus building and a newly opened two-floor shopping mall with a 90% accuracy of 2m and 3.2m achieved at two places.

Barrier coverage has been widely adopted to prevent unauthorized invasion of important areas in sensor networks. As sensors are typically placed outdoors, they are susceptible to getting faulty. Previous works assumed that faulty sensors are easy to recognize, e.g., they may stop functioning or output apparently deviant sensory data. In practice, it is, however, extremely difficult to recognize faulty sensors as well as their invalid output. We, in this paper, propose a novel fault-tolerant intrusion detection algorithm (TrusDet) based on trust management to address this challenging issue. TrusDet comprises of three steps: i) sensor-level detection, ii) sink-level decision by collective voting, and iii) trust management and fault determination. In the Step i) and ii), TrusDet divides the surveillance area into a set of fine- grained subareas and exploits temporal and spatial correlation of sensory output among sensors in different subareas to yield a more accurate and robust performance of barrier coverage. In the Step iii), TrusDet builds a trust management based framework to determine the confidence level of sensors being faulty. We implement TrusDet on HC- SR501 infrared sensors and demonstrate that TrusDet has a desired performance.

Wireless charging is a promising way to solve the energy constraint problem in sensor networks. While extensive efforts have been made to improve the performance of charging and communication in wireless rechargeable sensor networks (WRSNs), little has been done to address the operation scheduling problem. To fill this void, we propose a joint energy replenishment and scheduling mechanism so as to maximize the network lifetime while making strict sensing guarantees in the WRSN. We first formulate the problem in a general 2-D space and prove its NP-completeness. We then devise an f-approximate scheduling mechanism by transforming the classical minimum set cover problem and develop an optimal energy-replenish strategy based on the energy consumption of nodes returned by the scheduling mechanism. Large-scale simulation results validate our design and show a 39.2% improvement of network lifetime over a baseline method.

As an innovative mobility strategy, public bike-sharing has grown dramatically worldwide. Though providing convenient, low-cost and environmental-friendly transportation, the unique features of bike-sharing systems give rise to problems to both users and operators. The primary issue among these problems is the uneven distribution of bicycles caused by the ever-changing usage and (available) supply. This bicycle imbalance issue necessitates efficient bike re-balancing strategies, which depends highly on bicycle mobility modeling and prediction. In this paper, for the first time, we propose a spatio-temporal bicycle mobility model based on historical bike-sharing data, and devise a traffic prediction mechanism on a per-station basis with sub-hour granularity. We extensively evaluated the performance of our design through a one-year dataset from the world's largest public bike-sharing system (BSS) with more than 2800 stations and over 103 million check in/out records. Evaluation results show an 85 percentile relative error of 0.6 for both check in and check out prediction. We believe this new mobility modeling and prediction approach can advance the bike re-balancing algorithm design and pave the way for the rapid deployment and adoption of bike-sharing systems across the globe.

We present a system utilizing narrow band software defined radio to detect the moving human through walls, and give some motion details, such as motion orientation which includes relative moving direction. To achieve high accuracy, FindIt applies Short Time Fourier Transform (STFT) and statistical methods to received signals. In order to adapt to different environments, FindIt uses clustering and classification methods to determine thresholds. Moreover, FindIt provides user-friendly real-time detection results, which can be used as a trigger of high-level functions.

Wireless sensor networks have been used in many mobile applications such as wildlife tracking and participatory urban sensing. Because of the combination of high mobility and low-duty-cycle operations, it is a challenging issue to reduce discovery delay among mobile nodes, so that mobile nodes can establish connection quickly once they are within each other's vicinity. Existing discovery designs are essentially pairwise based, in which discovery is passively achieved when two nodes are prescheduled to wake up at the same time. In contrast, this work reduces discovery delay significantly by proactively referring wake-up schedules among a group of nodes. Since proactive references incur additional overhead, we introduce a novel selective reference mechanism based on spatiotemporal properties of neighborhood and the mobility of nodes. Our quantitative analysis indicates that the discovery delay of our group-based mechanism is significantly smaller than that of the pairwise one. Our testbed experiments using 40 sensor nodes and extensive simulations confirm the theoretical analysis, showing one order of magnitude reduction in discovery delay compared with legacy pairwise methods in dense, uniformly distributed sensor networks with at most 8.8 percent increase in energy consumption.

Limited energy in each node is the major design constraint in wireless sensor networks (WSNs). To overcome this limit, wireless rechargeable sensor networks (WRSNs) have been proposed and studied extensively over the last few years. In a typical WRSN, batteries in sensor nodes can be replenished by a mobile charger that periodically travels along a certain trajectory in the sensing area. To maximize the charged energy in sensor nodes, one fundamental question is how to control the traveling velocity of the charger. In this paper, we first identify the optimal velocity control as a key design objective of mobile wireless charging in WRSNs. We then formulate the optimal charger velocity control problem on arbitrarily-shaped irregular trajectories in a 2D space. The problem is proved to be NP-hard, and hence a heuristic solution with a provable upper bound is developed using novel spatial and temporal discretization. We also derive the optimal velocity control for moving the charger along a linear (1D) trajectory commonly seen in many WSN applications. Extensive simulations show that the network lifetime can be extended by 2.5× with the proposed velocity control mechanisms.

In this demo, we propose Phone meter which leverages the RF energy harvesting technologies to measure the strength of Electromagnetic Field (EMF). To this end, Phone meter combines EMF sensor with the smartphone through audio interface without any modifications to the phone. We fully implement the low-cost Phone meter and conduct extensive experiments to prove the functionality of Phone meter. Phone meter achieves about 13:7% relative error in average compared with the industrial-grade spectrum analyzer with significantly reduced the costs.

Although GPS has become a standard component of smartphones, providing accurate navigation during the last portion of a trip remains an important but unsolved problem. Despite extensive research on localization, the limited resolution of a map imposes restrictions on the navigation engine in both indoor and outdoor environments. To bridge the gap between the end position obtained from legacy navigation services and the real destination, we propose FollowMe, a "last-mile" navigation system to enable plug-and-play navigation in indoor and semi-outdoor environments. FollowMe exploits the ubiquitous, stable geomagnetic field and natural walking patterns to navigate the users to the same destination taken by an earlier traveler. Unlike existing localization and navigation systems, FollowMe is infrastructure-free, energy-efficient and cost-saving. We implemented FollowMe on smartphones, and evaluated it in a four-story campus building with a testing area of 2000m2. Our experimental results with 5 users show that 95% of spatial errors during navigation were 2m or less with at least 50% energy savings over a benchmark system.

Anomalies of the omnipresent earth magnetic (i.e., geomagnetic) field in an indoor environment, caused by local disturbances due to construction materials, give rise to noisy direction sensing that hinders any dead reckoning system. In this paper, we turn this unpalatable phenomenon into a favorable one. We present Magicol, an indoor localization and tracking system that embraces the local disturbances of the geomagnetic field. We tackle the low discernibility of the magnetic field by vectorizing consecutive magnetic signals on a per-step basis, and use vectors to shape the particle distribution in the estimation process. Magicol can also incorporate WiFi signals to achieve much improved positioning accuracy for indoor environments with WiFi infrastructure. We perform an in-depth study on the fusion of magnetic and WiFi signals. We design a two-pass bidirectional particle filtering process for maximum accuracy, and propose an on-demand WiFi scan strategy for energy savings. We further propose a compliant-walking method for location database construction that drastically simplifies the site survey effort. We conduct extensive experiments at representative indoor environments, including an office building, an underground parking garage, and a supermarket in which Magicol achieved a 90 percentile localization accuracy of 5 m, 1 m, and 8 m, respectively, using the magnetic field alone. The fusion with WiFi leads to 90 percentile accuracy of 3.5 m for localization and 0.9 m for tracking in the office environment. When using only the magnetism, Magicol consumes 9 × less energy in tracking compared to WiFi-based tracking.

The wireless rechargeable sensor network is a promising platform for long-term applications such as inventory management, supply chain monitoring, and so on. For these applications, sensor localization is one of the most fundamental challenges. Different from a traditional sensor node, a wireless rechargeable sensor has to be charged above a voltage level by the wireless charger in order to support its sensing, computation, and communication operations. In this work, we consider the scenario where a mobile charger stops at different positions to charge sensors and propose a novel localization design that utilizes the unique Time of Charge (TOC) sequences among wireless rechargeable sensors. Specifically, we introduce two efficient region dividing methods, Internode Division and Interarea Division , to exploit TOC differences from both temporal and spatial dimensions to localize individual sensor nodes. To further optimize the system performance, we introduce both an optimal charger stop planning algorithm for the single-sensor case and a suboptimal charger stop planning algorithm for the generic multisensor scenario with a provable performance bound. We have extensively evaluated our design by both testbed experiments and large-scale simulations. The experiment and simulation results show that by as less as five stops, our design can achieve sub-meter accuracy and the performance is robust under various system conditions.

Wireless rechargeable sensor network is a promising platform for long-term applications such as inventory management, supply chain monitoring and so on. For these applications, sensor localization is one of the most fundamental challenges. Different from traditional sensor node, wireless rechargeable sensor has to be charged above a voltage level by the wireless charger in order to support its sensing, computation and communication operations. In this work, we consider the scenario where a mobile charger stops at different positions to charge sensors, and propose a novel localization design that utilizes the unique Time of Charge (TOC) sequences among wireless rechargeable sensors. Specifically, we introduce two efficient region dividing methods, Inter-node Division and Inter-area Division, to exploit TOC differences from both temporal and spatial dimensions to localize individual sensor nodes. To further optimize the system performance, we introduce both an optimal charger stop planning algorithm for single sensor case and a suboptimal charger stop planning algorithm for the generic multisensor scenario with a provable performance bound. We have extensively evaluated our design by both testbed experiments and large-scale simulations. The experiment and simulation results show that by as less as 5 stops, our design can achieve sub-meter accuracy and the performance is robust under various system conditions.

Integrated with low-power micro-controllers and sensors, RFID-based wireless rechargeable sensor node is a very promising platform for applications such as inventory management, supply chain monitoring etc. Among other major research challenges, one of the most essential problems in such wireless rechargeable sensor networks is how to minimize the communication delay among RFID readers and RFID-based rechargeable nodes. While the existing works have mostly focused on the collision avoidance among RFID-based nodes, in this work we study an orthogonal approach which focuses on how to optimally plan the movement of the reader so as to minimize the communication delay in the network. To solve this problem, we introduce both an optimal solution for the linear reader movement pattern and an approximation solution for the generic two-dimensional reader move pattern with a provable approximation ratio. In addition, we also provide a solution for guaranteeing the quality of communication while minimizing the communication delay. We verify our observations through testbed experiments and extensively evaluate our design by both emulations and large-scale simulations. The results show our design can effectively reduce communication delay in wireless rechargeable sensor networks when compared with baseline solutions.

Access card authentication is critical and essential for many modern access control systems, which have been widely deployed in various government, commercial, and residential environments. However, due to the static identification information exchange among the access cards and access control clients, it is very challenging to fight against access control system breaches due to reasons such as loss, stolen or unauthorized duplications of the access cards. Although advanced biometric authentication methods such as fingerprint and iris identification can further identify the user who is requesting authorization, they incur high system costs and access privileges cannot be transferred among trusted users. In this work, we introduce a dynamic authentication with sensory information for the access control systems. By combining sensory information obtained from onboard sensors on the access cards as well as the original encoded identification information, we are able to effectively tackle the problems such as access card loss, stolen, and duplication. Our solution is backward-compatible with existing access control systems and significantly increases the key spaces for authentication. We theoretically demonstrate the potential key space increases with sensory information of different sensors and empirically demonstrate simple rotations can increase key space by more than 1,000,000 times with an authentication accuracy of 90 percent. We performed extensive simulations under various environment settings and implemented our design on WISP to experimentally verify the system performance.

Access card authentication is critical and essential for many modern access control systems, which have been widely deployed in various government, commercial and residential environments. However, due to the static identification information exchange among the access cards and access control clients, it is very challenging to fight against access control system breaches due to reasons such as loss, stolen or unauthorized duplications of the access cards. Although advanced biometric authentication methods such as fingerprint and iris identification can further identify the user who is requesting authorization, they incur high system costs and access privileges can not be transferred among trusted users. In this work, we introduce a sensory-data-enhanced authentication for access control systems. By combining sensory-data obtained from onboard sensors on the access cards as well as the original encoded identification information, we are able to effectively tackle the problems such as access card loss and stolen. Our solution is backward-compatible with existing access control systems and significantly increases the key spaces for authentication. We theoretically demonstrate the potential key space increases with simple sensor data and empirically demonstrate simple rotations can increase key space by more than 30, 000 times with an authentication accuracy of 95%. We performed extensive simulations under various environment settings and implemented our design on WISP to experimentally verify the system performance.

Wireless Sensor Networks have been used in many mobile applications such as wildlife tracking and participatory urban sensing. Because of the combination of high mobility and low-duty-cycle operations, it is a challenging issue to reduce discovery delay among mobile nodes, so that mobile nodes can establish connection quickly once they are within each other's vicinity. Existing discovery designs are essentially pair-wise based, in which discovery is passively achieved when two nodes are pre-scheduled to wake-up at the same time. In contrast, for the first time, this work reduces discovery delay significantly by proactively referring wake-up schedules among a group of nodes. Because proactive references incur additional overhead, we introduce a novel selective reference mechanism based on spatiotemporal properties of neighborhood and the mobility of the nodes. Our quantitative analysis indicates that the discovery delay of our group-based mechanism is significantly smaller than that of the pair-wise one. Our testbed experiments using 40 sensor nodes confirm our theoretical analysis, showing one order of magnitude reduction in discovery delay compared with traditional pair-wise methods with only 0.5%~8.8% increase in energy consumption.

Based on spatiotemporal properties of neighborhood and mobile properties of nodes in the networks, we propose Selective Reference Mechanism to trade off between the delay and overhead of neighbor discovery in low-duty-cycle WSNs. Extensive simulation and test-bed experiment confirm our theoretical analysis, showing as much as 35.4% increase in discovery probability, 38.6% reduction in discovery delay and 27.7% reduction in total energy consumption.

To bridge the gap between insufficiency of existing proximity authentication solutions and the increasing demand of high security guarantee for access control systems, we develope a WISP-based access control system combing electronic and mechanical authentication methods. In our authentication, encryption complexity is changeable and trusted users can share privileges with each other. During experiments, our system has achieved 95% authentication accuracy rate with up to 3 different users.