Open-set semi-supervised learning (OSSL) leverages unlabeled data containing both in-distribution (ID) and unknown out-of-distribution (OOD) samples, aiming simultaneously to improve closed-set accuracy and detect novel OOD instances. Existing methods either discard valuable information from uncertain samples or force-align every unlabeled sample into one or a few synthetic “catch-all” representations, resulting in geometric collapse and overconfidence on only seen OODs. To address the limitations, we introduce selective non-alignment, adding a novel “skip” operator into conventional pull and push operations of contrastive learning. Our framework, SkipAlign, selectively skips alignment (pulling) for low-confidence unlabeled samples, retaining only gentle repulsion against ID prototypes. This approach transforms uncertain samples into a pure repulsion signal, resulting in tighter ID clusters and naturally dispersed OOD features. Extensive experiments demonstrate that SkipAlign significantly outperforms state-of-the-art methods in detecting unseen OOD data without sacrificing ID classification accuracy.

As AI moves from data centers to robots and wearables, scaling ever-larger models becomes insufficient. Physical AI operates under tight latency, energy, privacy, and reliability constraints, and its performance depends not only on model capacity but also on how signals are acquired through controllable sensors in dynamic environments. We present Artificial Tripartite Intelligence (ATI), a bio-inspired, sensor-first architectural contract for physical AI. ATI is tripartite at the systems level: a Brainstem (L1) provides reflexive safety and signal-integrity control, a Cerebellum (L2) performs continuous sensor calibration, and a Cerebral Inference Subsystem spanning L3/L4 supports routine skill selection and execution, coordination, and deep reasoning. This modular organization allows sensor control, adaptive sensing, edge-cloud execution, and foundation model reasoning to co-evolve within one closed-loop architecture, while keeping time-critical sensing and control on device and invoking higher-level inference only when needed. We instantiate ATI in a mobile camera prototype under dynamic lighting and motion. In our routed evaluation (L3-L4 split inference), compared to the default auto-exposure setting, ATI (L1/L2 adaptive sensing) improves end-to-end accuracy from 53.8% to 88% while reducing remote L4 invocations by 43.3%. These results show the value of co-designing sensing and inference for embodied AI.

Variations in environment and sensor (ES) conditions— lighting, ISO, shutter, and aperture—cause domain shifts that degrade visual recognition. While a recent robustness benchmark, ImageNet-ES, shows that these shifts differ from conventional augmentations, collecting physically recaptured data is costly and hard to scale. We present CycleGAN-ES, a per-condition unpaired translation framework that simulates ES-style variations from a small set of real targets. Trained on Tiny-ImageNet and ImageNet-ES domain pairs with as few as 200 images per target domain and minimal tuning, CycleGAN-ES produces a synthetic counterpart, ImageNet-sES (IN-sES). The generated images exhibit high-fidelity ES effects both qualitatively and quantitatively, reproducing characteristic exposure and noise behaviors (e.g., highlight clipping at long exposure and increased high-ISO noise). In benchmark evaluations, augmenting training with ImageNet-sES improves robustness to ES shifts on ImageNet-ES, achieves complementary gains when combined with standard augmentation strategies, and transfers to other corruption domains such as ImageNet-C. The learned translators further transfer to new datsets (e.g., CIFAR-100) without retraining. To our knowledge, this is the first systematic ES-simulation study anchored to real recaptures at ImageNet scale, offering a practical, scalable route for injecting physically grounded ES diversity into training pipelines and enabling broader real-world robustness evaluation.

The rising prevalence of colorectal cancer necessitates early and accurate optical diagnosis of colorectal polyps. Despite advances in Computer-Aided Diagnosis (CAD) systems, challenges like data variability and inconsistent clinical performance hinder their widespread use. To address these limitations, we propose ColonOOD, an integrated CAD system for polyp localization, uncertainty-aware polyp classification, and Out-of-Distribution (OOD) polyp detection during colonoscopy. ColonOOD ensures robust classification of adenomatous, hyperplastic, and OOD polyps while providing calibrated uncertainty scores to support clinical decisions. Extensive evaluations across four medical centers and two public datasets demonstrate ColonOOD’s strong performance, achieving up to 79.69% classification and 75.53% OOD detection accuracy. This system offers reliable insights for endoscopists, marking a significant step toward broader clinical adoption of automated diagnostic tools in colorectal cancer care.

The growing use of 3D point cloud data in autonomous vehicles (AVs) has raised serious privacy concerns, particularly due to the sensitive information that can be extracted from 3D data. While model inversion attacks have been widely studied in the context of 2D data, their application to 3D point clouds remains largely unexplored. To fill this gap, we present the first in-depth study of model inversion attacks aimed at restoring 3D point cloud scenes. Our analysis reveals the unique challenges, the inherent sparsity of 3D point clouds and the ambiguity between empty and non-empty voxels after voxelization, which are further exacerbated by the dispersion of non-empty voxels across feature extractor layers. To address these challenges, we introduce ConcreTizer, a simple yet effective model inversion attack designed specifically for voxel-based 3D point cloud data. ConcreTizer incorporates Voxel Occupancy Classification to distinguish between empty and non-empty voxels and Dispersion-Controlled Supervision to mitigate non-empty voxel dispersion. Extensive experiments on widely used 3D feature extractors and benchmark datasets, such as KITTI and Waymo, demonstrate that ConcreTizer concretely restores the original 3D point cloud scene from disrupted 3D feature data. Our findings highlight both the vulnerability of 3D data to inversion attacks and the urgent need for robust defense strategies.

Domain shift remains a persistent challenge in deep-learning-based computer vision, often requiring extensive model modifications or large labeled datasets to address. Inspired by human visual perception, which adjusts input quality through corrective lenses rather than over-training the brain, we propose Lens, a novel camera sensor control method that enhances model performance by capturing high-quality images from the model’s perspective, rather than relying on traditional human-centric sensor control. Lens is lightweight and adapts sensor parameters to specific models and scenes in real-time (i.e., test-time input adaptation). At its core, Lens utilizes VisiT, a training-free, model-specific quality indicator that evaluates individual unlabeled samples at test time using confidence scores, without additional adaptation costs. To validate Lens, we introduce ImageNet-ES Diverse, a new benchmark dataset capturing natural perturbations from varying sensor and lighting conditions. Extensive experiments on both ImageNet-ES and our new ImageNet-ES Diverse show that Lens significantly improves model accuracy across various baseline schemes for sensor control and model modification, while maintaining low latency in image captures. Lens effectively compensates for large model size differences and integrates synergistically with model improvement techniques. Our code and dataset are available at github.com/Edw2n/Lens.git.

Sleep is essential for health and well-being, yet the gold-standard method for sleep stage assessment, polysomnography (PSG), requires overnight monitoring with numerous wired sensors and manual scoring, limiting its scalability and comfort in real-world settings. We present ViSS, the first end-to-end deep learning framework for fully non-contact sleep staging directly from raw infrared video. To capture the dual temporal structure of sleep, ViSS integrates per-epoch motion encoding and long-range sequence modeling in a compact hierarchical architecture, without relying on intermediate physiological signals. Evaluated on the largest infrared video dataset to date, including both healthy individuals and patients, ViSS achieves 80% accuracy and 0.78 macro F1-score using fewer than 7 million parameters. These results establish infrared video as a viable modality for automated sleep staging and highlight the potential of ViSS for scalable, privacy-preserving home-based assessment.

Current AI advances largely rely on scaling neural models and expanding training datasets to achieve generalization and robustness. Despite notable successes, this paradigm incurs significant environmental, economic, and ethical costs, limiting sustainability and equitable access. Inspired by biological sensory systems, where adaptation occurs dynamically at the input (e.g., adjusting pupil size, refocusing vision)—we advocate for adaptive sensing as a necessary and foundational shift. Adaptive sensing proactively modulates sensor parameters (e.g., exposure, sensitivity, multimodal configurations) at the input level, significantly mitigating covariate shifts and improving efficiency. Empirical evidence from recent studies demonstrates that adaptive sensing enables small models (e.g., EfficientNet-B0) to surpass substantially larger models (e.g., OpenCLIP-H) trained with significantly more data and compute. We (i) outline a roadmap for broadly integrating adaptive sensing into real-world applications spanning humanoid, healthcare, autonomous systems, agriculture, and environmental monitoring, (ii) critically assess technical and ethical integration challenges, and (iii) propose targeted research directions, such as standardized benchmarks, real-time adaptive algorithms, multimodal integration, and privacy-preserving methods. Collectively, these efforts aim to transition the AI community toward sustainable, robust, and equitable artificial intelligence systems.

Computer vision applications predict on digital images acquired by a camera from physical scenes through light. However, conventional robustness benchmarks rely on perturbations in digitized images, diverging from distribution shifts occurring in the image acquisition process. To bridge this gap, we introduce a new distribution shift dataset, ImageNet-ES, comprising variations in environmental and camera sensor factors by directly capturing 202k images with a real camera in a controllable testbed. With the new dataset, we evaluate out-of-distribution (OOD) detection and model robustness. We find that existing OOD detection methods do not cope with the covariate shifts in ImageNet-ES, implying that the definition and detection of OOD should be revisited to embrace real-world distribution shifts. We also observe that the model becomes more robust in both ImageNet-C and -ES by learning environment and sensor variations in addition to existing digital augmentations. Lastly, our results suggest that effective shift mitigation via camera sensor control can significantly improve performance without increasing model size. With these findings, our benchmark may aid future research on robustness, OOD, and camera sensor control for computer vision. Our code and dataset are available at https://github.com/Edw2n/ImageNet-ES.

Virtual try-on (VTO) superimposes clothing over user image or video, enhancing online shopping experience. On-device VTO can preserve user privacy but most VTO techniques cannot be run on resource-constrained devices due to excessive computation overhead. In this demo, we demonstrate a novel Android application for on-device video VTO referring to MIRROR, the state-of-the-art mobile VTO system. The application minimizes video generation time by splitting the process into two phases, achieving 0.76 minutes to convert 10-second-long video on Galaxy S24 Ultra. Our application was evaluated as 78.5 score (above average) in SUS usability test. A companion video is provided at: https://youtu.be/YTExc8W5BzM

Obstructive sleep apnea (OSA) is a prevalent sleep disorder, affecting approximately one billion individuals globally. In this study, we aim to address the limitations of Polysomnography (PSG), the gold standard for OSA diagnosis, by developing SlAction, a non-intrusive system that utilizes infrared videos for OSA detection in daily sleep settings. Considering the privacy-sensitive nature of sleep videos, SlAction is designed to analyze data directly on the camera-capturing device, eliminating the need to transmit video data to a server. With the collaboration of clinical experts, we extensively analyze the largest dataset worldwide that we collected, establishing correlations between OSA events and human motions during sleep. Our novel approach achieved an OSA prediction performance with an F1 score of 0.88. Notably, even when running on a low-spec CPU, our SlAction operates approximately 75 times faster than previous work evaluated on high-performance GPU servers.

While federated learning leverages distributed client resources, it faces challenges due to heterogeneous client capabilities. This necessitates allocating models suited to clients' resources and careful parameter aggregation to accommodate this heterogeneity. We propose HypeMeFed, a novel federated learning framework for supporting client heterogeneity by combining a multi-exit network architecture with hypernetwork-based model weight generation. This approach aligns the feature spaces of heterogeneous model layers and resolves per-layer information disparity during weight aggregation. To practically realize HypeMeFed, we also propose a low-rank factorization approach to minimize computation and memory overhead associated with hypernetworks. Our evaluations on a real-world heterogeneous device testbed indicate that HypeMeFed enhances accuracy by 5.12% over FedAvg, reduces the hypernetwork memory requirements by 98.22%, and accelerates its operations by 1.86X compared to a naive hypernetwork approach. These results demonstrate HypeMeFed's effectiveness in leveraging and engaging heterogeneous clients for federated learning.

Semi-supervised Learning (SSL) has received increasing attention in autonomous driving to reduce the enormous burden of 3D annotation. In this paper, we propose UpCycling, a novel SSL framework for 3D object detection with zero additional raw-level point cloud: learning from unlabeled de-identified intermediate features (i.e., "smashed" data) to preserve privacy. Since these intermediate features are naturally produced by the inference pipeline, no additional computation is required on autonomous vehicles. However, generating effective consistency loss for unlabeled feature-level scene turns out to be a critical challenge. The latest SSL frameworks for 3D object detection that enforce consistency regularization between different augmentations of an unlabeled raw-point scene become detrimental when applied to intermediate features. To solve the problem, we introduce a novel combination of hybrid pseudo labels and feature-level Ground Truth sampling (F-GT), which safely augments unlabeled multi-type 3D scene features and provides high-quality supervision. We implement UpCycling on two representative 3D object detection models: SECOND-IoU and PV-RCNN. Experiments on widely-used datasets (Waymo, KITTI, and Lyft) verify that UpCycling outperforms other augmentation methods applied at the feature level. In addition, while preserving privacy, UpCycling performs better or comparably to the state-of-the-art methods that utilize raw-level unlabeled data in both domain adaptation and partial-label scenarios.

Running deep learning models on resource-constrained edge devices has drawn significant attention due to its fast response, privacy preservation, and robust operation regardless of Internet connectivity. While these devices already cope with various intelligent tasks, the latest edge devices that are equipped with multiple types of low-power accelerators (i.e., both mobile GPU and NPU) can bring another opportunity; a task that used to be too heavy for an edge device in the single-accelerator world might become viable in the upcoming heterogeneous-accelerator world. To realize the potential in the context of 3D object detection, we identify several technical challenges and propose PointSplit, a novel 3D object detection framework for multi-accelerator edge devices that addresses the problems. Specifically, our PointSplit design includes (1) 2D semantics-aware biased point sampling, (2) parallelized 3D feature extraction, and (3) role-based group-wise quantization. We implement PointSplit on TensorFlow Lite and evaluate it on a customized hardware platform comprising both mobile GPU and EdgeTPU. Experimental results on representative RGB-D datasets, SUN RGB-D and Scannet V2, demonstrate that PointSplit on a multi-accelerator device is 24.7 × faster with similar accuracy compared to the full-precision, 2D-3D fusion-based 3D detector on a GPU-only device.

Dependability is an essential part of software engineering in the industry, which is the reason why most companies hire professional quality assurance (QA) engineers. Despite the presence of separate QA engineers, however, software developers are still required to perform basic QA processes by themselves before submitting their new software for peer review. When it comes to mobile system software, however, the basic QA process requires significant time and effort for writing different test scripts for various types of mobile devices, which degrades the productivity of developers. In this paper, we propose ScriptPainter, an easy-to-use automatic test script generator that enables a mobile phone to self-convert vision information on its screen into a test program; developers simply record a video while drawing a test scenario on a mobile touchscreen and ScriptPainter detects and tracks cursor locations in the video for generating a proper script. For accurate cursor detection and tracking with modest computation, ScriptPainter com-prises a set of techniques, such as frame sampling, deep neural network (DNN)-based detection, hough transform-based tracking with bounding box adaptation. ScriptPainter is evaluated on multiple Android smartphones, which shows that it analyzes a video 3.8x faster than pure DNN-based detection. Experiments with human developers show that ScriptPainter speeds up their job 8.3 times while providing even better accuracy compared to manual scripting.

Time Slotted Channel Hopping (TSCH), defined in the IEEE 802.15.4e standard requires complicated slot scheduling to enjoy its collision-free multihop communication capabilities fully. Recently, several autonomous scheduling algorithms that allocate slots without central control and additional control message exchange have been devised. However, autonomous algorithms are traffic oblivious, assigning the same number of slots regardless of traffic demands. Undifferentiated resource allotment is identified as the root cause that impairs the performance of TSCH networks. To address the issue, we propose A3, an autonomous and adaptive slot allocation scheme that adjusts the number of slots per slotframe according to varying traffic loads. A key component of A3 is receiver-side load estimation in real-time without any explicit control message exchange. A3 can be combined with any autonomous scheduling protocols, such as Orchestra and ALICE. With extensive evaluation on a large public testbed comprising 62 nodes, we verify that A3 significantly enhances autonomous scheduling algorithms' performance. It improves throughput by more than twice, PDR (Packet Delivery Ratio) by more than six times, and reduces the latency by ten times.

This work investigates a run-time scheduling system for Bluetooth Low Energy (BLE). The Bluetooth specifications propose control of the anchor point (i.e., start time of each connection event) by changing the WinOffset parameter of the control protocol data unit (PDU), which can be very useful when the master schedules multiple connections simultaneously. Since we had no access to the controller (i.e., which wraps the physical layer and link layer of Bluetooth), we could not figure out how WinOffset works or use it the way we wanted. However, with Zephyr's advent, there is room to modify the controller of BLE, and we can adjust anchor points as we want using WinOffset. To the best of our knowledge, this work is the first systematic study of this regime in the research community. We experimentally study how commercial devices actually schedule multiple connections and how this inefficient scheduling degrades BLE's performance. Based on the preliminary study, we propose BLEX that dynamically adjusts anchor points of multiple slaves to satisfy quality of service (QoS) requirements without violating the Bluetooth specifications or intervening the BLE host (i.e., software-defined higher layer). Through extensive performance evaluation on off-the-shelf BLE chips, we show that BLEX provides stable performance regardless of the traffic requirements of applications and shows significantly reduced QoS failures compared to the state-of-the-art scheduling schemes for BLE.

The upsurge of mobile traffic volume has led to dense deployment of IEEE 802.11 wireless local area networks (WLANs). To enhance spectral efficiency in dense WLAN deployments, a new 802.11 amendment, namely 802.11ax, introduces overlapping basic service set packet detection-based spatial reuse (OBSS PD-based SR) operation. Although the OBSS PD-based SR operation enables more simultaneous transmissions among OBSSs, whether it improves network performance is highly uncertain and depends on the mutual interference among the simultaneous transmission links. In this paper, we propose a link-aware SR scheme, named LSR, that facilitates simultaneous transmissions among OBSSs considering the mutual interference. LSR efficiently exploits SR transmission opportunities by selecting the appropriate link for the SR transmission while sufficiently protecting the ongoing transmission. Moreover, the proposed method allows BSSs to coordinate with each other in a distributed manner without incurring additional control message exchange. Through extensive ns-3 simulation, we demonstrate that LSR significantly improves the network performance compared with legacy 802.11 and the OBSS PD-based SR operation.

To avoid access to PFS, dedicated BB allocation is preferred despite of severe BB underutilization. Recently, new all-flash HPC storage systems with integrated BB and PFS are proposed, which speed up access to PFS. For this reason, we adopt BB over-subscription allocation method by allowing HPC applications to use BB only for I/O phase for improving BB utilization. Unfortunately, BB over-subscription aggravates I/O interference and demotion overhead from BB to PFS, resulting in degraded performance. To minimize the performance degradation, we develop an I/O scheduler to prevent I/O congestion and a new transparent data management system based on checkpoint/restart characteristics of HPC applications. With the proposed approach, not only the BB utilization can be improved, but also high performance of applications is achieved. In our experiments, we find that BB utilization is improved at least 2.2×, and more stable and higher checkpoint performance is guaranteed compared to other approaches. Besides, we achieve up to 96.4% hit ratio of restart requests on BB and up to 3.1× higher restart performance than others.

As the emerging Internet of Things (IoT) devices and applications flourish, demand for reliable and energy-efficient low-power wireless network protocols is surging. For this purpose, IEEE 802.15.4 standardized time-slotted channel hopping (TSCH), a promising and viable link-layer solution that has shown outstanding performance achieving over 99% reliability with low duty-cycles. However, it lacks one thing, flexibility. It is not adaptable to a wide variety of applications with varying traffic load and unpredictable routing topology due to its static timeslot scheduling. To this end, we propose OST, an On-demand Scheduling scheme for TSCH with traffic-awareness. In OST, each node dynamically self-adjusts the frequency of timeslots at run time according to time-varying traffic intensity. Moreover, it features on-demand resource allocation to handle bursty/queued packets in a timely manner. By doing so, OST aims to minimize its energy consumption while guaranteeing reliable packet delivery. We evaluate OST on a large-scale 72-node testbed, demonstrating that it achieves improvement of 60% in reliability and 52% in energy-efficiency compared to the state of the art.

The increasing congestion of the RF spectrum is a key challenge for low-power wireless networks using concurrent transmissions. The presence of radio interference can indeed undermine their dependability, as they rely on a tight synchronization and incur a significant overhead to overcome packet loss. In this paper, we present Harmony, a new data collection protocol that exploits the benefits of concurrent transmissions and embeds techniques to ensure a reliable and timely packet delivery despite highly congested channels. Such techniques include, among others, a data freezing mechanism that allows to successfully deliver data in a partitioned network as well as the use of network coding to shorten the length of packets and increase the robustness to unreliable links. Harmony also introduces a distributed interference detection scheme that allows each node to activate various interference mitigation techniques only when strictly necessary, avoiding unnecessary energy expenditures while finding a good balance between reliability and timeliness. An experimental evaluation on real-world testbeds shows that Harmony outperforms state-of-the-art protocols in the presence of harsh Wi-Fi interference, with up to 50% higher delivery rates and significantly shorter end-to-end latencies, even when transmitting large packets.

We present PC-RPL , a transmission power-controlled IPv6 routing protocol for low-power and lossy wireless networks that significantly improves the end-to-end packet delivery performance under heavy traffic compared to the standard RPL. We show through actual design, implementation, and experiments that a multihop wireless network can achieve better throughput and routing stability when transmission power and routing topology are “jointly and adaptively” controlled. Our experiments show that the predominant “fixed and uniform” transmission power strategy with “link quality and hop distance”–based routing topology construction (i.e., RPL) loses significant bandwidth due to hidden terminal and load imbalance problems. We design an adaptive and distributed control mechanism for transmission power and routing topology, named PC-RPL , on top of the standard RPL routing protocol for hidden terminal mitigation and load balancing. We implement PC-RPL on real embedded devices and evaluate its performance on a 49-node multihop testbed. PC-RPL reduces total end-to-end packet losses by approximately sevenfold without increasing hop distance compared to RPL with the highest transmission power, resulting in 17% improvement in aggregate bandwidth and 64% improvement for the worst-case node by successfully alleviating both hidden terminal and load imbalance problems.

Although low-power lossy network (LLN), at its early stage, commonly used asynchronous link layer protocols for simple operation on resource-constrained nodes, development of embedded hardware and time synchronization technologies made Time-Slotted Channel Hopping (TSCH) viable in LLN (now part of IEEE 802.15.4e standard). TSCH has the potential to be a link layer solution for LLN due to its resilience to wireless interference (e.g., WiFi) and multi-path fading. However, its slotted operation incurs non-trivial cell scheduling overhead: two nodes should wake up at a time-frequency cell together to exchange a packet. Efficient cell scheduling in dynamic multihop topology in wireless environments has been an open issue, preventing TSCH's wide adoption in practice. This work introduces ALICE, a novel autonomous link-based cell scheduling scheme which allocates a unique cell for each directional link (a pair of nodes and traffic direction) by closely interacting with the routing layer and using only local information, without any additional communication overhead. We implement ALICE on Contiki and evaluate its effectiveness on the IoT-LAB public testbed with 68 nodes. ALICE generally outperforms Orchestra (the state-of-the-art method) and even more so under heavy traffic and high node density, increasing throughput by 2 times with 98.3% reliability and reducing latency by 70%, route changes by 95%, and radio duty cycle by 35%. ALICE can serve as an autonomous scheduling framework, which paves the way for TSCH-based LLN to go on.

This paper presents MARVEL, a mobile augmented reality (MAR) system which provides a notation display service with imperceptible latency (<100 ms) and low energy consumption on regular mobile devices. In contrast to conventional MAR systems, which recognize objects using image-based computations performed in the cloud, MARVEL mainly utilizes a mobile device's local inertial sensors for recognizing and tracking multiple objects, while computing local optical flow and offloading images only when necessary. We propose a system architecture which uses local inertial tracking, local optical flow, and visual tracking in the cloud synergistically. On top of that, we investigate how to minimize the overhead for image computation and offloading. We have implemented and deployed a holistic prototype system in a commercial building and evaluate MARVEL's performance. The efficient use of a mobile device's capabilities lowers latency and energy consumption without sacrificing accuracy.

The emergence of low-power 32-bit Systems-on-Chip (SoCs), which integrate a 32-bit MCU, radio, and flash, presents an opportunity to re-examine design points and trade-offs at all levels of the system architecture of networked sensors. To this end, we develop a post-SoC/32-bit design point called Hamilton, showing that using integrated components enables a ~$7 core and shifts hardware modularity to design time. We study the interaction between hardware and embedded operating systems, identifying that (1) post-SoC motes provide lower idle current (5.9 μA) than traditional 16-bit motes, (2) 32-bit MCUs are a major energy consumer (e.g., tick increases idle current >50 times), comparable to radios, and (3) thread-based concurrency is viable, requiring only 8.3 μs of context switch time. We design a system architecture, based on a tickless multithreading operating system, with cooperative/adaptive clocking, advanced sensor abstraction, and preemptive packet processing. Its efficient MCU control improves concurrency with ~30% less energy consumption. Together, these developments set the system architecture for networked sensors in a new direction.

Although TCP has widespread adoption in the Internet, wireless sensor networks (WSNs) generally use simpler UDP-based protocols. The few existing TCP implementations for sensor network operating systems do not support all of the features of TCP. We present a full-scale TCP implementation for sensor networks, called TCPlp, based on the TCP protocol logic of the FreeBSD Operating System. Our implementation demonstrates that full-scale TCP can run within the resource constraints of a modern WSN platform, and serves as a vehicle to explore the benefits of using a full TCP stack in the WSN setting. We showcase TCPlp via three applications of TCP: (1) reliable data collection in the context of an application, (2) an interactive configuration/debug shell, and (3) a mote-based web server.

We show that a multihop wireless network can achieve better bandwidth and routing stability when transmission power and routing topology are jointly and adaptively controlled. Our experiments show that the predominant 'fixed and uniform' transmission power strategy with 'link quality and hop distance'-based routing topology construction loses significant bandwidth due to hidden terminal and load imbalance problems. We design an adaptive and distributed control mechanism for transmission power and routing topology, PCRPL, within the standard RPL routing protocol. We implement PC-RPL on real embedded devices and evaluate its performance on a 49-node multihop testbed. PC-RPL reduces total end-to-end packet losses ~7-fold without increasing hop distance compared to RPL with the highest transmission power, resulting in 17% improvement in aggregate bandwidth and 64% for the worst-case node.

For over a decade, multihop low power and lossy networks (LLNs) have mainly focused on delivering upward traffic from individual nodes to the servers for supporting remote monitoring applications. However, as part of Internet of Things(IoT), LLN applications and traffic patterns are being diversified. In this paper, we address performance issues of LLNs when delivering various traffic patterns. Specifically, we experimentally show that LLNs suffer from unreliable routes when downward traffic is dominant since current routing protocols do not utilize downward traffic for link quality update. To tackle the problem, we design a novel mechanism, named ORPL-DT, that updates link quality information by using both upward and downward traffic, and implement it on top of ORPL (defacto IPv6 opportunistic routing protocol). We evaluate its performance on a 31-node multihop testbed, showing that ORPL-DT improves performance in terms of packet delivery ratio, control overhead, and radio duty-cycle.

Bluetooth Low Energy (BLE) is one of the widely used low power wireless protocols due to its simplicity and low energy consumption. The BLE standard is being developed further to support a wide range of applications spanning smart homes, wearables, and myriad appliances as part of IoT (Internet of Things). These new applications bring forth an important challenge: connection maintenance with low energy consumption in dynamic channel environments. In this paper, we investigate effective solutions to this technical challenge. First, we show using experiments that the current design of using fixed connection intervals incurs significant performance degradation under dynamic link conditions. To overcome the problem, we mathematically find an optimal connection interval that minimizes energy consumption while maintaining connectivity for a given link condition. Then, we design a simple yet effective connection interval adaptation mechanism for BLE, named CABLE. We implement the proposed solution on real embedded devices, and using extensive testbed experiments and simulation verify that CABLE leads to significant performance improvement while providing resilient connectivity in dynamic link environments.

RPL is an IPv6 routing protocol for low-power and lossy networks (LLNs) designed to meet the requirements of a wide range of LLN applications including smart grid AMIs, industrial and environmental monitoring, and wireless sensor networks. RPL allows bidirectional end-to-end IPv6 communication on resource constrained LLN devices, leading to the concept of the Internet of Things (IoT) with thousands and millions of devices interconnected through multihop mesh networks. In this article, we investigate the load balancing and congestion problem of RPL. Specifically, we show that most of the packet losses under heavy traffic are due to congestion, and a serious load balancing problem appears in RPL in terms of routing parent selection. To overcome this problem, this article proposes a simple yet effective queue utilization based RPL (QU-RPL) that achieves load balancing and significantly improves the end-to-end packet delivery performance compared to the standard RPL. QU-RPL is designed for each node to select its parent node considering the queue utilization of its neighbor nodes as well as their hop distances to an LLN border router (LBR). Owing to its load balancing capability, QURPL is very effective in lowering queue losses and increasing the packet delivery ratio. We implement QU-RPL on a low-power embedded platform, and verify all of our findings through experimental measurements on a real testbed of a multihop LLN over IEEE 802.15.4. We present the impact of each design element of QU-RPL on performance in detail, and also show that QU-RPL reduces the queue loss by up to 84 percent and improves the packet delivery ratio by up to 147 percent compared to the standard RPL.

In this paper, we consider a protocol architecture that enables IPv6 routing protocol for low power and lossy networks (RPL) to run on top of Bluetooth Low Energy (BLE), aiming to provide a BLE-based multi-hop IoT network. In our approach to RPL over BLE, we propose to use both of advertising and data channels of BLE to create synergistic effects between RPL and BLE to jointly achieve high energy efficiency and reliable multi-hop routing. We design an adaptation layer between BLE and RPL (ALBER) which tightly couples RPL and BLE operations together. Specifically, ALBER provides RPL control message broadcast through BLE, RPL routing metric calculation that reflects BLE link quality, and routing table update that incorporates BLE connection management. We implement ALBER in a Linux kernel to realize RPL over BLE and compare its performance with that of RPL over IEEE 802.15.4 on a multi-hop testbed network. The performance results show that our architecture is not only feasible but also provides almost perfect packet delivery performance (↑100%) and reduces dutycycle up to 32% compared to RPL over IEEE 802.15.4 under varying link dynamics. Our research shows that RPL over BLE is a promising approach which can increase the utility and impactof BLE across different application domains.

We present a transmission power control scheme for RPL, the IPv6 Routing Protocol for Low-power lossy network [9], that controls the routing topology to achieve load balancing in a low-power multihop wireless network. We show that higher-than-required transmission power results in congestion and load balancing problems under heavy traffic, and transmission power cannot be optimized for reliable packet delivery when equal power is used by all nodes. To address these issues, we propose an adaptive and non-uniform transmission power controlled RPL, called PC-RPL, that significantly improves the end-to-end packet delivery performance compared to the standard RPL.

In low-power wireless networks, maintaining multihop connectivity is considered effective in constructing communication routes between individual nodes to a gateway. Since sensor networks are typically used for data collection, multihop routing protocols are designed to find routes optimal in upward directions. As sensor networks become widely applied to diverse applications, efficient downward traffic delivery also becomes important. To achieve this, we consider an asymmetric transmission power-based network (APN), where a power-supplied gateway uses high-power radios to cover the entire network via single-hop transmission, whereas common nodes use low-power transmissions. For effective APN operations, we propose a single-hop downlink protocol (SHDP) that consists of direct downlink transmission, local acknowledgment, neighbor forwarding, and contention resolution among the destination’s neighbors. We evaluate SHDP through mathematical analysis, simulations, and testbed experiments. Our proposal outperforms other competitive multihop routing protocols. Specifically, SHDP shows high packet delivery performance and lowers the duty cycle greatly while reducing the packet transmission overhead by >50%.

RPL is an IPv6 routing protocol for low-power and lossy networks (LLNs) designed to meet the requirements of a wide range of LLN applications including smart grid AMIs, industrial and environmental monitoring, and wireless sensor networks. RPL allows bi-directional end-to-end IPv6 communication on resource constrained LLN devices, leading to the concept of the Internet of Things (IoT) with thousands and millions of devices interconnected through multihop mesh networks. In this paper, we investigate the load balancing and congestion problem of RPL. Specifically, we show that most of packet losses under heavy traffic are due to congestion, and a serious load balancing problem exists in RPL in terms of routing parent selection. To overcome this problem, this paper proposes a simple yet effective queue utilization based RPL (QU-RPL) that significantly improves end-to-end packet delivery performance compared to the standard RPL. QU-RPL is designed for each node to select its parent node considering the queue utilization of its neighbor nodes as well as their hop distances to an LLN border router (LBR). Owing to its load balancing capability, QU-RPL is very effective in lowering the queue losses and increasing the packet delivery ratio. We verify all our findings through experimental measurements on a real testbed of a multihop LLN over IEEE 802.15.4.

Updating price tags in a large-scale market is a recurrent task, still performed manually in most markets. Given that human-errors can easily lead to customer complaints and accounting inaccuracies, the ability to autonomously reconfigure price tags can be of significant benefit. With the introduction of low-power display techniques such as electronic ink, applications of enabling electronic, wirelessly reconfigurable price tags show potential for future deployment. In this work, we examine networking architectures that can be applied in such scenarios. Through a series of preliminary pilot studies in an actual supermarket, we show that the performance of existing protocols are not ready to overcome the unique challenges of busy market environments. We identify underlying technical challenges and propose MarketNet, an asymmetric transmission power-based system designed for densely populated, obstacle-rich, downwards traffic-oriented environments. We evaluate MarketNet in a large indoor mar- ket visited by 5000+ customers per day. Our results show that MarketNet addresses the challenges of the target application and environment, while achieving higher packet delivery performance with noticeably lower radio duty-cycles than existing protocols such as RPL and SHDP.

In this demo, we present interoperability between Bluetooth Low Energy (BLE) and IPv6 routing protocol for low power and lossy networks (RPL). To make the operation of RPL over BLE feasible, we design an adaptation layer between BLE and RPL, termed ALBER. Specifically, we develop three technical features in ALBER which enable BLE to be combined with RPL. First, it broadcasts RPL control messages through BLE using advertising channels with low energy consumption. Second, it updates RPL routing table considering connection management of BLE. Lastly, it estimates link quality based on round trip time of link layer ping packets to provide routing metric for RPL. We implement our ALBER on Linux kernel. This demo will present an operation example of RPL over BLE using ALBER in a small scale multi-hop topology, where each node comprises a Raspberry Pi platform and a BLE dongle.

In this paper, we consider the use of an electric-supplied coordinator exploiting much higher transmission power than battery-supplied nodes in low power and lossy networks (LLNs). Since the coordinator can transmit via one hop instead of multiple hops over downlink, it is possible to reduce the communication overhead significantly. To take this advantage, we propose a single hop downlink protocol (SHDP) which comprises direct downlink transmission, local acknowledgement, neighbor forwarding, and mitigation of forwarding contention. Finally, the performance of the proposed SHDP is mathematically analyzed and evaluated by computer simulation, showing significant performance improvement over conventional multi-hop routing when applied to LLNs.

In wireless networks, frame retransmission is used to increase reliability with the use of additional wireless resources. Recently, various approaches have been proposed to reduce the overhead for retransmission. However, they still suffer from performance degradation due to unpredictable wireless channel variation during a frame transmission in fast-fading environments. In this paper, we propose an in-frame rate adaptation (InFRA) scheme as a solution for mitigating the retransmission overhead. In InFRA, a receiver estimates the channel SNR of incoming symbols and feeds it back to the sender. According to the feedback information, the sender adaptively changes its symbol transmission rate during a frame transmission. In this way, InFRA significantly increases the reliability of frame transmission even when the channel state changes within a frame transmission. To evaluate the effectiveness of InFRA, we mathematically analyze its performance and compare it with the existing schemes. Finally, our simulation results show that InFRA achieves significant throughput improvements over the other competitive schemes in fast fading channel environments.

In massive multiple-input multiple-out (M-MIMO) systems, conventional sounding schemes may suffer from pilot contamination of cell edge users or a lowered number of serviced users in a multi-cell scenario. In this paper, we propose a partial sounding resource reuse (PSRR) method which aims to seamlessly guarantee the quality of service (QoS) of mobile users by mitigating the pilot contamination as well as minimize the reduction in the number of serviced users. To this end, the PSRR divides each cell area into center and edge areas, and partially reuses sounding resources among users in neighboring edge areas. We use a Markov chain model to analyze the performance of the PSRR. Then we evaluate the accuracy of our analysis through simulations, and show that the PSRR considerably improves QoS performance over the conventional schemes.

Data gathering is one of the most popular applications in multi-hop wireless sensor networks. Since resources are limited, it is important to efficiently allocate the resource while providing desired performance. In this paper, we consider resource allocation in the uplink of multi-hop wireless networks with the use of data buffers of finite size in the presence of transmission error. We first analyze the probability of buffer occupation of each router and then derive the outage probability associated with buffer overflow and transmission latency. Exploiting the analytic results, we design a resource allocation algorithm that minimally utilizes the resource in multi-hop wireless sensor networks, while guaranteeing desired performance. Finally, the proposed scheme is verified by computer simulation.

In massive multiple-input multiple-out (M-MIMO) systems, the base station estimates each forward-link channel by using the reverse-link sounding pilot and the channel reciprocity property of time division duplex (TDD) operation. However, conventional non-cooperative sounding schemes, like a sounding sequence assignment scheme with reuse factor-1, cause cell edge users to suffer from pilot contamination in multi-cell scenarios. The pilot contamination problem becomes even worse in an environment where mobile users are travelling through cell edge areas frequently. To alleviate this problem, a cooperative sounding resource reuse method that assigns each cell a set of resources different from other neighboring cells by using a reuse factor-3 scheme, can be considered, but it significantly lowers the number of served users. In this paper, we propose a partial sounding resource reuse (PSRR) method which aims to seamlessly support the quality of service (QoS) of each mobile user by mitigating the pilot contamination and to minimize the reduction in the number of served users. To this end, the PSRR divides each cell area into center and edge areas, and applies a reuse factor-1 scheme for center users while a reuse factor-3 scheme for edge users. We use a Markov chain model to analyze the performance of the PSRR, and evaluate the accuracy of our analysis through simulations. Then we confirm that the PSRR considerably improves QoS performance over the conventional competitive schemes.

IEEE 802.15.4 networks employ a medium access control scheme based on carrier sense multiple access with collision avoidance (CSMA/CA). However, CSMA/CA may seriously experience contention collision, suffering from resource wastage in time-varying operation environments. In this paper, we consider the alleviation of contention collision in IEEE 802.15.4 networks by partitioning the transmission period into a number of sub-transmission periods. To this end, the network coordinator and its child devices measure the transmission performance in a distributed manner. When the measured performance is not satisfactory, the coordinator adjusts the number of subtransmission periods considering the operation environment. Each child device uses one of the sub-transmission periods for its signal transmission, effectively reducing the probability of contention collision. Simulation results show that the proposed scheme remarkably reduces packet collision and thus energy consumption in time-varying network environments.

IEEE 802.15.4 employs a medium access control protocol based on carrier sense multiple access with collision avoidance (CSMA/CA). Channel sensing may have critical impact on the system throughput and energy consumption as well. In this paper, we investigate the effect of channel sensing on the CSMA/CA performance in IEEE 802.15.4 star-topology configured networks. We show that the hidden node ratio (i.e., the probability that any two child nodes cannot hear each other in a star-topology configured network) is highly associated with the target false-alarm probability of the channel sensing detector. We show that the CSMA/CA performance of IEEE 802.15.4 network can significantly be improved by adjusting the hidden node ratio.

In this letter, we consider the mitigation of co-channel interference in Bluetooth piconets. Prior to sending packets, the proposed scheme checks whether the next channel for the frequency hopping (FH) is in use by other signals or not. Then, it periodically detects busy channels subject to WLAN interference based on the packet error rate and interference signal detection rate in a group-wise manner and changes the set of hopping frequency. For further provision of channels for the FH, it maximized the number of channels for the FH by detecting additional channels which belong to groups classifies as being interrupted by WLAN, but not being used. The analytic design and simulation results show that the proposed scheme is quite robust to the presence of multiple WLAN and other Bluetooth signals, compared to the conventional mitigation schemes.

In this paper, we consider interference mitigation in IEEE 802.15.4 beacon-enabled cluster-tree networks. The proposed scheme first detects the presence of interference by means of packet-error detection and/or channel sensing in a clusterwise manner. In the presence of interference, the coordinator (or cluster head) fast notifies the presence of interference to its end devices by repeatedly transmitting the channel-handoff command. Then, it avoids the interference by the use of temporary channel hopping through predetermined hopping channels. Finally, it selects the best one among the hopping channels, which will be used as the final handoff channel. Simulation results show that the proposed scheme provides marked performance improvement over conventional schemes even in the presence of heavy interference.

In this paper, we consider performance improvement of IEEE 802.15.4 systems in the presence of co-channel interference from other systems such as wireless local area networks (WLAN). The proposed scheme first detects the presence of interference by means of spectrum sensing or transmission performance testing. When the presence of interference signal is detected, the coordinator changes the carrier frequency for the signal transmission by means of frequency hopping through multiple channels which are predetermined according to the characteristics of interference signal. After signal transmission through these multiple channels for a while, the coordinator selects the best one among the multiple channels and then returns to the original transmission mode through the newly selected channel, efficiently alleviating the interference problem. Finally, the performance of the proposed scheme is verified by computer simulation in the presence of multiple WLANs.

In this paper, we consider channel sounding for multi-sector cooperative beamforming in othogonal frequency division multiplexing-based wireless cellular systems. To reduce sounding interference and signaling overhead, we allocate sounding physical resource units (PRUs) by exploiting the correlation characteristics in the frequency domain. The proposed scheme transmits sounding signal through PRUs which are not highly correlated to each other, significantly reducing the sounding interference and signaling overhead. Then, the whole channel information can be extracted by means of channel estimation based on the frequency correlation (e.g., minimum-mean-square-error estimation). Simulation results show that the proposed scheme provides remarkable performance improvement over conventional schemes in the presence of channel correlation in the frequency domain.