Accurate prediction of drug–drug interactions (DDIs) for newly synthesized compounds enables early, in-silico safety screening in drug discovery and formulary review. We target the cold-start regime, where (i) new compounds are topologically isolated on external biomedical knowledge graphs (KGs) and on the DDI graph, and (ii) sparse supervision hampers the learning of discriminative representations. We propose an early-fusion method (LINCS-DDI) that inserts shared substructure nodes to connect a molecular-fingerprint knowledge graph with the DDI graph, turning structural similarity into topological links, and providing two-hop connectivity directly from the Simplified Molecular Input Line Entry System (SMILES) without prior inclusion in external KGs. Building on this substrate, we introduce Native Dual-View Contrastive Learning (NDV-CL): within a single pass of a flow-based graph neural network (GNN), forward and reverse message-passing representations of the same drug pair are treated as deterministic positives, while label-guided negatives (screened using only training-split interaction labels) are mined within the induced subgraph, improving representation quality without stochastic augmentations. Under strict cold-start settings on two open-source datasets, LINCS-DDI improves macro-F1 by up to 4.1% over the best baseline and reduces contrastive overhead by up to 66%. These properties make the approach suitable for routine, large-scale preclinical DDI triage, prioritizing high-risk combinations for wet-lab validation, and informing pharmacovigilance pipelines.

Monitoring respiratory rate variability (RRV) via smartphone microphones represents a significant opportunity for accessible digital health. However, extracting clean respiratory signals from acoustic recordings in noisy real-world environments remains a major obstacle. This paper presents a Frame-bin Adaptive Stochastic Modeling (FASM) framework with an Attention-based Time-Frequency Masking Network (ATFRespNet) for robust RRV estimation using commodity smartphones. FASM adaptively partitions the acoustic field into localized frame-bins and introduces a coefficient-based stochastic modeling strategy that precisely distinguishes physiological reflections and separates RRV from motion artifacts through a stochastic instantaneous frequency (SIF) method. ATFRespNet further enhances the respiration-related spectral regions through an attention-driven masking mechanism, while an autoencoder (AE) refines the waveform morphology for accurate rate estimation. We evaluated our model on a diverse dataset of real-world recordings containing various environmental noises. The results demonstrate that our method achieves state-of-the-art accuracy in estimating respiratory rate.

Health monitoring is a critical component of modern healthcare, requiring continuous or periodic measurement of physiological parameters to accurately assess personal health status. Advances in wearable technology have significantly improved the accessibility and convenience of such monitoring. Among various form factors, earables offer unique advantages: they can capture rich biosignals, provide stable and motion-resistant measurements, ensure long-term comfort, maintain discreteness, and integrate seamlessly with everyday audio functionalities. By investigating the latest technological advances and application cases in ear-worn devices, this survey reviews the current state of earable technology in health monitoring, identifies gaps and opportunities, and suggests directions for future research and development. We first explore the multifaceted role of earables in health monitoring, including measurement of physiological parameters, activity monitoring, and healthcare applications. We then summarize the challenges of robustness, context-awareness, and signal fidelity, and outline six future directions-dynamic monitoring, context-aware processing, multimodal fusion, semantic activity understanding, personalized adaptation, and explainable AI-to advance earable health monitoring.

Bluetooth Low Energy (BLE) is a prevalent technology in various applications due to its low power consumption and wide device compatibility. Despite its numerous advantages, the encryption methods of BLE often expose devices to potential attacks. To fortify security, we investigate the application of Physical-layer Key Generation (PKG), a promising technology that enables devices to generate a shared secret key from their shared physical environment. Although extensively investigated, PKG is generally discussed in the context of Wi-Fi, and existing solutions for BLE demonstrate significantly lower performance. To bridge this gap, we propose a distinctive approach that capitalizes on the inherent characteristics of BLE to facilitate efficient PKG. We utilize the constant tone extension within BLE protocols to extract comprehensive physical layer information and introduce an innovative method that employs Legendre polynomial quantization for PKG. This method facilitates the exchange of secret keys with a high key matching rate and a high key generation rate. The efficacy of our approach is validated through extensive experiments on a software-defined radio platform, underscoring its potential to enhance security in the rapidly expanding field of BLE applications. A pilot study on commercial off-the-shelf BLE devices further validates the system's practicality, revealing important trade-offs between performance and hardware constraints in real-world deployments.

In this paper, we present Ear2Pos, a novel framework for full-body pose estimation using only two Inertial Measurement Units (IMUs) integrated into earbuds. Unlike traditional motion capture systems requiring multiple sensors, Ear2Pos leverages a minimal setup to achieve high-accuracy 3D motion reconstruction. The system incorporates a dualcoordinate framework and Transformer-based modeling to predict joint positions and rotations. Additionally, we propose a personalized skeletal parameterization mechanism, utilizing extracted bone lengths from a single image to enhance individual adaptability. Extensive evaluations demonstrate that Ear2Pos achieves state-of-the-art accuracy in pose estimation when using two sensors, outperforming other methods in upper-body motion prediction with an average joint position error of 5.04 cm. Furthermore, we explore clinical applications, particularly in gait analysis for cervical spondylotic myelopathy (CSM) patients, showcasing the frameworks potential for rehabilitation assessment. These findings indicate that Ear2Pos is a promising lightweight solution for non-invasive motion capture, offering robust performance in both research and real-world applications.

On low-power, low-cost Internet of Things (IoT) edges, coarse-grained entropy source-based physical-layer key generation (PKG) is often used, which results in a very low bit generation rate (BGR). In this paper, a novel PKG scheme, Hinge, designed to adapt to varying environmental conditions is introduced to optimize the trade-off between the bit mismatch rate (BMR) and BGR using fine-grained entropy sources on IoT devices. Hinge predicts channel reciprocity levels from one side and dynamically adjusts the quantization strategy, maintaining a low BMR while maximizing BGR. Compared with existing PKG solutions on Bluetooth devices, Hinge yields significant improvements in BGR, with a comparable BMR. Through extensive experiments, Hinge showcases its potential for providing a secure and efficient key generation mechanism for IoT devices in complex real-world scenarios.

In this paper, we propose a dynamic data transmission strategy for smart home environments that aims to optimize the Quality of Experience (QoE) by adaptively adjusting the data upload frequency based on the predicted trends in sensor data. Using the home wireless sensors monitoring dataset, we implement a deep learning model for accurate time series forecasting. In addition, an anomaly detection mechanism is used to identify critical events, requiring more frequent data uploads when important changes are detected. The QoE is quantified through a weighted average of several influencing factors, including data timeliness, timely upload of critical events, and transmission frequency. Our optimization objective is to maximize QoE while minimizing the number of transmissions, with an emphasis on reducing energy consumption through intelligent scheduling. The results demonstrate that our approach effectively balances data timeliness, transmission efficiency, and energy savings, leading to improved user satisfaction in smart home applications.

Bluetooth Low Energy (BLE) Mesh is a pivotal multi-hop self-organizing network in the Internet of Things (IoT) domain, offering low power consumption, low cost, and robustness. This paper presents a comprehensive study on the communication performance of BLE-Mesh using commercial off-the-shelf devices, focusing on the impact of key mesh parameters such as transmission power, packet interval, and network structure on performance. Through extensive indoor and outdoor experiments, we quantify the impact of these parameters and conduct a detailed study. Our findings provide insights into the actual communication range of BLE-Mesh, the effect of node design on overall network performance, and the configuration for optimal performance. The research contributes to the establishment of a BLE-Mesh network in real-world environments, answering critical questions for practitioners, and offering a reference for future BLE-Mesh deployments. This work furthers our understanding of the characteristics, challenges, and future directions of BLE-Mesh, setting the stage for advancements in IoT applications such as smart offices and homes.

Bluetooth Low Energy (BLE) is a prevalent technology in various applications due to its low power consumption and wide device compatibility. Despite its numerous advantages, the encryption methods of BLE often expose devices to potential attacks. To fortify security, we investigate the application of Physical-layer Key Generation (PKG), a promising technology that enables devices to generate a shared secret key from their shared physical environment. We propose a distinctive approach that capitalizes on the inherent characteristics of BLE to facilitate efficient PKG. We harness the constant tone extension within BLE protocols to extract comprehensive physical layer information and introduce an innovative method that employs Legendre polynomial quantization for PKG. This method facilitates the exchange of secret keys with a high key matching rate and a high key generation rate. The efficacy of our approach is validated through extensive experiments on a software-defined radio platform, underscoring its potential to enhance security in the rapidly expanding field of BLE applications.

User identification, especially multiuser identification, plays an important role in Internet of Things (IoT)-enabled smart spaces. The early wearable or vision-based solutions either cause discomfort or suffer from privacy leakage, and the radio frequency (RF)-based methods are appreciated in recent years. Compared with other RF technologies, the millimeter wave (mmWave) has the merit of high spatial resolution and has been widely employed in wireless sensing. In this article, we present a multiuser gait identification system based on micro-Doppler calibration (MCGait) using a commodity mmWave radar. With the raw signals as the input, MCGait first extracts the point clouds with a pipeline of signal preprocessing and separates them using a spatial cluster algorithm for multitarget tracking. Then, MCGait conducts a velocity calibration with a virtual radar-based method and calibrates temporal gait micro-Doppler features for each user, so as to eliminate the negative effect of gait direction dynamics. Finally, the calibrated features are fed into a neural network to identify all the users. We implement MCGait on a commodity 77-GHz mmWave radar and conduct extensive experiments to validate its performance. The experimental results show that the proposed MCGait can achieve up to 98.50% single-user recognition accuracy, and over 95.45% identification accuracy for up to four users.

The rapid proliferation of Internet of Things (IoT) devices has led to the deployment of numerous sensor nodes in various industrial scenarios. These nodes play a crucial role in collecting, relaying, processing, and transmitting data through wireless communication. The interconnections between these nodes form diverse topologies, each exhibiting varying degrees of robustness against different types of attacks. To make the robustness of topology more robustness, this article proposes a robust topology generation method for IoT nodes. The method leverages the proximal policy optimization (PPO) algorithm in reinforcement learning, combined with a discrete action space that closely aligns with real-world deployment environments. By utilizing PPO and discrete actions, the method effectively optimizes the topology of IoT nodes, considering the constraints and limitations of the deployment scenario. In addition, this article introduces a novel metric called$S$-value to evaluate the robustness of the generated topology. Unlike existing metrics, the$S$-value provides a more practical and meaningful assessment of topology robustness, taking into account the ability of the topology to withstand attacks and maintain connectivity to the server even when nodes with high degrees fail. Experimental results demonstrate the effectiveness of the proposed method in generating robust topologies for IoT nodes.

As a natural interaction approach, handwriting acts as an essential role in human–computer interaction. In order to achieve ubiquitous and reliable applications, a handwriting recognition system should be easy to use without any intrusion and robust to environment changes, which cannot be satisfied in most of existing approaches. In this article, we presentRFPad, a device-free handwriting recognition system with a tag square consisting of four low-cost passive radio frequency identification (RFID) tags. With such an ingenious tag square, we transform the flat surface into a handwriting platform, and build a geometry-based theoretical model between the finger positions and the tags' phase variations so that the finger trajectory can be accurately tracked and handwriting can be recognized with the phases segmented from continuous signals. We implement a prototype ofRFPadusing commercial off-the-shelf RFID devices and conduct extensive experiments to evaluate its performance. Experiment results show that ourRFPadcan track finger movement with an average error of 1 cm and achieve average recognition accuracy of 94.08$\%$for all 26 handwriting capital letters.

Electronic signatures are widely used in financial business, telecommuting, and identity authentication. Offline electronic signatures are vulnerable to copy or replay attacks. Contact-based online electronic signatures are limited by indirect contact such as handwriting pads and may threaten the health of users. Consider combining hand shape features and writing process features to form electronic signatures, the article proposes an in-air handwritten signature verification system with millimeter-wave(mmWave) radar, namely mmHSV. First, the biometrics of the handwritten signature process are modeled, and phase-dependent biometrics and behavioral features are extracted from the mmWave radar mixture signal. Secondly, a handwritten feature recognition network based on few-sample learning is presented to fuse multi-dimensional features and determine user legitimacy. Finally, mmHSV is implemented and evaluated with commercial mmWave devices in different scenarios and attack mode conditions. Experimental results show that the mmHSV can achieve accurate, efficient, robust and scalable handwritten signature verification. Area Under Curve (AUC) is 98.96%, False Acceptance Rate (FAR) is 5.1% at the fixed threshold, AUC is 97.79% for untrained users.

Biometric sensing has become a widely concerned authentication technology. Existing image-based methods are susceptible to light conditions and have privacy exposure risks, while contact authentication methods are not conducive to epidemic prevention requirements in public places. In this article, we propose a palmprint-based identification system by collecting backscattered signals of the inaudible acoustic signals, namely AcoPalm. AcoPalm does not require special hardware and contact operation for user authentication. First, frequency modulated continuous wave (FMCW) on acoustic signals are designed to extract static contours and palmprint changes and to model the unique biological characteristics of the individual palm. Second, a palmprint authentication model based on PENN is proposed to achieve high-precision multiuser authentication without mass training data. Finally, the system performance is evaluated in multiple smartphones and three scenarios. The experimental results show that AcoPalm can resist replay attack and imitation attack, and the authentication accuracy can reach 96.22%. Furthermore, AcoPalm achieves satisfactory experience in availability and practicality.

In recent years, wireless sensor networks have been used in a wide range of indoor localization-based applications. Although promising, the existing works are dependent on a large number of anchor nodes to achieve localizations, which brings the issues of increasing of the cost and additional maintenance. Inspired by the cross-technology communication, an emerging technique that enables direct communication among heterogeneous wireless devices, we propose an anchor-free distributed method, which leverages the installed Wi-Fi APs to calculate the distance instead of traditional anchor nodes. More specifically, for the asymmetric coverage of Wi-Fi and ZigBee nodes, we first design a progressive method, where the first unknown node estimates its location based on two Wi-Fi APs and a sink node, then once achieving its position, it acts as the alternative sink node of the next hop. This process is repeated until the new members can obtain their positions. Second, as a low-power technology, ZigBee signal may be submerged in strong signals such as Wi-Fi. To overcome this problem, a prime number is deployed to be the Wi-Fi broadcasting period based on the numerical analysis theory. Among lots of prime numbers, we have the opportunity to select an appropriate one with the relatively small packet collisions. Last, numerical simulations and experiments are performed to evaluate the proposal. The evaluation results show that the proposal can achieve decimeter level accuracy without deploying any anchor node. Moreover, the proposal demonstrates the anti-interference ability in the crowded open spectrum environment.

Shopping behavior information acquisition and mining are of great importance for marketing and merchandizing strategies developing. With a single tag on each commodity, the previous pattern-based works only can be considered as behavior recognition solutions with behavior streams as output for each tag. In this article, we introduce a completely new mode of dual tags and propose a fine-grained shopping behavior acquisition solution called as fine-grained shopping behavior information acquisition (FGSA). FGSA enables shopping behaviors recognition by taking advantage of the phase calibrated Doppler shift of dual tags and extracts fine-grained behavior information for each behavior from the fusion Doppler spectrums with only one antenna. We implement a prototype of FGSA with COTS radio-frequency identification (RFID) devices and conduct extensive experiments to evaluate its performance. The experimental results demonstrate that FGSA achieves excellent performance in terms of both behavior recognition and information extraction.

Consumer Internet-of-Things platforms, especially in smart home, have received widespread attention. A large number of end-to-end and edge-to-end wireless communication links are subject to significant security and privacy risks. Key generation using radio link side-channel information is a burgeoning technology to solve this problem. This article designs a symmetric key generation method without environmental constraints, namely, MobiKey. The basic idea is taking the human mobility in home or the swing of the antenna to generate symmetric keys based on channel reciprocity. We leverage the practical phase information as the channel feature to generate symmetric keys between two communicating parties. To overcome the effects of noise on both communication sides and get a better performance, MobiKey harnesses the adaptive quantization method to make the bit generation rate and bit matching rate higher. MobiKey also uses the adaptive inconsecutive samples to increase the confusion of the adversary. Moreover, we propose the D-Gray code to enhance the randomness of quantization. MobiKey is implemented on different commercial devices, and the results show that it has comparative advantages in key generation consistency, randomness, and security. In addition, the performance in preventing predictable channel attacks and eavesdropping attack is discussed.

RFID systems nowadays are operated at large-scale in terms of both occupied space and tag quantity. One may have prior knowledge of the complete set of tags (denoted by N) and any set of wanted tags (denoted by M) within the complete set, i.e., M ⊆ N. Then here comes an open problem: when one is particularly interested in a subarea of the system, how to collect information (not simply tagIDs) from a wanted subset (denoted by dM) of the interrogated tags (denoted by dN) in that subarea? This issue has great significance in many practical applications but appears to be challenging when there is a stringent time constraint. In this work, we first establish the lower-bound of this problem, and show a straightforward polling solution. Then, we propose a novel polling protocol called LocP, which consists of two phases: the Tags-Filtering phase and the Ordering-and-Reporting phase. LocP employs Bloom Filter twice to significantly reduce the scale of candidate tags in the Tags-Filtering phase. In the Ordering-and-Reporting phase, tags determine their own transmission time-slots according to the allocation vectors iteratively broadcasted by the reader. LocP thus achieves a delicate tradeoff between time and polling accuracy. We conduct extensive simulations to evaluate the performance of LocP. The results demonstrate that LocP is highly efficient in terms of information collection time, leading to convincing applicability and scalability of large-scale RFID systems.

Under the influence of multipath effects and small scale fading, the robustness and reliability of the existing human detection methods based on radio frequency signals are easy to be impaired. In this paper, we propose a novel design that using the multi-layer filtering of channel state information (CSI) to identify moving targets in dynamic environments and analyze the gait periodicity of human. We employ an efficient CSI subcarrier feature difference to the multi-layer filtering method leveraging principal component analysis (PCA) and discrete wavelet transform (DWT) to eliminate the noises. Furthermore, we propose a profile matching mechanism for human detection and a periodicity analysis mechanism for human gait taking advantage of the above design. We evaluated it with the commodity Wi-Fi infrastructures in different environments. Experimental results indicate that our approach performs identification of human with an average accuracy of 94%.

Emergency navigation is an emerging application of wireless sensor networks with significant research and social value. In order to ensure the safe and timely navigation of the evacuees, most of the existing works model navigation as a path-planning problem or movement decision support problem and adopt different metrics, such as the shortest route, the minimum exposure path, and the maximum safe distance. Without sufficient consideration of the dynamics of danger, the existing approaches are likely to cause users to move back and forth during navigation, known as oscillation. Frequent oscillations inevitably result in the user remaining in danger for a longer period of time, amplification of the user's panic, and eventual decrease in the chances of survival. In this paper we take users' oscillations in the dynamic environments into account and quantify the local success rate of navigation using a metric called ENO (Expected Number of Oscillations). We then propose OPEN, an oscillation-free navigation approach that minimizes the probability of oscillation and guarantees the success rate of emergency navigation. We implement OPEN and evaluate its performance through the trace from our system and extensive simulations. The results demonstrate that OPEN outperforms the current state-of-the-art approaches with respect to user safety and navigation efficiency.

Emergency navigation is an emerging application of wireless sensor networks with significant research and social values. In order to ensure the safety and timeliness of navigation for the users, most of the existing works model navigation as a path-planning problem and adopt different metrics, such as the shortest route, the minimum exposure path, and the maximum safe distance. Without sufficient consideration of the dynamics of danger, the existing approaches are likely to cause users to move back and forth during navigation, known as oscillation. Frequent oscillations inevitably result in the user remaining in danger for a longer period of time, amplification the user's panic, and eventual decrease in the chances of survival. In this paper we take users' oscillations in the dynamic environments into account and quantify the local success rate of navigation using a metric called ENO (Expected Number of Oscillations). We then propose OPEN, an oscillation-free navigation approach that minimizes the probability of oscillation and guarantees the success rate of emergency navigation. We implement OPEN and evaluate its performance through test-bed experiments and extensive simulations. The results demonstrate that OPEN outperforms the current state-of-the-arts approaches with respect to user safety and navigation efficiency.