With the rapid evolution of societal and technological trends leading up to 2030, the limitations of existing 5G mobile communication networks have become increasingly evident. Emerging applications demand unprecedented performance from 6G networks, including extremely high data transfer rates, ultra-low latency, exceptional reliability, scalability, energy efficiency, and mobility. This paper explores relevant visions for 6G and comprehensively reviews its current developmental status. We specifically identify and analyze three key application scenarios driving 6G requirements: Integrated Sensing and Communication (ISAC), energy saving, and Artificial Intelligence(AI)-related technologies. Furthermore, the paper examines a range of potential future application scenarios and their associated demand directions. The analysis aims to provide a scientific basis and strategic guidance for the development of 6G technology, which is expected to pioneer new scenarios, offer innovative experiences, and support economic growth by enriching the global communication service.
Enhancing microvascular imaging improves tumor prediction and early assessment. Tissue harmonic imaging (THI) and super-harmonic imaging (SHI), leveraging the nonlinear effects of acoustic waves with dual-frequency transducers, enable high-precision imaging with improved contrast-to-tissue ratio (CTR). However, current dual-frequency systems are often bulky, expensive, and complex. In this paper, a portable ZYNQ-based dual-frequency ultrafast ultrasound imaging system is proposed, capable of simultaneously exciting and receiving signals. System performance was evaluated through low and high frequency pulse transmission tests and dual-frequency tissue harmonic imaging (DF-THI) experiments. Results demonstrate flexible pulse frequency adjustment, 80V excitation at both frequencies, and plane wave electronic beam steering. DF-THI experiments at 6000 frames per second achieved a CTR of 38.4 dB after 15angle coherent compounding. Utilizing only a ZYNQ-based acquisition board and a PC, the system demonstrates strong portability and leaves an ARM-based interface for future onboard RF processing and image display, showing potential for portable ultrafast nonlinear and super-harmonic imaging.
On-device learning is becoming increasingly important for enabling adaptive and privacy-preserving AI directly on resource-constrained edge devices. However, traditional deep learning approaches based on backpropagation (BP) remain computationally expensive and memory-intensive, making them impractical in such environments. To address these limitations, the Forward-Forward (FF) algorithm has been proposed as an alternative, replacing the backward pass with two forward passes to reduce memory overhead and computational complexity. In this paper, we introduce Sparse-FF, a novel variation of the FF algorithm that incorporates a stochastic selection mechanism to update only a subset of weights per training step. By dynamically selecting a fraction of neurons for activation and weight updates, Sparse-FF significantly reduces computational and memory requirements while maintaining competitive performance. We provide a formal analysis of its computational complexity, demonstrating its efficiency compared to both BP and FF. In addition, we conducted extensive experiments on the MNIST dataset to evaluate its accuracy and training speed. Our results show that Sparse-FF achieves a favorable balance between efficiency and model performance, making it a promising alternative for real-time edge learning applications where resources are limited.
This study introduces a wide-range tensile force detection system employing optoelectronic principles along the Z-direction, enabling accurate measurement of vertical mechanical signals through optoelectronic conversion mechanisms. The system comprises an elastic body, support layer, optoelectronic devices, and conditioning circuitry. Light signals emitted by light-emitting diodes (LEDs) are reflected by a metallic reflective coating on the inner wall of the elastic body and subsequently received by photodiodes (PDs). When tensile force induces deformation in the elastic body, the resulting photocurrent output reflects the deformation status of the inner wall, thereby enabling characterization of Z-axis tensile forces. The optoelectronic chip is fabricated using GaN-based semiconductor optoelectronic processes, with an active low-pass filter circuit converting optical signals to electrical signals. Experimental results demonstrate that at 25 degrees C, the system achieves tensile force characterization within 0-1.1 MPa, featuring a sensitivity of 0.36 mV/kPa, nonlinearity error < +/- 5.56% FS, repeatability error < +/- 6.83% FS, and overall accuracy < +/- 8.83% FS. These findings confirm that the optoelectronic sensing-based measurement system effectively characterizes tensile forces with excellent output stability and reliability.
Efficient and flexible wireless power transfer (WPT) is essential for the development of next-generation wearable electronics. This work presents a wireless power-sharing scheme based on magnetic resonance coupling (MRC-WPSS) that enables simultaneous and controllable power distribution between two passive loads using only passive receiving circuits. By adjusting the input resonant capacitance on the transmitter side and configuring for loose coupling, the system achieves tunable and overlapping resonance profiles, allowing independent control of the power delivered to each receiver. Both simulation and experimental results confirm that varying the input capacitance enables seamless power sharing, allowing one receiver to be selectively powered or the load to be equally distributed. Physical experiments with fabric-based resistive heaters demonstrate the system’s capabilities, achieving independent temperature control and validating power sharing in real-time. The average power transfer efficiency reaches 15.65%, sufficient for practical applications in multi-zone wearable heating. This fully passive approach eliminates the need for active or complex receiver-side circuitry, providing a simple and scalable solution for distributed wearable electronics and other remote actuation scenarios.
This paper proposes an elastic LC-based z-tensile force sensor, which can be applied in 0-100 N z-tensile force or 0-1 MPa mechanical pressure measurement. Combined with the planar spiral inductor, the reduction in capacitance caused by tensile force would manifest as a shift to higher resonant frequencies. The influence of the capacitor dimensions on the resonant frequency is also totally analyzed. The sensor is fabricated through the standard flexible printed circuits (FPC) technology, with a 10 mm by 10 mm force-sensing area. An elastic 90-MPa polyurethane (PU) sheet is chosen as the dielectric layer. Experiment results show that the sensor has a resonant frequency around 110 MHz, and a sensitivity of 3.27 kHz/N (327 kHz/MPa). The calculated nonlinearity error is 2.1%, demonstrating that this LC-based z-tensile force sensor can effectively characterize tensile force values with good stability.
Digital healthcare (eHealth) is a key vertical for future sixth-generation networks (6G), especially in the wake of the Covid-19 pandemic, which emphasized the need for widespread telemedicine solutions. The advent of 6G, thanks to the technologies that will characterize it, offers several growth opportunities for the eHealth sector. For example, artificial intelligence (AI) is currently widely used in healthcare. However, the privacy of data and information is not always guaranteed by the network architectures that support such applications. In this paper, we present a novel federated learning (FL) framework that combines digital twin (DT) simulation and edge intelligence to enable distributed learning while preserving data privacy. We demonstrate this by simulating a realistic edge-learning scenario in which electrocardiogram (ECG) data from patients are transmitted to distributed virtual twins, and show that selectively including only the most informative clients based on prediction uncertainty significantly improves both convergence speed and model generalization. Specifically, the proposed DT strategy outperforms both conventional and purely entropy-based approaches on key evaluation measures by using past trends and local training performance to predict future client utility. The proposed architecture provides a versatile and flexible foundation for implementing AI-driven and privacy-friendly healthcare systems.
This paper presents a hybrid architecture solution based on the EMQX IoT platform for battery management systems (BMS). The system utilizes the STM32F103 as the core controller to handle data processing for the BQ76930 battery monitoring chip. Additionally, the system extends BMS functionalities through the Arduino Mega 2560, integrating a GPS module and a TJC4848T040_011 serial display for human-machine interaction. The key features of the system include: 1) Data transmission based on the reliable and low-overhead MQTT protocol; 2) Remote monitoring of battery data via the cloud; 3) Real-time tracking of BMS location through a cross-platform app (Android/iOS). Experimental results demonstrate that the system can effectively transmit data and retrieve readings in a networked environment. The GPS module achieves an accuracy of 2.5 meters in practical conditions, offering low power consumption and reduced hardware costs compared to traditional industrial solutions. By utilizing the STM32F103 chip for control, the system efficiently transmits data and retrieves readings, reducing overall hardware costs and power consumption compared to conventional industrial solutions. This architecture provides a highly reliable and scalable BMS cloud collaboration solution for Electric Vehicles (EV) and mobile energy storage devices.
Wireless sensor networks consist of independent, resource-limited nodes that work together to achieve a certain goal, with applications in both civil and military domains, such as remote environmental monitoring and target tracking. These nodes are deployed in unattended and unsecured environments, rendering them vulnerable to attacks like node replication, where adversaries capture nodes, create multiple replicas with identical credentials, and strategically place them to carry out malicious activities. In this paper, we propose a novel protocol, the "Entangle and Challenge", to resist such node replication attacks. Because of the monogamy of entanglement, only two genuine nodes can be maximally entangled at any given time. Replica nodes are detected by challenging two genuine nodes to verify that they remain maximally entangled. Our protocol demands minimal quantum hardware and channel requirements, making it particularly appealing for near-term quantum networks.
Wellness and comfort are crucial factors that impact occupant health, productivity, and overall satisfaction, and the advent of the Internet of Things (IoT) has brought about a paradigm shift in building management, revolutionizing the integration of advanced sensors and data analytics. This convergence has enabled the development of intelligent building systems that improve energy efficiency and optimize thermal comfort for occupants. In this paper, we propose a novel framework that uses IoT technologies with multivariate analysis (MVA) for the maximization of indoor wellness using IoT data collected from a real environment and correlating them with satisfaction score, obtained through questionnaires. To demonstrate the robustness of the model, we will proceed in two steps. First, with no action control, we will show how the model could be used to effectively identify/classify the environments according to data gathered from IoT, leveraging the soft independent modeling of class analogies (SIMCA) classification method. Secondly, we will show how the control of the temperature(T) could strongly enhance wellness using the same model to predict the optimal indoor temperature. A partial least squares (PLS) regression model is employed for maximizing occupant thermal comfort while optimizing heating, ventilation, and air conditioning (HVAC) energy efficiency. The results demonstrate that the model effectively generalizes across different environments, achieving a satisfaction success rate of 76.47% and significantly reducing the predicted percentage of dissatisfied (PPD) in accordance with Fanger’s thermal comfort model. These findings underscore the model’s robustness, showing a competitive approach to deep learningbased algorithms.
The rapid increase in the elderly population globally and within Singapore has underscored the need for effective solutions that promotes both independence and safety for elderly individuals. Many elderly face risks such as getting lost, experiencing falls, or encountering emergencies without immediate assistance, particularly those with dementia. Current tracking and monitoring technologies often face limitations in scalability, accuracy, and affordability, especially in complex indoor environments. To address these challenges, we developed the Elderly Tracking and Monitoring System (ETMS), a hybrid IoT network integrating LoRa and painlessMesh technologies for reliable indoor and multi-floor tracking. ETMS enables caregivers to receive real-time alerts via geofencing when an elderly individual exits predefined safe zones, with a simple interface that supports efficient monitoring. Extensive testing in a simulated residential setup demonstrated the system’s accurate indoor localization and stable performance across multiple floors, despite minor signal interference. ETMS provides a cost-effective, scalable solution, promoting independent living while alleviating the monitoring burden on caregivers.
This paper presents an IoT-based intelligent parenting system that integrates edge sensing, psychological temperament modeling, large language models (LLMs), and user feedback to deliver personalized and responsive infant care. Unlike existing solutions that rely on passive monitoring or generic recommendations, our system features a closed-loop, multi-agent architecture that supports individualized, explainable parenting interventions. The architecture follows a cloud-edge-user collaborative paradigm: multimodal sensors deployed at the edge and edge agents continuously monitor infant behavior and environmental conditions; on the cloud, an LLM agent performs context-aware reasoning with function calling, while a DeepSearch QA agent retrieves temperament-aligned parenting knowledge. A caregiver-facing app enables temperament profiling, query submission, and feedback collection, completing a human-in-the-loop cycle. To support fine-grained personalization, we construct a centralized knowledge base from validated parenting resources, continuously enriched with de-identified Q&A samples approved by users. Infant temperament profiles influence both retrieval and reasoning, ensuring targeted and interpretable responses. In experimental evaluations, our system achieved a 90% recall rate in personalized knowledge retrieval, significantly outperforming baseline RAG systems (65%). For response generation, Top-1 human approval reached 80%, highlighting the system’s potential as a trustworthy and adaptive AI assistant for infant care. We simulate the full pipeline within a smart crib use case to validate the closed-loop interaction and personalization capability under realistic care conditions.
The development of 6G wireless networks emphasizes reducing communication overhead without compromising information accuracy. Semantic communication enables the transmission of latent data representations, minimizing the need for raw data while improving efficiency. However, wireless networks are susceptible to noise, path loss, and other channel impairments, which can degrade information quality under compressed representations. This paper proposes an adaptive multi-bottleneck semantic communication model that adjusts bottleneck sizes based on channel conditions to optimize efficiency and mitigate noise. The proposed model is evaluated using realistic channel simulations and demonstrates improved performance across varying conditions, balancing efficiency and robustness for next-generation communication systems.
With the rapid growth of China's low-altitude economy (LAE), there is an increasing demand for reliable unmanned aerial vehicles (UAVs) and electric vertical takeoff and landing (eVTOL) systems, which drives the development of low-cost and high-precision inertial measurement units (IMUs), including accelerometers and gyroscopes. Interferometric fiber-optic gyroscopes (IFOGs) have been widely adopted in aerospace and navigation systems; however, achieving high accuracy in a compact and power-efficient implementation remains a key challenge for urban air mobility applications, placing new demands on both optical and circuit design. This paper presents a low-noise integrated CMOS analog front-end (AFE) circuit for photodetection in IFOGs. The AFE comprises a transimpedance amplifier (TIA) that converts photocurrent to voltage, followed by filtering and driving stages. Compared with conventional discrete implementations using PIN-FET detectors, the proposed AFE offers higher integration and lower power consumption while maintaining competitive noise performance. Fabricated in a 0.18 mu m CMOS process, the AFE was tested in a commercial IFOG by replacing its original front-end components. Experimental results demonstrate a zero-bias stability of 0.085 degrees/h and an angular random walk (ARW) of 0.0048 degrees/vh at room temperature, validating the effectiveness of the proposed design.
We focus on an integrated all-optical quantum random number generator (QRNG) based on the bi-phase states of degenerate optical parametric oscillators (DOPOs), implemented using a silicon nitride microresonator driven by dual pumps. In our numerical analysis, we present a comparative study of phase bifurcation in DOPOs initialized under four distinct noise schemes. As a representative case, we simulate random number generation in the Kerr-based microresonator at a rate of 10 MHz for each noise model by using stochastic differential equations.
To address the low-carbon imperative in the power sector, this paper introduces a two-layer optimization framework that explicitly couples the electricity and carbon markets. The upper layer treats the carbon price as a decision variable, with the objective of maximizing social welfare while accounting for carbon emission penalties, enabling flexible constraints on total carbon emissions. The lower layer treats the electricity price as its decision variable and conducts generation adjustments through processes that integrate carbon cost constraints, thus facilitating both day-ahead and real-time market clearing. The simulation results indicate that, compared to a baseline that ignores carbon cost, the proposed mechanism markedly enhances renewable-energy integration and streamlines power-resource allocation.
With the growing importance of passive wireless sensors in the Internet of Things (IoT), efficient sensor signal detection has become increasingly critical. This paper presents a fast time-domain signal analysis circuit designed for passive wireless sensors, utilizing a FPGA as the main control module for frequency measurement. The design is based on a closed-loop control principle similar to a phase-locked loop (PLL), which reduces the reliance on high-speed analog-to-digital converters (ADC) through mixing and phase detection operations, thereby enabling rapid measurement. The system achieves a maximum measurement error of less than 0.6% and a stabilization time of less than 3ms. Furthermore, the measurement results are unaffected by changes in readout distance. This method demonstrates significant advantages in both speed and accuracy, making it highly suitable for LC sensor signal detection.
This work introduces a dual-band shared-aperture phased array antenna system that combines subarray architecture with ultra-wideband radiating elements for Ku/Ka-band operation. Functional diversity is achieved by integrating ultra-wideband antenna element operating at two different frequencies into the shared-aperture. To minimize the number of transmit/receive (Tx/Rx) channels, two types of triangular-grid-based rectangular planar subarrays are employed. The two subarrays operate in the 14-18 GHz and 30-36 GHz bands, respectively and achieves a +/- 40 degrees-scanning range. The low-frequency subarray consists of 8 radiators, while the high-frequency subarray consists of 2 radiators. This architecture significantly reduces the number of Tx/Rx channel with wideband and wide-scanning performance.
This paper addresses the challenge of multiple unmanned aerial vehicles target assignment and dynamic tracking within communication-constrained environments. We propose a framework based on computational intelligence (CI), which integrates two key components. First, for target assignment, we introduce a distributed Hungarian algorithm (DHA). The DHA leverages local information exchange and iterative optimization to achieve efficient assignment, thereby mitigating the reliance on global information while preserving the optimality characteristics of the traditional Hungarian algorithm. Second, for target tracking, we employ a two-stage deep reinforcement learning (DRL) algorithm based pre-trained clone learning. This approach enables the UAV team to learn adaptive strategies for achieving sustained cooperative tracking. Simulation results demonstrate that the proposed framework yields both efficient assignment and accurate tracking performance in dynamic environments.
The growing demand for smaller, lighter, and more embedded hardware has made Physical Unclonable Functions (PUFs) a promising solution for authentication in Internet of Things (IoT) applications. Traditional PUF authentication methods often rely on Error Correction (EC), which can be computationally intensive and time-consuming. Given the time-critical nature of authentication in security-sensitive IoT systems, there is a need for efficient alternatives. Machine Learning (ML) techniques have emerged as a viable option for bypassing EC by leveraging pre-trained models. This paper proposes a novel approach for authenticating DRAM PUFs using Random Forest (RF) classifiers. RFs, which are ensembles of independently trained decision trees, leverage collective voting to improve classification accuracy. This approach enhances confidence, reduces bias, and offers greater transparency, as the decision-making process of decision trees is inherently interpretable. Our results demonstrate that RF classifiers provide robust performance, comparable to or superior to other ML approaches commonly employed for these tasks, offering a reliable and efficient alternative to EC-based methods.
