Email alert
Current Articles
2025, Volume 47, Issue 4
Display Method:
2025,
47(4):
859-875.
doi: 10.11999/JEIT241116
Abstract:
Significance Unmanned Aerial Vehicles (UAVs) have attracted significant attention due to their flexibility, high mobility, and potential for widespread applications across various industries. The integration of UAVs with Integrated Sensing and Communication (ISAC) technology enables the combination of sensing and communication capabilities on a single platform, facilitating high-quality data collection, processing, and real-time communication, particularly in complex environments. This integration offers substantial benefits in both communication and environmental sensing, addressing key challenges in emerging fields, particularly in low-altitude economic scenarios such as smart cities, geomatics, and emergency rescue. Progress This paper provides a systematic survey of UAV-enabled ISAC networks, offering a comprehensive discussion on their underlying principles and the integration of communication and sensing tasks. The first section introduces the foundational principles and characteristics of ISAC technology, including a review of waveform sensing-communication integration and multi-modal sensing-communication technologies. The paper also examines recent efforts toward standardizing ISAC technology and emphasizes the importance of sharing and co-scheduling hardware and spectrum resources to improve overall system efficiency. Subsequently, the paper explores two main network architectures for deploying ISAC devices on UAVs and ground stations. First, it investigates sensing-assisted communication tasks, where the deployment of ISAC devices within UAV communication networks ensures efficient resource allocation, improved coverage, and enhanced communication performance, particularly in challenging environments. Second, it discusses sensing-communication fusion tasks, where UAV-enabled ISAC networks integrate functions such as positioning, edge computing, and data caching. UAVs play a pivotal role in combining these functionalities to optimize overall system performance. Through UAV-enabled ISAC technology, the system’s capacity to collect environmental data, perform real-time communication, and support intelligent decision-making in complex, dynamic conditions is significantly enhanced. Additionally, the paper surveys the current state of key UAV-enabled ISAC technologies, focusing on two main aspects: sensing-enabled techniques and resource allocation strategies. From the perspective of sensing-enabled technologies, advanced techniques such as massive MIMO, collaborative sensing, near-field communication, and multi-modal sensing notably improve UAVs’ sensing precision and coverage in dynamic environments, thereby facilitating the successful execution of various tasks. Furthermore, the paper examines resource allocation techniques, addressing the challenges of distributing energy, spectrum, and processing power within energy-constrained UAV systems. It also covers the integration of wireless energy harvesting, Reconfigurable Intelligent Surfaces (RIS), and advanced communication techniques, such as covert communication, which enable UAVs to operate more efficiently in challenging environments with limited energy resources. Conclusions UAV-enabled ISAC technology is progressing rapidly and holds significant potential to transform the integration of communication and sensing tasks within UAV networks. By capitalizing on UAV mobility and versatility, ISAC networks facilitate the seamless integration of communication and environmental sensing on a single platform. This integration enhances UAV performance and adaptability in complex environments while improving resource utilization, ensuring the efficient operation of UAV networks in applications such as smart cities, geographic surveying, and emergency response. Prospects Although significant progress has been made in developing UAV-enabled ISAC networks, several challenges persist. Energy limitations, complex transmission environments, and network security are critical issues that must be addressed for UAVs to operate effectively in dynamic and diverse environments. Future research will need to focus on overcoming these challenges by integrating emerging technologies such as wireless energy harvesting and RIS, which can enhance energy efficiency and network performance. Furthermore, geographic information-enabled technologies, such as radio maps, will increasingly play a crucial role in optimizing UAV deployment and navigation, particularly in complex environments. In addition, integrating covert communication techniques into UAV networks offers a promising avenue for improving the security and privacy of UAV-based communication systems, especially in sensitive applications such as defense and surveillance. The future of UAV-enabled ISAC networks will depend on addressing challenges related to energy constraints, environmental complexity, and security concerns, while enhancing the efficiency and effectiveness of communication and sensing tasks. As these technologies mature, UAVs will become even more integral to emerging low-altitude economies, fostering the development of smart cities, efficient disaster response systems, and intelligent traffic management.
2025,
47(4):
876-887.
doi: 10.11999/JEIT241151
Abstract:
Significance The communication–sensing–computing integrated network, a central direction in the development of Sixth-Generation (6G) mobile communication systems, represents a shift toward intelligent network coordination. This architecture addresses key challenges in secure data transmission, efficient resource allocation, and intelligent network control. These capabilities are essential for supporting emerging applications in an era defined by pervasive connectivity and artificial intelligence. Progress This paper analyzes three key technologies of the communication–sensing–computing integrated network. First, a collaborative architecture integrating communication and sensing is proposed, which combines cloud–fog–edge computing and blockchain technologies to ensure secure data transmission and storage. Second, high-precision sensing and interference management are examined, including adaptive sensing mechanisms based on demand, a dual-layer optimized spectrum-sharing framework, and strategies for mitigating mutual interference in integrated systems. Third, Artificial Intelligence (AI)-driven frameworks for resource allocation are presented, including dynamic strategies for optimization and scheduling, which enhance multi-dimensional resource efficiency and improve network adaptability to support future intelligent, secure, and high-efficiency integrated networks. Conclusions The integrated architecture and methods presented in this paper form the technical foundation of the 6G communication–sensing–computing integrated network. The blockchain-enhanced framework provides robust security, whereas the adaptive sensing mechanisms and interference management strategies enable improved performance in complex network environments. The AI-driven resource allocation framework further enhances network operation by significantly improving resource utilization efficiency and adaptability. Prospects Future research on integrated communication–sensing–computing networks should focus on core technologies such as collaborative mechanisms across communication, sensing, and computing; heterogeneous data processing methods; and intelligent resource scheduling. These technologies are critical for addressing challenges related to resource optimization, interoperability, and dynamic adaptation in complex network environments. Through continued research and technological advancement, next-generation wireless systems aim to realize an efficient, reliable, and intelligent communication–sensing–computing integrated framework for 6G networks.
2025,
47(4):
888-908.
doi: 10.11999/JEIT240806
Abstract:
Significance The Internet of Things (IoT) has become a transformative technology in intelligent systems across diverse domains, including smart cities, industrial automation, and healthcare. However, traditional IoT systems, which isolate sensing, communication, computation, and intelligence, often face inefficiencies in resource utilization, increased latency, and scalability issues. These challenges are further intensified by the dynamic and resource-constrained requirements of future 6G IoT applications. Integrated Sensing, Communication, Computation, and Intelligence (ISCCI) offers a novel paradigm that unifies these functionalities, aiming to meet the needs of low-power transmission, multimodal sensing, low-latency computation, and distributed intelligence. By overcoming the fragmentation inherent in existing architectures, ISCCI can optimize resource allocation, enhance system adaptability, and support the development of next-generation IoT systems. Progress The development of 6G IoT depends on significant advancements across four key areas: communication, sensing, computation, and intelligence. These advancements aim to address the challenges of low-power transmission, multimodal sensing, low-latency computation, and distributed intelligence, enabling efficient resource utilization and robust system operation for future IoT applications. Specifically: (1) Communication: Low-power communication is essential for sustainable operation in large-scale IoT systems. Backscatter communication plays a critical role by allowing devices to transmit data without generating their own signals, thereby significantly reducing energy consumption. Simultaneously, Wireless Power Transfer (WPT) technology provides an energy-efficient solution for powering passive IoT devices, ensuring long-term operation even in resource-constrained environments. The integration of backscatter communication and WPT enhances passive IoT systems’ functionality and reduces maintenance requirements. (2) Sensing: Multimodal sensing integrates data from various sensor types, enabling accurate perception of complex environments. This approach supports real-time monitoring and adaptive decision-making in dynamic IoT scenarios. For instance, environmental sensing technologies can extract valuable insights from ambient signals, enhancing system awareness and facilitating efficient resource allocation. (3) Computation: Low-latency computation frameworks, such as edge computing, are essential for reducing reliance on centralized cloud servers. By offloading computational tasks to edge nodes, these frameworks improve system responsiveness and enable real-time processing. Additionally, joint communication and computation optimization techniques allow efficient task allocation, balancing resource constraints and application demands in heterogeneous IoT environments. (4) Intelligence: Distributed intelligence is achieved through collaborative learning frameworks like federated learning. By enabling IoT devices to train machine learning models collaboratively without sharing raw data, this approach ensures data privacy while promoting scalable intelligence across the network. Furthermore, real-time decision-making algorithms enable IoT systems to dynamically adapt to varying conditions, ensuring robust and efficient operation. Conclusions This paper proposes a future ISCCI IoT architecture comprising terminal nodes, fusion network elements, and service centers, and highlights four key enabling technologies that support the ISCCI IoT paradigm: (1) Environmental backscatter communication and sensing coexistence technology: At the terminal level, this technology enables IoT devices to use ambient signals for both communication and sensing, reducing the need for dedicated resources and improving energy efficiency. (2) Cloud-edge-device collaborative sensing task processing: Building on terminal-level capabilities, this technology orchestrates hierarchical processing architectures to optimize the allocation of sensing tasks across cloud, edge, and device layers. It ensures real-time performance and scalability by dynamically allocating tasks based on computational demands and latency requirements. (3) Intelligent computing for enhanced sensing prediction: To further enhance system adaptability, this technology integrates intelligent computing capabilities into the IoT network architecture. It improves sensing accuracy and enables proactive resource allocation, ensuring efficient and adaptive system operation in dynamic and unpredictable environments. (4) Sensing-communication-energy unified waveform design: At the core of the ISCCI paradigm, this technology develops multifunctional waveforms that simultaneously support sensing, communication, and energy transfer functionalities. IoT nodes can use the harvested wireless energy to power their sensing and communication functions, extending the network’s operational lifespan. Additionally, environmental sensing results can aid in channel reconstruction, reducing signaling overhead associated with channel information acquisition. By leveraging efficient sensing-communication-energy unified waveforms, the ISCCI paradigm is expected to significantly enhance system adaptability, scalability, and resource efficiency in resource-constrained environments. Prospects Despite these advancements, ISCCI systems still face several challenges. Future research will focus on the following areas to enable the practical deployment of ISCCI systems: (1) Theoretical Foundations: The absence of comprehensive theoretical models for ISCCI integration hinders the development of unified performance metrics. Future research should prioritize formulating mathematical models that account for the complex interactions between sensing, communication, computation, and intelligence. Additionally, there is a need for tools to evaluate system performance across diverse real-world scenarios. Research should also aim to integrate communication, sensing, and computation metrics, which are currently assessed using incompatible indicators. (2) Network Architecture: Designing flexible and adaptive network architectures is essential to support ISCCI functionalities. In large-scale IoT scenarios, such as smart cities, multi-base cooperative sensing networks with adaptive switching strategies are crucial for overcoming challenges like weak echoes and interference. Moreover, synchronizing massive node information presents significant challenges, as even small timing errors can severely impact accuracy. Research should focus on methods to address synchronization and improve network performance by leveraging cooperative edge nodes, as well as multi-node collaborative transmission and interaction mechanisms. (3) Interference Management: Managing interference is critical in dense IoT environments. Advanced algorithms for resource allocation, interference cancellation, and multi-user coordination are required to mitigate interference effects. AI-driven techniques can optimize terminal scheduling and resource allocation to avoid frequency interference. Additionally, edge computing capabilities will be essential for identifying and suppressing external interference, ensuring reliable communication, and enhancing ISCCI system performance. Addressing these challenges will be key to unlocking the full potential of ISCCI and ensuring its successful deployment in 6G IoT systems.
2025,
47(4):
909-920.
doi: 10.11999/JEIT241142
Abstract:
Objective Integrated Sensing and Communication (ISAC) is one of the six scenarios of 6G confirmed by the International Telecommunication Union (ITU). In particular, by enabling separable sensing transceivers, bi-static sensing is free from self-interference and can leverage ubiquitous network devices, making it an essential scenario for ISAC. However, bi-static sensing faces challenges due to non-idealities, including Timing Offset (TO), Timing Drift (TD), and Carrier Frequency Offset (CFO), which significantly affect signal detection and parameter estimation. Therefore, the suppression of sensing non-idealities is a key research area, as it directly influences the reliability of sensing results. Many researchers use proprietary datasets to investigate and suppress these non-idealities, which complicates fair and unified evaluations of different methods and technologies. Moreover, such reliance on specific experimental conditions hinders the reproducibility of relative studies. To support the development and standardization of ISAC techniques, a measured ISAC sensing signal dataset based on the 5G air interface has been constructed. This dataset enables the parallel comparison of various studies and facilitates research implementation even in the absence of specific experimental conditions. Methods This dataset utilizes Universal Software Radio Peripherals (USRPs), to operate in the sub-6 GHz frequency band and to run the 5G New Radio (NR) physical layer protocol stack, with the DeModulated Reference Signal (DMRS) in Physical Downlink Shared CHannel (PDSCH) reused as the sensing signal for data acquisition. The physical layer protocol stack is developed based on the NR protocol Release 15. The dataset comprises 2 scenarios and 2 sensing modes, resulting in a total of 8 data groups. The two sensing modes are bi-static and mono-static sensing, allowing for independent research on either sensing mode as well as comparative studies between the two. For mono-static sensing, a single USRP serves as the Base Station (BS), transmitting and receiving the sensing signal. For bi-static sensing, two USRPs are used: one acts as the BS and the other acts as the User Equipment (UE), with the BS transmitting the sensing signal and the UE receiving it. For both sensing modes, the transmitter uses a signal panel antenna, while the receiver is equipped with an antenna array consisting of 8 antenna elements. These 8 antenna elements correspond to 8 radio channels in the receiver, facilitating 8-channel reception. For each scenario and sensing mode, Channel State Information (CSI) from the 8 channels is provided over a continuous 30-second period, capturing both the moving sensing target and the background environment. Additionally, data corresponding only to the background environment is also included in this dataset. In each scenario, the positions and orientations of the transmitting and receiving antennas, as well as the moving trajectory of the sensing target, remain unchanged for both sensing modes. This ensures that the ground truth remains identical for both mono-static and bi-static sensing, enabling comparative research between the two sensing modes. Results and Discussions To provide a clearer demonstration of the dataset, this paper presents the delay spectrums and delay-Doppler spectrums of typical sensing signals using the classical 2-Dimensional Discrete Fourier Transformation (2D-DFT) algorithm, with corresponding analyses and descriptions. The delay-Doppler spectrums of mono-static sensing are much clearer (Fig. 7 ), with the sensing target easily detectable. However, the delay-Doppler spectrums of bi-static sensing exhibit significant dispersion (Fig. 8 ), which results from sensing non-idealities and hinders signal detection and parameter estimation. Therefore, suppressing sensing non-idealities is critical for improving bi-static sensing performance. As an example, this paper provides a reference path method in the delay domain, based on the oversampling Inverse Discrete Fourier Transformation (IDFT) algorithm, to mitigate sensing non-idealities in bi-static sensing and to validate the reliability and effectiveness of the dataset. The results demonstrate that the reference path method effectively suppresses the impact of sensing non-idealities (Fig. 9 ), yielding acceptable position measurements for the sensing target in bi-static sensing (Fig. 10 ). However, further research is needed to develop comprehensive solutions to address sensing non-idealities, which is the primary motivation for releasing this dataset. Conclusions Currently, there is a lack of an effective, standardized, and flexible dataset for sensing signals in ISAC based on air interfaces. Datasets derived from air interfaces in practical systems are critical foundations for research on bi-static sensing signal processing in 6G ISAC. To address this gap, this paper constructs and publicly releases an ISAC dataset based on the 5G air interface. The data is collected using USRPs running the 5G NR physical layer protocol stacks. Users can apply segmentation, decimation, or sliding-window extraction to the data to meet specific research needs. This dataset supports research on sensing non-idealities, signal detection, parameter estimation, clutter elimination, and sensing signal design. It facilitates independent research on mono-static and bi-static sensing, as well as comparative studies between the two sensing modes. Future efforts will focus on maintaining and expanding the dataset to include more complex scenarios, such as outdoor environments, low-altitude scenarios, and collaborative sensing.
2025,
47(4):
921-931.
doi: 10.11999/JEIT240644
Abstract:
Objective The increasing demand for wireless communication has led to a significant scarcity of spectrum resources, while the inherent coupling between communication and sensing systems allows for shared spectrum utilization. Integrated Sensing and Communication (ISAC) thus shows considerable promise for future applications. However, most existing studies focus on optimizing either communication or sensing performance, treating the other as a constraint, which limits system flexibility. This approach becomes particularly problematic in complex multi-user, multi-target scenarios, where balancing both functionalities is essential. Additionally, previous works often assume symmetrically distributed radar targets in small quantities, simplifying optimization but diverging from practical asymmetric and dense target distributions. To address these limitations, this study explores ISAC systems with asymmetric multi-target configurations, aiming to improve flexibility and practicality through joint optimization of communication and sensing performance optimization. Methods This study adopts an MIMO radar framework in which orthogonal transmit signals maximize waveform Degrees of Freedom (DoFs) in proportion to the antenna count. A beamforming matrix is designed to detect targets across multiple directions while allocating distinct waveforms for communication and sensing tasks. In contrast to conventional antenna configurations, the proposed scheme utilizes all antennas for radar detection, enhancing sensing performance. To address the limitations of single-objective optimization, a Pareto optimization framework is introduced, allowing for weighted trade-offs between the communication Weighted Sum Rate (WSR) and radar beam pattern error. This framework is adaptable to dynamic scenarios. To handle the non-convexity of the optimization problem, a hybrid algorithm combining Weighted Minimum Mean Square Error (WMMSE) and Semidefinite Relaxation (SDR) is proposed. Specifically, the WSR and radar error maximization problem is first reformulated as a Mean Square Error (MSE) minimization problem, followed by SDR-based relaxation of constraints for tractable solutions. Results and Discussions As shown in (Fig. 2 , Fig. 3 ): (a) The proposed beamforming design demonstrates superior flexibility compared to single-objective optimization, enabling adaptable balancing of communication and sensing performance across scenarios by adjusting the Pareto weight factor. (b) Compared to separated deployment schemes, the proposed method utilizes more antennas for sensing, concentrating transmit power in specific directions to enhance target detection capability. (c) With comparable radar performance, the communication WSR of the proposed scheme shows an 11.6% improvement over shared deployment configurations.(Fig. 4 ) further illustrates the radar detection error under varying SNRs. Regardless of the performance weight values, the radar detection error decreases with increasing transmit power, indicating that higher power improves system performance. Under constant transmit power, a smaller performance weight results in higher radar detection accuracy, as more power is allocated to radar performance optimization. For a more comprehensive comparison, (Fig. 5 ) shows the beamforming patterns under different transmit power levels for the separated deployment scheme. In this scheme, as transmit power increases, the radar detection error actually increases. This occurs because the system optimizes detection performance in a specific direction, achieving optimal precision there. As shown in the figure, as power increases, the antenna power becomes concentrated in the direction of the target at 0°, significantly improving resolution in that direction, while detection performance in other directions deteriorates. This indicates that the separated deployment scheme is limited in its ability to meet detection requirements for multiple targets simultaneously. (Fig. 6 ) demonstrates that, at all transmit power levels, the proposed scheme exhibits clear advantages in communication performance. (Fig. 8 , Fig. 9 ) analyze the impact of target quantity on radar detection error, confirming robustness in multi-target asymmetric scenarios. When transmit power and weight factors remain unchanged, increasing the number of radar detection targets leads to an increase in radar detection error. This happens because total power remains constant, and adding more detection targets reduces the power allocated to each target. This effect becomes more pronounced as the number of targets grows. However, this error increase can be mitigated by increasing transmit power. Simulation results show that the proposed scheme consistently outperforms other methods under different target numbers, demonstrating its ability to efficiently utilize limited power and maintain low detection errors, even as the number of targets increases. (Fig. 10 ) reveals directional limitations, as beam patterns at edge angles exhibit weak directivity, complicating peripheral target detection. Algorithm convergence curves (Fig. 11 ) and Pareto frontiers for communication-radar trade-offs (Fig. 12 ) confirm the stability and flexibility of the proposed scheme. Conclusions This study addresses the limitations of single-objective optimization (communication or radar performance) and constrained radar degrees of freedom by proposing a weighted joint beamforming design for ISAC. The ground base station, equipped with dual functionalities, optimizes the WSR and radar beam pattern error. By adjusting the Pareto weight factor, flexible performance trade-offs between communication and sensing are achieved, improving adaptability to diverse scenarios. Experimental results demonstrate that, under optimized weights and transmit SNR, the proposed scheme reduces radar detection error by 36.2% and enhances communication SINR by 1 dB compared to separated deployment strategies. These advancements validate the effectiveness of the joint optimization framework in practical asymmetric multi-target environments, providing a robust foundation for next-generation ISAC systems.
2025,
47(4):
932-947.
doi: 10.11999/JEIT241114
Abstract:
Objective The Sixth-Generation (6G) mobile communications network will evolve from being human-centered to agent-centered, enabling a deep integration of multi-dimensional functions such as communication, sensing, and computation. It will further advance key physical layer technologies, including large arrays, broad bandwidth, multi-frequency bands, and multi-node collaboration, giving rise to the Integrated Sensing And Communications (ISAC) system. The ISAC system will support communication services while leveraging communication signals to sense and monitor comprehensive information about the physical world. This will empower a range of services, including low-altitude economy, digital twins, Internet of Vehicles, industrial Internet, and smart cities. However, realizing comprehensive sensing of the physical world while maintaining communication performance remains a critical challenge that requires further research. Methods This study presents a framework to deconstruct the physical world into static environments, dynamic targets, and various object materials. The static environment, which includes buildings, roads, trees, and other structures, constitutes the majority of the physical world. Sensing the static environment is fundamental for the sensory system’s understanding of the physical world. A multi-user, multi-Base Station (BS), and active-passive fusion approach for Static Environment Reconstruction (SER) is proposed. This method constructs two-dimensional or three-dimensional maps of the physical world by analyzing environmental scattering point data collected during communication between the BS and the user. Dynamic targets within the static environment, including pedestrians, vehicles, and drones, contribute to the spatiotemporal movement of the physical world. Sensing these dynamic targets is vital for enabling the sensory system to support various applications in production and daily life. A Dynamic Target Sensing (DTS) technology is proposed, leveraging multi-BS collaboration and multi-feature fusion. This technology actively transmits detection signals from the BS and receives target echo signals, enabling the monitoring of the existence, position, velocity, and category of dynamic targets. Object materials, such as metal, wood, and fabric, influence the propagation laws and interaction patterns of electromagnetic signals in the physical world. Thus, sensing the material properties of objects is crucial for the sensory system’s analysis of the fundamental laws governing the physical world. To address this, a material recognition technology based on multi-BS collaboration is proposed, which identifies the material properties of target objects by analyzing the electromagnetic coefficients of the scattering points in the BS-object-user channel. Results and Discussions Building on theoretical research, this paper presents the development of a universal ISAC hardware prototype platform based on RF System-on-Chip (RFSoC) and Field Programmable Gate Array (FPGA). With the implementation of a self-developed ISAC baseband algorithm, the platform enables real-time sensing of dynamic targets and accurate mapping of static environments. Conclusions This paper proposes a synesthesia ISAC framework based on the concept of "separation of dynamic and static," which thoroughly analyzes the interaction between the physical world and electromagnetic wave signals. It decomposes the sensing of the physical world into SER, DTS, and Object Material Recognition (OMR), thereby providing substantial support for the ultimate goal of synesthesia—accurately replicating the real physical world into a digital twin.
2025,
47(4):
948-957.
doi: 10.11999/JEIT240716
Abstract:
Objective Traditional approaches typically adopt a disjoint design that improves specific performance aspects under particular scenarios but often proves inadequate for addressing complex tasks in dynamic environments. Challenges such as real-time task offloading, efficient resource scheduling, and the simultaneous optimization of sensing, communication, and computing performance remain significant. The Integrated Sensing, Communication, and Computing (ISCC) architecture has been proposed to address these issues. In complex scenarios, the diversity of task types and varying requirements lead to inflexible offloading policies, limiting the system’s ability to adapt to real-time network changes. Moreover, computational uncertainty can undermine the robustness of resource scheduling, potentially resulting in performance degradation or task failure. Effectively addressing challenges like high user energy consumption and computational uncertainty while maintaining service quality is crucial for optimizing future network nodes. As network environments grow increasingly complex and user demands for high performance, low latency, and robust reliability rise, the optimization of resource efficiency and the achievement of mutual benefit across sensing, communication, and computing functions become urgent and critical. To meet this challenge, it is essential to advance the system towards higher intelligence and multi-dimensional connectivity. Furthermore, research on robust offloading in ISCC networks remains limited and warrants further investigation. Methods To address high user energy consumption and computational uncertainty in ISCC networks under complex scenarios, a robust resource allocation and decision optimization scheme is proposed. The goal is to minimize the total energy consumption of users. The proposed scheme takes into account common constraints and computational uncertainty commonly encountered in practical applications, offering a viable optimization approach for ISCC network design. First, to tackle the challenge of accurately predicting task complexity, potential biases arising from resource allocation and processing estimations are analyzed. These biases reflect real-world unpredictability, where task size can be measured but completion time remains uncertain, potentially leading to resource waste or performance degradation. To mitigate this, a robust computational resource allocation problem is formulated to manage the uncertainty caused by task offloading effectively. Second, the problem of minimizing users’ total energy is established by jointly optimizing task offloading ratios, beamforming, and resource allocation, subject to constraints such as power consumption, processing time, and radar estimation information rate. However, due to the multi-variable, non-convex, and NP-hard nature of this optimization problem, traditional methods fail to provide efficient solutions. To address this, a Markov decision process is modeled, and an optimization algorithm based on Soft Actor-Critic (SAC) is proposed. Results and Discussions The simulation results demonstrate that the proposed SAC-based algorithm outperforms existing methods in terms of performance and flexibility in dynamic and complex scenarios. Specifically, the learning rate affects the convergence speed of the algorithm, but its impact on final performance is minimal (Fig. 3 ). Compared to the Proximal Policy Optimization (PPO) and Advantage Actor-Critic (A2C) algorithms, the proposed algorithm achieves faster training speeds. Thanks to its flexible and unique design, the proposed algorithm exhibits stronger exploration capabilities and remains more stable during training (Fig. 4 ). The robust design enhances adaptability, resulting in higher overall reward values (Fig. 5 ). In terms of total user energy consumption, the proposed algorithm reduces energy use by approximately 9.57% compared to PPO and by 40.72% compared to A2C. As the number of users increases and more users access the network, signal interference intensifies, transmission rates decrease, and task offloading costs rise. In such scenarios, the proposed algorithm shows greater flexibility in policy adjustment, maintaining energy consumption at a relatively low level, outperforming both PPO and A2C. This advantage becomes more pronounced as the number of users grows or load pressure increases (Fig. 6 ). Overall, the proposed algorithm offers a robust and efficient solution for resource allocation and optimization in dynamic and complex environments, demonstrating exceptional adaptability and reliability in multi-user and multi-task scenarios. These results not only highlight the superior performance of the SAC algorithm but also highlight its potential in addressing multi-variable, non-convex problems. Conclusions This paper presents an optimization algorithm based on SAC, which not only achieves outstanding performance in terms of energy consumption, latency, and task offloading efficiency but also demonstrates excellent scalability and adaptability in multi-user, multi-task, and complex scenarios. A robust computational resource allocation scheme is proposed to address the uncertainty in offloading decisions. Simulation results show that the proposed algorithm can adapt to complex and dynamic network environments through flexible policy decisions, providing both theoretical support and a technical reference for further research on ISCC networks in such scenarios. Future research could explore incorporating multi-base station collaboration to enhance the robustness of ISCC networks, enabling them to better handle even more complex network environments.
2025,
47(4):
958-967.
doi: 10.11999/JEIT240446
Abstract:
Objective In the future, 6G will usher in a new era of intelligent interconnection and the integration of virtual and physical environments. This vision relies heavily on the deployment of numerous communication and sensing devices. However, the scarcity of frequency resources presents a significant challenge for sharing these resources efficiently. Integrated Sensing and Communication (ISAC) technology offers a promising solution, enabling both communication and sensing to share a common set of equipment and frequency resources. This allows for simultaneous target detection and information transmission, positioning ISAC as a key technology for 6G. ISAC research can be divided into two main approaches: Coexisting Radar and Communication (CRC) and Dual-Function Radar Communication (DFRC). The CRC approach designs separate systems for radar and communication, aiming to reduce interference between the two; however, this leads to increased system complexity. The DFRC approach integrates radar and communication into a single system, simplifying the design while still achieving both radar detection and communication functions. As a result, DFRC is the primary focus of ISAC research. Waveform design is a crucial component of ISAC systems, with two primary strategies: non-overlapped resource waveform design and fully unified waveform design. The fully unified design can be further classified into three types: sensing-centric, communication-centric, and joint design. Previous research has predominantly focused on sensing-centric or communication-centric designs, which limit the flexibility of the integrated waveform in balancing communication and sensing performance. Additionally, limited research has addressed ISAC in marine environments. This paper investigates waveform design for ISAC in marine environments, proposing a joint design approach that uses a weighting coefficient to adjust the communication and sensing performance of the integrated waveform. Methods Considering the characteristics of the marine environment, this paper proposes using Unmanned Aerial Vehicles (UAVs) as nodes in the ISAC system, owing to their flexibility, portability, and cost-effectiveness. The integrated waveform transmitted by UAVs can both communicate with downlink users and detect targets. The communication performance is evaluated using the achievable sum rate, while the sensing performance is assessed by the error between the covariance matrix of the integrated waveform and the standard radar covariance matrix. The optimization objective is to maximize the weighted sum of these two performance indices, subject to the constraint that UAV power does not exceed the maximum allowable value. The weighting coefficient represents the ratio of communication power to sensing power. Due to the non-convex rank-1 constraint and objective function, the optimization problem is non-convex. This paper decomposes the non-convex optimization problem into a series of convex subproblems using the Successive Convex Approximation (SCA) algorithm. The local optimal solution of the original problem is obtained by solving these convex subproblems. The communication and sensing performance of the integrated waveform can be adjusted by varying the weighting coefficient. The performance of the weighted integrated waveform design in a marine environment is simulated, and the results are presented. Results and Discussions Simulation results indicate that the integrated beam pattern exhibits two large lobes: one directed towards the target for detection, and the other towards the communication user (Fig.4). As the weighting coefficient increases, the lobes directed towards the communication users become more pronounced, reflecting the increased emphasis on communication performance. Furthermore, as the weighting coefficient increases, the sensing performance error (smaller error indicates better sensing performance) initially increases slowly before rising more rapidly. Meanwhile, the achievable sum rate of communication increases sharply. Eventually, both the sensing performance error and the communication sum rate curves flatten out (Fig. 6 ). Since the UAV’s maximum power is limited to 10 W, further increases in the weighting coefficient beyond a certain point lead to diminishing returns in communication performance, as power constraints limit further improvement. At this point, the sensing performance error remains stable. Conclusions This paper investigates the waveform design for UAV-enabled ISAC systems in marine environments. A wireless propagation model for UAVs in such environments is developed, and an integrated waveform optimization method based on a weighted design is proposed. The SCA algorithm is used to solve the convex approximation. Simulation results demonstrate that when the weighting coefficient is between 0.2 and 0.5, the integrated waveform ensures strong communication performance while maintaining good sensing performance.
2025,
47(4):
968-978.
doi: 10.11999/JEIT240995
Abstract:
Objective With the rapid development of Low Earth Orbit (LEO) satellite constellations and Integrated Sensing And Communication (ISAC) systems, performance optimization faces increasing challenges due to fixed power distribution, spectrum limitations, and interference between communication and sensing functions. This study proposes an optimization method based on multi-dimensional resource joint scheduling to address these constraints in LEO satellite environments. The method enhances the combined performance of communication and sensing by leveraging the high satellite visibility of LEO constellations. The optimization focuses on improving communication reach, data rate, radar mutual information, and positioning accuracy while ensuring efficient resource allocation. Methods The optimization problem is formulated as a multi-variable joint problem, incorporating satellite selection, subchannel function allocation, and power distribution. To address the complexity of this Mixed-Integer NonLinear Programming (MINLP) problem, it is decoupled into subproblems and solved iteratively using the Block Coordinate Descent (BCD) method. Satellite selection is optimized using a modified Multi-Population Genetic Algorithm (MPGA), which accounts for communication link quality, sensing capabilities, and satellite geometric distribution. Subchannel allocation and power distribution are iteratively optimized to maximize system performance while maintaining a balance between communication and sensing tasks. Results and Discussions The proposed optimization method is evaluated through simulations against benchmark schemes. Results indicate that, under the same resource constraints, the method enhances integrated communication and sensing performance by over 7% (Fig. 5 ). Improvements are observed in communication efficiency, radar detection mutual information, and positioning accuracy. Additionally, the number of cooperating satellites significantly affects system performance, though gains diminish beyond an optimal threshold (Fig. 4 ). This highlights the importance of strategic satellite selection and coordination to balance performance gains with complexity and resource usage. Moreover, the results confirm the convergence of the proposed method, demonstrating consistent performance across multiple scenarios (Fig. 3 ). Conclusions This study proposes an optimization approach for ISAC systems in LEO satellite constellations, addressing challenges related to resource allocation, power distribution, and interference management. The multi-dimensional resource joint scheduling method enhances overall system performance by optimizing satellite selection, subchannel allocation, and power distribution. Simulation results demonstrate that: (1) The proposed optimization method improves integrated communication and sensing performance in LEO satellite ISAC systems, achieving a performance gain of over 7% compared to benchmark solutions. (2) The multi-dimensional resource joint scheduling approach effectively balances communication and sensing tasks by optimizing satellite selection, subchannel function allocation, and power distribution, thereby mitigating interference and resource constraints. (3) The number of cooperating satellites significantly influences system performance. However, beyond an optimal threshold, additional satellites yield diminishing returns, emphasizing the need for efficient satellite coordination. This study assumes ideal sensing capabilities; future research should incorporate real-world constraints, such as satellite mobility and environmental factors, to enhance the practical applicability of the proposed approach.
2025,
47(4):
979-990.
doi: 10.11999/JEIT240566
Abstract:
Objective As wireless communication technology advances, sensing and communication systems are shifting toward higher frequency bands, larger antenna arrays, and miniaturization. This integration of hardware architecture, channel characteristics, and signal processing enables wireless infrastructure to support environmental sensing in addition to communication. Technologies such as millimeter-wave communication, Reconfigurable Intelligent Surface (RIS), and Integrated Sensing And Communication (ISAC) facilitate this development. Although extensive research has examined RIS applications in ISAC systems, expanding the RIS aperture fundamentally alters electromagnetic field characteristics, extending the near-field range. Unlike far-field scenarios, near-field communication and sensing exhibit more complex channel structures, posing challenges for RIS-assisted millimeter-wave systems. To address these challenges, this study proposes an ISAC framework and develops a nested tensor-based Simultaneous Localization And Communication (SLAC) scheme. This approach localizes scattering points and users while detecting information symbols in near-field environments, eliminating the need for dedicated pilot signals. Methods First, a near-field spherical wave transmission model is established. To mitigate the complexity introduced by spatial path variations across reflection units, a channel model based on the second-order Taylor approximation is derived, incorporating distance, direction of arrival, and angle of arrival. Next, to fully utilize the time redundancy of Khatri-Rao Space-Time (KRST) coding, the received signal is formulated as a nested tensor model comprising outer and inner PARAFAC tensors, enabling the development of a nested tensor-based SLAC scheme. For the outer PARAFAC tensor, an Alternating Least Squares (ALS) algorithm is employed for channel matrix estimation and information symbol detection. For the inner PARAFAC model, a two-stage algorithm is used for channel parameter estimation and User Equipment (UE) and scatterer localization. The Minimum Description Length (MDL) method determines the number of transmission channel paths. In the first stage, the ALS method decomposes the PARAFAC model to estimate channel parameters. In the second stage, the ESPRIT algorithm is applied to refine parameter estimation and perform localization. Finally, the estimated channel parameters are used to determine the locations of the UE and scatterer points. Results and Discussions The proposed scheme first utilizes the multi-dimensional resources of the ISAC scenario and the KRST coding method to structure the received ISAC signals into a fourth-order nested tensor. Leveraging the algebraic properties of the nested tensor and the second-order Fresnel approximation of the near-field channel model, the nested tensor-based SLAC scheme is designed to enable near-field localization of scattering points and UE, as well as information symbol detection. Simulation results demonstrate that the proposed scheme achieves superior ISAC performance compared with existing methods (Fig. 2 , Fig. 3 , Fig. 4 ). Performance improves as the number of subcarriers increases (Fig. 2 , Fig. 3 ). Additionally, the scheme maintains high localization accuracy and symbol detection performance even under higher-order modulation (Fig. 5 , Fig. 6 ). Further improvements in ISAC performance are observed with an increased number of time slots and coding length (Fig. 7 , Fig. 8 ). The results also indicate good convergence across various parameter configurations. Conclusions This paper proposes an RIS-assisted ISAC millimeter-wave near-field transmission scheme based on a nested tensor model and develops a nested tensor-based SLAC scheme leveraging the second-order Fresnel approximation of the near-field channel model. The constructed nested tensor model exhibits an algebraic structure, enabling the proposed scheme to operate without dedicated pilot signals. Moreover, the model integrates multiple dimensions of sensing and communication signals, enhancing information symbol detection and target localization accuracy by extracting additional useful information. Simulation results demonstrate that the proposed method achieves good convergence across various parameter configurations. Compared with existing methods, it exhibits superior sensing performance. Under higher-order modulation, it maintains excellent information symbol detection and achieves high-precision channel state information recovery, providing centimeter-level localization accuracy. Furthermore, the method is scalable and can be applied to larger-scale systems, such as expanding RIS or increasing the number of antennas. However, system scalability increases computational complexity, particularly for higher-order tensor models. To address this, optimizing the algorithm structure, such as introducing tensor-based closed-form algorithms (e.g., higher-order singular value decomposition), is a promising approach.
2025,
47(4):
991-1003.
doi: 10.11999/JEIT240601
Abstract:
Objective: Next-generation communication networks will enhance converged “endogenous sensing” and communication service capabilities by improving information transmission. Integrated Sensing and Communication (ISAC) is a key technology for achieving the 6G vision and has attracted significant Attention from both academia and industry. The integration of ISAC with emerging technologies, such as Reconfigurable Intelligent Surface (RIS) and Movable Antenna (MA), is currently a hot research topic. Because the same waveforms are used for both communication and target sensing, ISAC systems are more vulnerable to information leakage. Unlike Physical Layer Security (PLS)-based designs, it is necessary not only to prevent the signals of legitimate users from being eavesdropped but also to hide the existence of communication behavior activities from malicious targets. This paper examines a generic Integrated Sensing and Covert Communication (ISCC) system involving multiple sensing targets (wardens) and multiple covert users. To facilitate communication between the Base Station (BS) and legitimate users, a simultaneously transmitting and reflecting RIS with movable elements (ME-STAR-RIS) is deployed. Inspired by the MA concept, the ME-STAR-RIS features movable elements that allow for Flexible And Passive Beamforming (FAPB). A key challenge is to design a rational architecture that minimizes the control cost of the ME-STAR-RIS. Our goal is to create an effective beamforming and element deployment strategy for this system and to investigate the benefits of element-level movement at the STAR-RIS. Methods: First, a Discrete Element Position (DEP)-based coupled phase-shift model for STAR-RIS is proposed. This model aims to reduce control costs associated with the movability and phase shifts of STAR-RIS elements. Then, a joint beamforming optimization problem is formulated based on this model. The goal is to jointly optimize active beamforming at the ISAC BS and flexible passive beamforming (including element positions, phase shifts, and amplitude coefficients) at the ME-STAR-RIS. This is intended to maximize the probing beam gain at the sensing target while adhering to covert communication quality constraints. The problem formulated is non-convex and presents strong coupling, making it challenging to solve. To address this, we develop an effective algorithm leveraging Semi-Definite Program (SDP), Block Coordinate Descent (BCD), Successive Convex Approximation (SCA), and Penalty Convex-Concave Procedure (PCCP) techniques. By introducing auxiliary variables and employing the SDP method, the original problem can be transformed into a more manageable Augmented Lagrangian form. Our approach features a two-layer iterative algorithm. In the inner loop, the element placement problem is modeled as a binary integer programming problem, using a penalty-based SCA method to solve it. In the outer layer, a penalty-based BCD method is proposed to maintain constraints on the coupled STAR-RIS phase shift upon convergence. Results and Discussions: The simulation results validate the effectiveness of the proposed algorithm and provide significant insights. The performance evaluation indicates that the STAR-RIS with 15 movable elements achieves 80% of the performance of a fixed full-array STAR-RIS with 30 elements while halving the required elements. This highlights the potential for a limited number of movable elements to approximate the performance of a fully fixed array. Furthermore, the proposed algorithm consistently converges to a high-performance smooth point, meeting constraints on array element positions and phase shift differences. The results also show that moving the elements leads to a narrower and stronger detection beam, enhancing the system’s performance. Additionally, the findings reveal a trade-off between communication, sensing, and covert presence. Specifically, as the communication Signal-to-Interference-Noise Ratio (SINR) threshold increases, the sensing performance decreases. Due to covert communication constraints, beamforming design freedom is limited, requiring additional system resources for covertness, which ultimately reduces overall sensing performance. Conclusions: This paper examines the ME-STAR-RIS-assisted pass-sense integrated system through the lens of covert communication. The BS senses target nodes and communicates with legitimate users via an ME-STAR-RIS. To ensure data security, it is essential to conceal communication activities from potential targets. A joint active-passive covert beamforming scheme designed for the ME-STAR-RIS-assisted ISAC system is designed. This scheme aims to maximize probing power while maintaining covert communication quality. This paper serves as an initial exploration of the STAR-RIS with movable elements. Simulation results indicate that element-level mobility offers advantages for the STAR-RIS-assisted ISAC system. Several issues warrant further investigation, including channel estimation, non-ideal Channel State Information (CSI), and optimization of array element positions in practical settings.
2025,
47(4):
1004-1013.
doi: 10.11999/JEIT240606
Abstract:
Objective With the widespread application of Unmanned Aerial Vehicle (UAV) communication technology in military and civilian domains, ensuring secure information transmission within UAV networks has received increasing attention. Covert communication is an effective approach to conceal information transmission. However, existing methods, such as digital beamforming, improve covert communication performance but increase system size and power consumption. This study proposes a UAV short-packet covert communication method based on Time Modulated Planar Array (TMPA) sensing. A TMPA-UAV covert communication system architecture is introduced, along with a two-dimensional Direction of Arrival (DOA) estimation method. A covert communication model is established, and a closed-form expression for the covert constraint is derived using Kullback-Leibler (KL) divergence. Based on the estimated angle of Willie, the TMPA switching sequence is optimized to maximize signal gain in the target direction while minimizing gain in non-target directions. Covert throughput is selected as the optimization objective, and a one-dimensional search method determines the optimal data packet length and transmission power. Results and Discussions Simulations show that the Root Mean Square Error (RMSE) for DOA estimation in both directions approaches 0°, with RMSE decreasing significantly as the Signal-to-Noise Ratio (SNR) increases (Fig. 4 ). With a fixed elevation angle and azimuth angles varying between 0° and 60°, a comparison between the proposed method and the traditional DOA estimation method for time-modulated arrays indicates that the proposed method reduces DOA estimation error to the order of 0.1°, significantly improving accuracy. Beamforming simulations based on the estimation results (Fig. 6 ) show a SideLobe Level (SLL) below –30 dB and a beamwidth of 5°, meeting design requirements. Covert communication simulations reveal the existence of an optimal data packet length that maximizes covert throughput (Fig. 7 ). A stricter covert tolerance imposes tighter constraints on covert communication (Fig. 8 ), requiring Alice to use lower transmission power and shorter block lengths to communicate covertly with Bob. When the beamforming error angle is small, the system maintains high covert throughput (Fig. 9 ). Within a UAV flight height range of 50 m to 90 m, covert throughput remains low; however, when the height exceeds 90 m, throughput increases rapidly. Beyond 130 m, UAV height has little effect on maximum covert throughput, and performance reaches its optimal state. Therefore, controlling UAV flight height appropriately is crucial for effective communication between legitimate links. Conclusions This study proposes a TMPA-based multi-antenna UAV sensing-assisted covert communication system for short packets. A TMPA-based DOA estimation method is introduced to determine the relative position of non-cooperative nodes. The Compressed Sensing (CS) algorithm optimizes the beam radiation pattern, maximizing gain at the legitimate destination node while creating nulls at the non-cooperative node's location. A closed-form expression for covert constraints is derived using KL divergence, and covert throughput is maximized through the joint optimization of packet length and transmission power. Simulations analyze the relationships between the number of array elements, covert tolerance, beam direction error angles, UAV height, and covert throughput. Results indicate that an optimal packet length maximizes covert throughput. Additionally, increasing the number of array elements and relaxing covert constraints can improve covert throughput. Practical system design should comprehensively optimize these factors.
2025,
47(4):
1014-1025.
doi: 10.11999/JEIT240640
Abstract:
Objective Breathing rate is a vital physiological indicator of human health. Abnormal changes in this rate can signify diseases like chronic obstructive pulmonary disease, sleep apnea syndrome, and nocturnal hypoventilation syndrome. Timely and accurate detection of these changes can help identify health risks early, enable professional medical intervention, and optimize treatment timing, thereby improving overall health. However, current detection methods often face limitations due to noise interference and “blind spot” issues, which impact accuracy and robustness. To address these challenges, this paper employs Wi-Fi devices to measure indoor human breathing rates using Integrated Sensing And Communication (ISAC) technology. By combining Variational Modal Decomposition (VMD) and Hilbert-Huang Transform (HHT), a new breathing rate sensing algorithm is proposed. This approach aims to enhance detection accuracy and robustness, resolve the “blind spot” problem in existing technologies, and offer an efficient and reliable solution for health monitoring. Methods Wi-Fi links with high environmental sensitivity were selected to construct the Channel State Information (CSI) ratio model. Subcarriers of the filtered CSI ratio time series were projected, and amplitude and phase information were combined to generate a candidate set of breathing mode signals. For each subcarrier, the sequence with the highest short-term breath noise ratio, determined by periodicity, was identified as the final breath pattern. A threshold was then applied to select relevant subcarriers. Time-frequency analysis using VMD and HHT eliminated modal components unrelated to the human breath rate, and the remaining components were reconstructed. Principal Component Analysis (PCA) was applied for dimensionality reduction, selecting components accounting for over 99% of the variance. The ReliefF algorithm was subsequently used to reconstruct the breath signal into a fused signal, from which the breathing rate was calculated using a peak detection algorithm. Results and Discussions Experiments were conducted in two scenarios: a conference office and a corridor. In both setups, a pair of transceivers was deployed, with a 2-meter distance maintained between the transmitter and receiver. The transmitter used one omnidirectional antenna, and the receiver had three antennas positioned perpendicular to the ground. Participants were seated on the vertical bisector of the Line Of Sight (LOS) path, synchronizing their breathing with a metronome as CSI data were recorded. Each test lasted 1 minute, with a confirmed breathing rate of 16 bpm. System parameters used in the experiments are detailed in Table 1. In the conference office scenario, this paper collected data at various distances from the participant to the transceiver. As illustrated in Figure 9, the Mean Estimation Accuracy (MEA) of our algorithm remains above 97%, even when the participant is 5 meters away. In contrast, the MEA of the other two methods drops by 4% and 5%, respectively. As the sensing distance increases, the multipath effect intensifies, leading to a gradual weakening of the reflected signal and greater noise interference. This impact significantly challenges the breathing detection accuracy of the other methods. The algorithm presented in this paper incorporates a VMD-HHT time-frequency analysis step. This enhancement allows for effective signal decomposition and feature extraction, markedly improving the accuracy of detecting the target breathing signal. Moreover, the method exhibits strong adaptability and robustness, effectively addressing noise interference and multipath effects in complex environments, thus demonstrating more stable performance. In the corridor scenario, we evaluated the algorithm’s performance at varying distances. The average absolute error of the algorithm was measured with distances ranging from 2 meters to 5 meters. At 2 meters, the Mean Absolute Error (MAE) recorded was 0.37 bpm, and even at 5 meters, the MAE only increased to 0.45 bpm, remaining below 0.5 bpm. As the distance between the target and transceiver increased from 3 to 5 meters, the MAE gradually rose. This trend is attributed to the further attenuation of the signal reflected from the human target, along with the escalating multipath and signal attenuation effects in the environment. Conclusions The experimental results indicate that the MEA of this sensing method exceeds 97% in both the conference office and corridor scenarios. This effectively addresses the "blind spot" issue present in current technologies. The enhanced accuracy and robustness of the algorithm outperform existing sensing schemes. Moreover, this method broadens the application of ISAC in breathing detection and opens new avenues for developing intelligent health management systems in the future.
2025,
47(4):
1026-1042.
doi: 10.11999/JEIT240719
Abstract:
Objective Vehicular networks, as key components of intelligent transportation systems, are encountering increasing spectrum resource limitations within their dedicated 25 MHz communication band, as well as challenges from electromagnetic interference in typical communication environments. To address these issues, this paper integrates cognitive radio technology with radar sensing and introduces Hybrid-Reconfigurable Intelligent Surface (H-RIS) to jointly optimize radar sensing, data transmission, and computation. This approach aims to enhance spectrum resource utilization and the Joint Throughput Capacity (JTC) of vehicular networks. Methods A phased optimization approach is adopted to alternately optimize power allocation, time allocation, and reflection components in order to identify the best solution. The data transmission capacity of secondary users is characterized by defining a performance index for JTP. The problem is tackled through a two-stage optimization strategy where power allocation, time allocation, and reflection element optimization are solved iteratively to achieve the optimal solution. First, a joint optimization problem for sensing, communication, and computation is formulated. By jointly optimizing time allocation, H-RIS reflection element coefficients, and power allocation, the goal is to maximize the joint throughput capacity. The block coordinate descent method decomposes the optimization problem into three sub-problems. In the optimization of reflection element coefficients, a stepwise approach is employed, where passive reflection elements are fixed to optimize active reflection elements and vice versa. Results and Discussions The relationship between joint throughput and the number of iterations for the proposed Alternating Optimization Iterative Algorithm (AOIA) is shown (Figure 4 ). The results indicate that both algorithms converge after a finite number of iterations. The correlation between the target secondary user’s joint throughput and radar power is presented (Figure 5 ). In the H-RIS-assisted Integrated Sensing Communication and Computation Vehicle-to-Everything (ISCC-V2X) scenario, the joint throughput of the Aimed Secondary User (ASU) is maximized through optimal power configuration (Figure 5 ). The comparison of the target secondary user’s joint throughput with radar system power for the proposed algorithm and baseline schemes is shown (Figure 6 ), demonstrating that the proposed method significantly outperforms random Reconfigurable Intelligent Surfaces (RIS) and No-RIS schemes under the same parameter settings. Furthermore, the proposed H-RIS optimization scheme outperforms both Random H-RIS and traditional passive optimization RIS in terms of joint throughput.The relationship between the target secondary user’s joint throughput and the number of H-RIS reflection elements is illustrated (Figure 7 ). The results show that the proposed scheme provides a significant performance improvement over both Random RIS and No-RIS schemes under the same parameter settings. The relationship between the transmit power of the target secondary user’s joint throughput and the transmit power of the ASU is depicted (Figure 9 ), highlighting that joint throughput increases with transmit power in all scenarios. The relationship between joint throughput and the number of active reflection elements for the proposed algorithm and other benchmark schemes is shown (Figure 10 ), demonstrating that joint throughput increases with the number of active reflection elements in H-RIS scenarios, with the proposed scheme exhibiting a faster growth rate than Random H-RIS. The relationship between ASU joint throughput, radar sensing time, and radar power is presented (Figure 11 ), revealing that an optimal joint time and power allocation strategy exists. This strategy maximizes ASU joint throughput while ensuring H-RIS presence and sufficient protection for the primary user. Conclusion To address the challenges of spectrum resource scarcity and low data transmission efficiency in vehicular networks, this paper focuses on improving the joint throughput of intelligent vehicle users, enhancing spectrum utilization, and achieving efficient data transmission in the H-RIS-assisted ISCC-V2X scenario. A joint optimization method for vehicular network perception, communication, and computation based on H-RIS is explored. The introduction of H-RIS aims to enhance data transmission efficiency while considering the interests of both primary and secondary users. The joint optimization problem for the target secondary user’s perception, communication, and computation is analyzed. First, the joint allocation scenario for the H-RIS-assisted ISCC-V2X system is constructed, introducing the signal model, radar perception model, communication model, and computation model. Using these models, a joint optimization problem is formulated. Through alternating optimization, the optimal H-RIS reflection element coefficients, time allocation vector, and power allocation vector are derived to maximize the joint throughput. Simulation results demonstrate that the incorporation of H-RIS significantly improves the joint throughput of the target secondary user. Furthermore, an optimal power allocation scheme is identified that maximizes the joint throughput. When both time allocation and power allocation are considered jointly, simulations show the existence of an optimal scheme that maximizes the joint throughput of the target secondary user.
2025,
47(4):
1043-1051.
doi: 10.11999/JEIT240842
Abstract:
Objective To address the issue of limited V2X communication and sensing paths in 6G dense urban environments, an RIS-assisted ISAC-V2X system framework is proposed. Considering vehicle mobility under Non-Line-of-Sight (NLOS) conditions, the Extended Kalman Filter (EKF) algorithm is utilized to track and predict the positions of moving vehicles by combining real-time Channel State Information (CSI) from the ISAC echo signals. A multi-vehicle power allocation optimization scheme based on Non-Orthogonal Multiple Access (NOMA) is introduced to enhance the downlink communication sum rate while maintaining sensing accuracy. The Karush-Kuhn-Tucker (KKT) conditions are incorporated as a feedback mechanism to prevent the system from converging to a local optimum. Simulation results demonstrate that the proposed system outperforms the traditional RIS-assisted ISAC-V2X system in terms of both communication and sensing performance. Methods This study establishes an RIS-assisted ISAC-V2X-NOMA system model. Considering vehicle mobility in NLOS conditions, the EKF algorithm is employed to track and predict vehicle locations base on real-time CSI from the ISAC signals. Subsequently, a multi-vehicle power allocation optimization scheme based on NOMA is proposed, with the KKT conditions introduced to avoid local optima and ensure global optimality. To comprehensively evaluate channel estimation performance, 1000 Monte Carlo simulations are conducted, and performance analyses are carried out on MATLAB with comparisons to traditional RIS-assisted ISAC-V2X systems under different scenarios, ultimately validating the superiority of the proposed system. Results and Discussions The sensor tracking performance of the proposed system is presented, which indicate that the introduction of RIS significantly improves the angle and distance tracking accuracy. As the number of RIS reflection elements increases, the system’s Root Mean Square Error (RMSE) decreases, validating the effectiveness of RIS in complex dynamic environments. Then, the communication performance analysis between the proposed system and the traditional system under different antenna configurations is presented, where one can observe that the communication sum rate increases as the vehicle approaches the RIS surface and decreases as it moves away, which can be also improved by increasing the number of antennas. In dense environments with limited resources, the proposed system obviously outperforms the traditional system in terms of communication sum rate under the same RIS configuration. Finally, one can also observe that power allocation optimization using NOMA allows more efficient resource management and reduced inter-user interference, further improving communication rates. These results demonstrate the significant advantages of the proposed system in terms of both communication and sensing performance in V2X systems. Conclusions This paper proposes an RIS-assisted ISAC-V2X-NOMA system framework. By utilizing RIS to dynamically adjust the propagation path of ISAC signals and designing an EKF-based vehicle tracking and prediction method, efficient real-time vehicle sensing and communication are achieved. Furthermore, a multi-vehicle power allocation optimization scheme based on NOMA is proposed to enhance communication rate and resource utilization. The results suggest that the proposed system not only reduces pilot signal overhead but also enhances the overall system performance.
2025,
47(4):
1052-1065.
doi: 10.11999/JEIT240853
Abstract:
Objective With the rapid increase in wireless devices and the growing demand for sensing services, Integrated Sensing And Communication (ISAC) has become a key technology to address spectrum scarcity. ISAC systems enable joint communication and sensing by sharing spectrum and hardware resources, thereby improving both spectral and energy efficiency. They also exploit the complementary properties of sensing and communication to enhance system performance. However, due to spectrum sharing and the broadcast nature of wireless signals, ISAC systems face major security risks. Physical Layer Security (PLS) has emerged as an effective approach for enhancing ISAC security. PLS designs transmission strategies based on the randomness and diversity of wireless channels to reduce eavesdropping risks and enhance security. Intelligent Reflecting Surfaces (IRS), a core technology for next-generation wireless networks, can manipulate the propagation environment of wireless signals by adjusting reflection phases. IRS enables more stable communication and sensing links, extends coverage, increases accuracy, and strengthens the overall security of ISAC systems. It thus offers a promising solution to PLS challenges in ISAC. However, when eavesdroppers are highly mobile, rapid changes in location and Channel State Information (CSI) hinder the acquisition of accurate channel data and real-time secure transmission. Leveraging ISAC’s sensing capabilities to track mobile eavesdroppers is therefore critical for ensuring security. This paper proposes an IRS-assisted ISAC system that enhances secure transmission by integrating PLS strategies in scenarios where rapidly moving aerial sensing targets act as potential eavesdroppers. Methods This study establishes an IRS-assisted ISAC system model comprising an ISAC base station, a rapidly moving aerial sensing target, a legitimate user, and an IRS equipped with multiple reflective elements. The system utilizes the base station’s sensing capability to estimate the location and dynamic state of the sensing target via radar echoes. An Extended Kalman Filtering (EKF) is used to track and predict the target’s trajectory in real time. Based on the predicted trajectory, a joint optimization problem is formulated to maximize the system’s secrecy rate. The formulation accounts for the tracking performance constraints of EKF, the transmission power budgets of both the base station and the legitimate user, and the IRS phase shift constraints. The optimization variables include the base station’s beamforming vector, the IRS reflective beamforming configuration, and the transmission power of the legitimate user. To improve real-time performance and security, the problem is designed as a non-convex optimization. This is decomposed into three sub-problems using an alternating optimization framework. The sub-problems are then solved using Successive Convex Approximation (SCA), Dinkelbach’s algorithm, and Majorization–Minimization (MM) methods. Results and Discussions Simulation results confirm the effectiveness of the proposed method in target tracking, system security, and performance enhancement. The trajectory prediction error of the proposed approach is substantially lower than that of radar echo-based estimation methods. Additionally, the EKF-based tracking algorithm achieves accuracy comparable to Particle Filtering (PF), while reducing computational complexity and conserving system resources. The convergence of the proposed algorithm is also verified. Under three different settings for the number of IRS reflection elements, the algorithm converges within five iterations, indicating stable and efficient convergence behavior. The results further show that the system’s secrecy rate increases with the number of transmit antennas. This improvement arises from the additional spatial degrees of freedom provided by the antennas, which enable the base station to generate more focused beams toward the sensing target. These beams enhance interference directed at the target during detection, thereby improving secure transmission. The secrecy rate also increases significantly with the number of IRS reflection elements. A larger number of elements allows the IRS to exploit additional spatial freedom, achieving higher beamforming gains and improving secure communication performance. In scenarios involving mobile sensing targets, the proposed method yields greater secrecy rate improvements than radar echo-based approaches. This advantage is attributed to the EKF’s ability to estimate the target’s position more accurately and in real time, enabling timely adjustment of beamforming strategies and enhancing security. Moreover, the optimized IRS configuration outperforms random phase shift designs, particularly in large-scale IRS deployments. Optimizing the IRS phase shift matrix contributes to higher secrecy rates and improved communication performance. Conclusions This paper presents a secure transmission scheme for an IRS-assisted ISAC system, targeting scenarios in which a rapidly moving sensing target serves as a potential eavesdropper. The ISAC base station leverages its sensing capability to extract the target’s state parameters from radar echoes and applies EKF to track and predict the target’s trajectory in real time. Based on this tracking, an optimization model is constructed to maximize the system’s secrecy rate by jointly optimizing the uplink user’s transmission power, the base station’s transmit and receive beamforming vectors, and the IRS phase shift matrix. To solve this problem efficiently, an alternating iterative optimization framework is adopted, which decomposes the non-convex objective into three independent sub-problems. These sub-problems are addressed using SCA, Dinkelbach transformation, and MM methods. Simulation results demonstrate that the proposed approach effectively detects the sensing target, maintains robust tracking performance, and ensures secure communication. Moreover, compared with the scenario without IRS, the IRS-assisted design achieves a substantially higher secrecy rate, highlighting both the advantages of IRS deployment in ISAC systems and the effectiveness of the proposed algorithm.
2025,
47(4):
1066-1075.
doi: 10.11999/JEIT240885
Abstract:
Objective The reciprocity, time variability, and unpredictability of wireless channels enable physical-layer key generation, a promising technology for B5G/6G systems due to its independence from third-party involvement and inherent quantum-resistant properties. In millimeter-wave Multiple Input Multiple Output (MIMO) systems, channel sparsity imposes stringent constraints on key capacity, particularly in quasi-static propagation environments. While Reconfigurable Intelligent Surface (RIS) technology enhances channel time variability and increases key capacity, it also leads to an exponential increase in pilot overhead with the number of transceiver antennas and RIS elements. To mitigate pilot overhead, Compressive Sensing (CS) techniques have been employed by leveraging channel sparsity and reformulating channel estimation as a sparse signal recovery problem. However, existing CS-based key generation schemes require prior knowledge of channel sparsity, which may not reflect actual dynamic channel conditions. Additionally, these approaches typically rely on grid-based discrete modeling, where Angles of Departure (AoDs) and Angles of Arrival (AoAs) are quantized into predefined grids, leading to key mismatches. To address these challenges, this study proposes an RIS-assisted key generation scheme based on Atomic Norm Minimization (ANM) for RIS-assisted millimeter-wave MIMO systems. Methods The proposed method presents a novel key extraction approach based on virtual AoDs and virtual AoAs for RIS-assisted millimeter-wave MIMO systems. First, the problem of virtual channel parameter estimation in RIS-assisted millimeter-wave MIMO cascaded channels is formulated as a continuous sparse signal recovery problem. An optimization problem is then constructed using ANM, where ANM serves as the objective function and pilot observation error as the constraint. The Multiple Signal Classification (MUSIC) algorithm is integrated to enhance channel sparsity and achieve super-resolution angle estimation, thereby extracting high-precision virtual AoDs and AoAs as key parameters. Based on these parameters, a comprehensive key generation scheme is proposed, incorporating quantization, information reconciliation, and privacy amplification. Additionally, the key capacity of the proposed scheme is theoretically derived, with a closed-form expression provided based on the distribution of virtual AoDs/AoAs. Finally, Monte Carlo simulations are conducted to validate the effectiveness of the proposed scheme. Comparative analysis with existing schemes demonstrates its advantages in terms of key inconsistency, mutual information per bit, key generation rate, and pilot overhead. Results and Discussions Analysis of the simulation results indicates that the proposed scheme improves pilot overhead, Bit Disagreement Rate (BDR), mutual information per bit, and Secret Key Rate (SKR). These metrics primarily assess channel information extraction and key generation performance. For channel estimation accuracy, the Normalized Mean Square Error (NMSE) of the estimated virtual angles is used as an evaluation metric, where a lower NMSE indicates higher accuracy. Compared to other schemes, the proposed approach consistently achieves a lower NMSE, particularly for short pilot lengths. Even with \begin{document}$ {N_{\text{p}}} = 4 $\end{document} Fig. 3 ), demonstrating superior handling of sparse signals. This contributes to reduced pilot overhead and improved estimation accuracy. Key generation performance is evaluated using BDR, mutual information per bit, and SKR. Compared to schemes using the channel response matrix, employing virtual AoAs and AoDs as random keys results in a lower BDR (Fig. 4 ) and higher bit-wise mutual information (Fig. 6 ) across various Signal-to-Noise Ratio (SNR) conditions, demonstrating robustness in both high and low SNR environments. This advantage arises from the inherent sparsity of millimeter-wave channels, where primary propagation paths are clearly distinguishable. Unlike the channel response matrix, angle information is less susceptible to environmental factors and minor physical variations, providing a more stable key source. Compared with traditional CS-based schemes, the proposed approach overcomes grid constraints, reducing the key inconsistency rate by 47.7% under low SNR conditions (5 dB). Additionally, when the number of propagation paths remains constant, BDR decreases as the number of antennas increases. Conversely, when the number of antennas is fixed, BDR increases as the number of paths (L) grows (Fig. 5 ). This occurs because a higher number of paths increases the complexity of distinguishing AoAs and AoDs, leading to greater estimation error. Furthermore, a larger number of paths generates more key bits, causing BDR accumulation across paths, which raises the overall BDR. However, as the number of antennas increases, the sparsity of millimeter-wave MIMO channels becomes more pronounced (L is smaller), further amplifying the advantages of the proposed scheme. Additionally, by utilizing virtual angles as the key source, the proposed scheme maintains a high SKR even under low SNR conditions, further demonstrating its potential for practical applications (Fig. 7 ). Conclusions The proposed method employs ANM to formulate the cascaded channel estimation problem between the Base Station (BS) and User Equipment (UE) as a gridless sparse signal recovery problem. By integrating the MUSIC algorithm, the method jointly estimates virtual AoDs and AoAs, overcoming traditional grid-based constraints and eliminating the explicit assumption of channel sparsity. Therefore, high-precision channel parameters are extracted as key sources. Simulation results demonstrate that, compared to conventional CS-based methods, the proposed scheme reduces the BDR by 47.7% at an SNR of 5 dB while significantly lowering pilot overhead. Additionally, its performance advantage becomes more pronounced as the antenna array size increases. The proposed scheme offers a robust solution for key generation in RIS-assisted millimeter-wave MIMO systems, eliminating the need for prior sparsity knowledge and mitigating grid quantization errors.
2025,
47(4):
1076-1084.
doi: 10.11999/JEIT241111
Abstract:
Objective The widespread adoption of Unmanned Aerial Vehicles (UAVs) across civilian and military domains has introduced significant privacy and security challenges. Robust UAV identification and localization technologies are essential to address these concerns. While Radio Frequency Fingerprint Identification (RFFI) techniques based on deep learning show promise, their practical deployment is hindered by excessive model complexity, prolonged training periods, and limited generalization capabilities. This research presents a novel UAV identification and localization methodology utilizing Mel-Frequency Cepstral Coefficients (MFCC) and Gated Recurrent Unit (GRU) architecture that achieves superior accuracy with enhanced computational efficiency. Methods The proposed framework comprises several key components: (1) UAV video transmission signal acquisition via USRP N210 software-defined radio platform; (2) MFCC feature extraction to characterize distinctive radio frequency fingerprints; (3) GRU-based classification for UAV identification; and (4) Regularized Orthogonal Matching Pursuit (ROMP) algorithm implementation for three-dimensional localization parameter estimation. Comprehensive experimental evaluation assessed classification accuracy, computational complexity, training efficiency, and localization precision. Results and Discussions Experimental validation demonstrates that the proposed methodology achieves 98% UAV identification accuracy. The implemented GRU architecture contains only 1.6 k parameters and requires merely 9 seconds for training completion, representing significant reductions in model complexity and computational overhead (Table 2 ). For localization tasks, the system achieves three-dimensional positioning error below 1 meter. Robustness assessment through classification tests on 10 identical wireless modules from the same manufacturer at varying distances (1 m, 2 m, 3 m, and 5 m) yielded identification accuracies of 100%, 98%, 98%, and 99%, respectively (Table 3 ). These results confirm the method’s exceptional performance in both identification and localization applications. Conclusions This research introduces an efficient and accurate UAV identification and localization methodology based on MFCC features and GRU architecture. The approach substantially reduces model complexity and training requirements while maintaining high identification accuracy and precise localization capabilities. Experimental validation confirms its feasibility and robustness for practical deployment. Future research directions include algorithm optimization for real-time processing and extension to diverse UAV platforms and operational environments.
2025,
47(4):
1085-1092.
doi: 10.11999/JEIT241152
Abstract:
Objective The low-altitude economy, driven by the widespread adoption of drones and other unmanned aerial vehicles (UAVs), supports a range of applications across industries such as aerial imaging, precision agriculture, disaster response, and logistics. Precise Three-Dimensional (3D) positioning is essential to ensure the safety and efficiency of UAV operations in these scenarios. However, conventional cellular-based positioning approaches rely heavily on dedicated pilot signals, which impose significant overhead and limit 3D positioning accuracy. Integrated sensing and communication (ISAC) technology offers a promising alternative by enabling receivers to extract positioning information from reflected communication signals, thereby reducing dependence on pilot signals. Furthermore, a distributed network of ISAC transceivers can enhance sensing coverage and data diversity, improving localization performance. Building on these principles, this study proposes a 3D positioning algorithm based on distributed ISAC networks. The algorithm achieves high positioning accuracy without additional pilot signal overhead, demonstrating strong potential to support UAV applications within the low-altitude economy. Methods Motivated by the principles of distributed radar systems, this study proposes a cooperative 3D positioning method for UAVs within a distributed ISAC network comprising both ground base stations (BSs) and UAV-mounted BSs. Firstly, each ISAC receiver—whether ground-based or UAV-mounted—independently collects communication signals reflected by the target UAV from multiple ISAC transmitters. Secondly, each ISAC receiver serves as an edge computing node and derives a coarse estimate of the UAV’s 3D coordinates. Specifically, the receiver applies the MUltiple SIgnal Classification (MUSIC) algorithm to estimate the time delays of Orthogonal Frequency-Division Multiplexing (OFDM) signals refracted by the UAV. This is accomplished by exploiting the common delay steering vector structure across different ISAC transmitters. The resulting time-delay estimates are input into an ellipse-based positioning algorithm to obtain the initial 3D position of the UAV. By processing signals from at least three BSs, the UAV’s position can be triangulated via the intersection of ellipses. The coarse 3D position estimates are then transmitted to a central computing unit, where a weighted averaging method refines them to achieve higher accuracy. This hierarchical approach ensures robust localization performance, even when individual edge estimates are degraded due to noise, interference, or geometric limitations that prevent reliable ellipse-based estimation. To evaluate the proposed algorithm, the Cramér–Rao Lower Bound (CRLB) of the 3D positioning error is derived under the assumption of circularly symmetric complex Gaussian noise in the communication channel model. Results and Discussions The combined use of ground and aerial BSs yields lower position estimation errors compared to networks consisting solely of ground BSs (Fig. 2 ). This improvement arises from the additional orientation information provided by UAV-mounted BSs, which enhances the accuracy of height estimation and thereby improves overall 3D localization performance. In low Signal-to-Noise Ratio (SNR) conditions, positioning accuracy declines significantly due to the failure to distinguish the noise subspace from the signal subspace. This impairs the formation of spatial spectrum peaks, degrading the performance of the MUSIC algorithm (Fig. 3 ). Under such conditions, errors in time delay estimation become the primary source of positioning inaccuracy. At high SNR, despite reduced noise influence, the hyperbolic positioning algorithm may still converge to a local optimum due to suboptimal initial position selection, resulting in persistent estimation errors. In comparison, the proposed algorithm maintains superior performance across both low and high SNR regimes (Fig. 3 ). As the path gain increases, the Position Error Bound(PEB) of the estimation error decreases, indicating improved theoretical positioning accuracy(Fig. 4 ). Moreover, increasing the number of receiving stations significantly enhances localization performance. When SNR variation among BSs exceeds differences in geometric distribution, SNR-based weighted averaging yields better positioning results than Geometric Dilution of Precision (GDOP)-based averaging (Fig. 5 ). Conclusions Monte Carlo simulations confirm that the proposed algorithm achieves high-accuracy 3D positioning without requiring dedicated pilot signals. The results further indicate that UAV-mounted BSs, when functioning as ISAC transceivers and edge computing centers, can effectively support ground BSs in estimating the 3D position of target UAVs. Compared with positioning algorithms that use only ground BSs, this configuration notably improves altitude estimation accuracy.
2025,
47(4):
1093-1103.
doi: 10.11999/JEIT240970
Abstract:
Objective Unmanned Aerial Vehicle (UAV)-assisted millimeter-Wave (mmWave) communication enables high-speed data transmission in diverse on-demand service and emergency scenarios. However, mmWave signals are inherently sensitive to blockage, leading to significant path loss that adversely affects system throughput. Existing anti-blockage strategies primarily rely on the probabilistic blockage model for UAV deployment, which is often limited in accurately reflecting real-time blockage status. To address this issue, a UAV anti-blockage trajectory planning method based on blockage prediction is proposed for scenarios where the geographic information of building obstacles (e.g., location, shape, and size) is unknown and mobile user positioning contains errors. This method enables accurate prediction of the blockage status for UAVs and users at any location, including unvisited areas. A three-Dimensional (3D) UAV trajectory is then designed through an iterative process that alternates between trajectory optimization and blockage prediction to mitigate blockage effects, thereby improving user throughput. Methods Utilizing the locations of the UAV and users, the link blockage is predicted by designing a geometric feature vector that incorporates Taylor expansion terms to account for position errors. Based on this prediction, the UAV’s 3D trajectory is optimized iteratively to avoid blockages and enhance user throughput. A Double Deep Q-Network (DDQN)-based deep reinforcement learning algorithm is employed to address this challenge. During decision-making, the UAV selects an action based on the current Q-value estimate and blockage prediction while maintaining an exploratory capability through a greedy strategy. As the UAV provides communication services, it continuously collects new blockage status data and periodically updates the blockage prediction model, improving prediction accuracy. The iterative interaction between action selection and prediction accuracy refinement enhances overall system performance. Results and Discussions The proposed anti-blockage UAV 3D trajectory design algorithm alternates between blockage prediction and trajectory optimization. Simulation results indicate that the designed feature vector for user positioning errors improves blockage prediction accuracy (Fig. 5 ). This improvement arises because incorporating Taylor expansion terms yields a feature vector that better approximates the actual user position in the presence of errors compared to one without these terms. Examples of UAV 3D trajectories in mmWave communication networks are demonstrated in two different urban environments (Fig. 6 ). The UAV selects actions based on user locations, resulting in an irregular flight path that aligns with user distribution. The relationship between user throughput and UAV transmit power under different blockage status methods when applying the proposed trajectory planning algorithm is illustrated (Fig. 7 ). In all cases, user throughput increases with higher UAV transmit power, as greater power enhances communication performance. Notably, employing the proposed blockage prediction model achieves higher throughput and closely approximates the performance of the method using real blockage status. This is because the prediction model reduces blockage uncertainty by accurately predicting blockages, thereby better matching the actual environment. The algorithm complexity comparison is presented in Table 3 . The RBS+DDQN (Real Blockage Status + DDQN-Based Path Planning) benchmark algorithm requires complete prior knowledge of the 3D geographic information of buildings, which may pose challenges in real-world applications due to complex data processing and potential latency issues. Compared with existing algorithms that use the probabilistic blockage model, the proposed algorithm, although relatively more complex, does not require building geographic information and achieves higher throughput despite errors in mobile user positioning. Its performance is close to the ideal algorithm with real blockage status, where full geographic information is available (Fig. 8 and Fig. 9 ). Therefore, the proposed algorithm achieves a balance between complexity and performance. Conclusions This study proposes an anti-blockage UAV 3D trajectory design algorithm for scenarios where prior building information is unavailable and mobile users have positioning errors. By incorporating a Taylor expansion error term into the feature vector, blockage prediction accuracy is enhanced using UAV and user location data. The blockage prediction model accurately determines blockage status at any position, including unvisited areas. Subsequently, the DDQN algorithm optimizes the UAV trajectory to avoid building blockages, thereby maximizing user throughput. Simulation results demonstrate that introducing the Taylor expansion feature vector improves blockage prediction accuracy in the presence of mobile user positioning errors. Furthermore, although the proposed algorithm has relatively high complexity, it achieves higher user throughput, effectively balancing complexity and performance.
2025,
47(4):
1104-1112.
doi: 10.11999/JEIT240827
Abstract:
Objective With the ongoing advancement of the Internet of Things (IoT) and the increasing integration of communication devices, there is a growing need for miniaturized, low-profile, and polarized antennas. To meet these needs, Antenna-in-Package (AiP) technology, which integrates the functional modules of the Radio Frequency (RF) system and enables multi-functional design, has emerged as a key development for wireless system miniaturization. Currently, two main antenna packaging technologies are used: Monolithic Microwave Integrated Circuit (MMIC) and Multi-Chip Module (MCM). MMIC faces limitations due to material and process constraints, making it difficult to integrate a large number of passive components. In contrast, MCM facilitates the integration of multiple IC chips onto a single substrate. It employs Surface-Mount Technology (SMT) for antenna design, using advanced microelectronic assembly and interconnection techniques to combine these components into a complete circuit system. Low-Temperature Co-Fired Ceramic (LTCC) technology is crucial in MCM, offering high-density 3D interconnect capabilities, low loss, high-temperature resistance, and other benefits, which make it widely used in communication applications. Integrated antennas based on LTCC technology offer high integration, compact size, light weight, and broad applicability, making them a focus of global research. While microstrip patch antennas are suitable for low-profile, circular polarization, a key challenge is using LTCC technology to widen the bandwidth of these antennas and integrate them with transceiver modules. This paper explores the development of a compact, wideband planar circularly polarized antenna using LTCC technology, integrated with a transceiver chip to form a miniaturized transceiver module. This innovation extends the use of AiP technology in IoT systems and has significant engineering implications. Methods To meet the increasing demand for low-profile integrated antennas in wireless transceiver systems for the IoT, this paper investigates and presents a circularly polarized integrated antenna based on LTCC technology. The antenna uses a 3D laminated LTCC structure to integrate the 3 dB coupler feed, printed radiating patch, Bluetooth chip, and associated peripheral control circuits. A detailed analysis of the LTCC laminate structure leads to several design enhancements, including structural hollowing, an integrated feed structure, and clearance processing. These improvements effectively expand the antenna bandwidth and significantly increase its gain, while preserving the low-profile circular polarization characteristics. The antenna is then integrated with an RF chip for packaging. Experimental results confirm that the RF transceiver module is compact, supports a long communication range, and meets the specific demands of IoT applications, demonstrating significant engineering potential, reliability, and high practical value. Results and Discussions Based on transmission line theory, this paper proposes a hollow structure in the LTCC laminate process to reduce the effective dielectric constant, significantly expanding the antenna bandwidth. The integration of the 3 dB coupler feed structure, printed radiating patch, Bluetooth chip, and peripheral control circuits allows for seamless integration with the transceiver module. Simulation results show that the antenna’s impedance bandwidth spans from 2.15 to 2.59 GHz, with a return loss of less than –10 dB and an axial ratio below 3 dB. The antenna is fabricated using LTCC technology, with dimensions of 0.37λ0×0.37λ0×0.33λ0 (λ0 is the free space wavelength at the central frequency). Measured results closely match the simulation data (Figure 8 ), confirming that the design effectively broadens the bandwidth while maintaining a compact size, thus validating the proposed hollow structure. Finally, the antenna is integrated with the RF circuit substrate to form a complete transceiver system (Figure 11 ). Test results demonstrate that the circularly polarized antenna offers excellent engineering application potential and high practical value (Tables 3 and 4 ). Conclusions This paper presents the design of a circularly polarized integrated antenna based on LTCC technology. The device integrates a 3 dB coupler feed structure, printed radiating patch, Bluetooth chip, and peripheral control circuits, with the hollow structure effectively broadening the antenna bandwidth. The fabricated antenna meets the performance requirements for Bluetooth systems, with an axial ratio and return loss that align with system specifications. The antenna’s size is 0.37λ0×0.37λ0×0.33λ0, demonstrating its low-profile characteristics. The hollow structure at the base integrates well with the transceiver chip, enhancing its engineering application potential. Finally, the seamless integration of the antenna with the wireless transceiver chip forms a complete, highly functional module. Measured results confirm its excellent circular polarization and practical characteristics. The proposed design approach provides valuable insights for the future development of integrated AiP solutions.
2025,
47(4):
1113-1125.
doi: 10.11999/JEIT240390
Abstract:
Objective The growing reliance on data in today’s digital age highlights the importance of effective data management across industries. Federated Learning (FL), an innovative approach, facilitates data collaboration and joint model development while maintaining privacy. However, existing homomorphic encryption-based security schemes for FL present several limitations. In some cases, FL servers may falsify aggregation results, leading to inaccurate models and subsequent issues, such as decision-making errors and erosion of trust in the system. Furthermore, servers may collude with users to steal private data, resulting in privacy breaches and potential misuse, including illegal marketing or cyberattacks. These issues undermine public confidence in data security and limit the broader adoption of FL, thus impeding innovation and efficiency gains. Many current schemes also depend heavily on trusted Third PArties (TPA) for key generation, introducing high communication overhead and diminishing model training efficiency, which discourages users from sharing data. This study proposes an optimized solution utilizing a distributed key generation protocol to prevent collusion and reduce third-party dependency. It integrates the Chinese Remainder Theorem (CRT) to lower communication costs and introduces auxiliary nodes to ensure aggregation accuracy. Additionally, an incentive mechanism is designed to encourage users to share high-quality private data. Collectively, these measures address key challenges in existing systems, offering a safer, more efficient, and reliable framework for the widespread adoption of FL. Methods The proposed FL scheme is collusion-resistant, privacy-preserving, and verifiable, integrating a distributed key generation protocol to achieve interactive key generation. This method enables users to encrypt data using their private keys while requiring collaborative decryption from multiple participants, thereby eliminating the reliance on TPA. It effectively prevents server collusion involving fewer than n–1 users and incorporates a fault-tolerant mechanism to address potential user disconnections. Enhanced data security and reduced communication overhead are achieved by employing randomized model processing, combined with CRT based dimensionality reduction prior to encryption. Specifically, each user superimposes a random model of identical dimensions onto their local model, uploads the randomized model to the server for aggregation, and then decomposes the random model into a public matrix and a low-dimensional vector. After applying CRT to reduce the vector’s dimensionality, homomorphic encryption is performed, reducing the data that must be encrypted and uploaded. The scheme also introduces auxiliary nodes and utilizes a bilinear aggregate signature algorithm, enabling each user to independently verify the aggregation results provided by the server, ensuring correctness and verifiability. Additionally, an incentive mechanism based on data characteristics, such as quality and richness, encourages participation from users with high-quality data. By dynamically calculating and distributing rewards after task completion, the mechanism effectively promotes the active sharing of high-quality data by users. Results and Discussions The proposed scheme is comprehensively evaluated through extensive experiments. The results demonstrate improvements in both model accuracy and training efficiency. As shown in (Fig. 4 ), the scheme achieves slightly higher accuracy on the MNIST dataset compared to FedAvg and three other approaches. This improvement is attributed to the incentive mechanism, which effectively encourages participation from users with high-quality data. Additionally, the preprocessing steps involving model randomization and CRT-based dimensionality reduction prior to encryption enhance communication efficiency, as evidenced by the time overhead comparison in (Fig. 5 ). The experimental evaluation of the designed verification scheme, shown in (Fig. 6 ), reveals that the verification time of user will not increase with the increase of the number of users. Even with 30 users, the verification time increases by only 1%, despite a 30% dropout rate, when compared to scenarios with no user dropouts. (Fig. 7 ) further confirms the verification time advantages of the proposed scheme over other verification approaches. Finally, the reward allocation mechanism demonstrates desirable fairness characteristics, as shown in (Fig. 8 ), where users contributing high-quality data consistently receive proportionally greater rewards throughout the training process. Conclusions The proposed privacy-preserving FL scheme, based on homomorphic encryption and verifiable mechanisms, effectively reduces the computational overhead associated with homomorphic encryption through optimized model parameter processing. This ensures data privacy while preventing excessive communication costs. Additionally, the framework incorporates a distributed key generation protocol to eliminate reliance on trusted third-party institutions and integrates the Diffie-Hellman key exchange protocol with Shamir’s secret sharing algorithm. This combination enables users to independently verify aggregation results provided by the server while supporting user dropouts and preventing collusion. To further encourage data contributions from users with high-quality data, an incentive mechanism is introduced, employing rational reward strategies to attract such users. Experimental results demonstrate excellent performance in model convergence speed and prediction accuracy, with verification time for aggregation results remaining stable regardless of the number of users. However, this study does not address potential malicious behaviors, such as individual users uploading erroneous or deceptive model updates that could compromise global model accuracy and fairness. Future work will focus on developing mechanisms to identify and mitigate attacks from malicious users while maintaining data privacy protection.
2025,
47(4):
1126-1139.
doi: 10.11999/JEIT240798
Abstract:
Objective With the advancement of information technology, information security concerns have become increasingly significant, making covert communication technology a critical area of focus. Existing schemes face limitations regarding embedding rate, anti-detection, and communication efficiency. To address these issues, steganographic embedding methods based on Generative Adversarial Networks (GANs) have gained considerable attention. This study utilizes the iterative training of GAN and steganalysis adversarial networks to generate stego-images with enhanced anti-detection capabilities. This approach aims to meet the concealment requirements for secure information transmission, while also improving the communication efficiency and security of the information exchange. Methods This study proposes a blockchain-based covert communication scheme utilizing image multilevel steganography. First, a multiple adversarial network for steganography is constructed, generating stego-images with enhanced anti-detection capabilities through the adversarial iterative training of GAN and steganalysis adversarial networks. Next, a reversible data hiding method in the ciphertext domain, based on location map information, is employed to embed the hidden data into the stego-images, resulting in a stego-images that contains the complete hidden information. Finally, the ciphertext image is stored in the InterPlanetary File System (IPFS) to assign it a unique identity, and then mapped to an address in the blockchain to enable covert transmission. Results and Discussions To evaluate the effectiveness of the proposed scheme in terms of anti-steganography capability, invisibility, embedding capacity, and communication delay, simulation experiments are conducted. Regarding anti-steganography capability, the stego-images generated by the proposed scheme demonstrate strong anti-detection performance, outperforming the WOW and HILL algorithms (Fig. 7 ). In terms of concealment, the reversible data hiding method in the ciphertext domain, based on location map and spatial domain information, offers high concealment, effectively protecting the image content while enabling lossless restoration (Table 5 , Table 6 , Table 7 ). Concerning embedding capacity, the steganography algorithm in this scheme exhibits a high embedding capacity, with an average embedding rate exceeding that of the PBTL, IPBTL, and ERLC-BMPR algorithms (Fig. 9 ). Finally, in terms of communication delay, the proposed scheme results in low covert communication delay, outperforming the DVANET, BDLV, and L-TCM algorithms (Fig. 10 ). Conclusions This paper proposes a blockchain-based covert communication scheme utilizing image multilevel steganography. Simulation experiments validate its advantages in information embedding rate, anti-steganography detection capability, concealment, and communication delay. The results demonstrate the following: 1. In terms of anti-steganography ability, the anti-detection performance of stego-images generated by SRNet+Zhu-Net significantly exceeds that of the WOW and HILL methods; 2. Regarding invisibility and embedding capacity, the proposed reversible data hiding method in the encrypted domain, based on location map and spatial domain information, achieves a high embedding rate and lossless recovery, outperforming the PBTL, IPBTL, and ERLC-BMPR methods; 3. In terms of communication efficiency, this scheme significantly reduces communication delay by combining blockchain and IPFS. Future research will focus on homomorphic encryption and identity authentication mechanisms to further enhance the security of on-chain data.
2025,
47(4):
1140-1149.
doi: 10.11999/JEIT240441
Abstract:
Objective Traditional classifiers tend to optimize overall accuracy when dealing with imbalanced data sets, often resulting in poor classification performance for minority class samples. Among the available strategies, oversampling methods are widely used due to their strong generalization ability. However, conventional oversampling techniques frequently generate new samples with high overlap rates and limited validity, particularly near decision boundaries. To address this issue, this study proposes an adaptive oversampling approach that selects sub-boundary samples—those located near the boundary samples—for sample generation. In addition, the nearest-neighbor parameter space is constrained to refine the synthetic sample region. This method improves the classifier’s performance when learning from imbalanced data sets. Methods This study first identifies the maximum safe like-neighbors of positive class samples and classifies these samples as either hazardous or safe. The local density of each sample is then calculated, and hazardous samples—those more difficult to classify—are further categorized as either boundary samples or outliers. To provide the classifier with more informative positive class samples, “sub-boundary points” are preferentially selected as root samples using a weighted composite factor. The K-value in the K-nearest neighbor algorithm is adaptively adjusted based on the maximum safe nearest neighbor of each sample to improve neighbor selection. Outliers are oversampled randomly within a hypersphere to generate new samples while minimizing increases in spatial complexity. Results and Discussions To evaluate the feasibility and generalization of the proposed method, Logistic Regression (LR) and Support Vector Machine (SVM) classifiers are employed as base classifiers. The range of the distance adjustment coefficient is first determined by comparing results across selected datasets (Table 3 ). Once the range is established, the effect of different weight adjustment coefficients on performance is assessed (Table 4 ). The proposed method is then compared with six existing oversampling techniques across 13 datasets. For most datasets, the proposed method achieves higher values in more than half of the five evaluation metrics considered (Tables 5 and 6 ). These results demonstrate that the proposed approach effectively improves classifier performance on imbalanced data sets. Conclusions This study introduces the maximum safe nearest neighbor number and local density to classify minority class samples into safe samples, boundary samples, and outliers. A weighted sampling probability, based on both local density and the maximum safe nearest neighbor number, is used to guide adaptive K-nearest neighbor oversampling of safe and boundary samples. Random oversampling within a hypersphere is applied to outliers to preserve informative but rare samples. Comparative experiments confirm that the proposed method performs well across datasets with varying imbalance ratios and remains competitive under highly imbalanced conditions.
2025,
47(4):
1150-1160.
doi: 10.11999/JEIT240886
Abstract:
Objective The fusion of InfRared (IR) and VISible light (VIS) images is critical for enhancing visual perception in applications such as surveillance, autonomous navigation, and security monitoring. IR images excel in highlighting thermal targets under adverse conditions (e.g., low illumination, occlusions), while VIS images provide rich texture details under normal lighting. However, existing fusion methods predominantly focus on optimizing performance under uniform illumination, neglecting challenges posed by dynamic lighting variations, particularly in low-light scenarios. Additionally, computational inefficiency and high model complexity hinder the practical deployment of state-of-the-art fusion algorithms. To address these limitations, this study proposes a scene-adaptive knowledge distillation framework that harmonizes fusion quality across daytime and nighttime conditions while achieving lightweight deployment through structural re-parameterization. The necessity of this work lies in bridging the performance gap between illumination-specific fusion tasks and enabling resource-efficient models for real-world applications. Methods The proposed framework comprises three components: a teacher network for pseudo-label generation, a student network for lightweight inference, and a light perception network for dynamic scene adaptation (Fig. 1 ). The teacher network integrates a pre-trained progressive semantic injection fusion network (PSFusion) to generate high-quality daytime fusion results and employs Zero-reference Deep Curve Estimation (Zero-DCE) to enhance nighttime outputs under low-light conditions. The light perception network, a compact convolutional classifier, dynamically adjusts the student network’s learning objectives by outputting probabilistic weights (Pd, Pn) based on VIS input categories (Fig. 3 ). The student network, constructed with structurally Re-parameterized Vision Transformer (RepViT) blocks, utilizes multi-branch architectures during training that collapse into single-path networks during inference, significantly reducing computational overhead (Fig. 2 ). A hybrid loss function combines Structural SIMilarity (SSIM) and adaptive illumination losses (Eq. 8–15), balancing fidelity to source images with scene-specific intensity and gradient preservation. Results and Discussions Qualitative analysis on the MSRS and LLVIP datasets demonstrates that the proposed method preserves IR saliency (highlighted in red boxes) and VIS textures (green boxes) more effectively than seven benchmark methods, including DenseFuse and PSFusion, particularly in low-light scenarios (Fig. 4 , Fig. 5 ). Quantitative evaluation reveals superior performance in six metrics: the method achieves SD scores of 9.728 7 (MSRS) and 10.006 7 (LLVIP), AG values of 6.5477 (MSRS) and 4.7956 (LLVIP), and SF scores of 0.0670 (MSRS) and 0.0648 (LLVIP), outperforming existing approaches in contrast, edge sharpness, and spatial detail preservation (Table 1 ). Computational efficiency is markedly improved, with the student network requiring only 0.76 MB parameters and 4.49 ms runtime on LLVIP, representing a 98.8% reduction in runtime compared to PSFusion (380.83 ms) (Table 2 ). Ablation studies confirm the necessity of RepViT blocks and adaptive illumination loss, as removing these components degrades SD by 16.2% and AG by 60.8%, with other evaluation metrics also experiencing varying degrees of decline,respectively (Table 3 , Fig. 6 ). Conclusions This work introduces a scene-adaptive knowledge distillation framework that unifies high-performance IR-VIS fusion with computational efficiency. Key innovations include teacher knowledge distillation for illumination-specific pseudo-label generation, RepViT-based structural re-parameterization for lightweight inference, and probabilistic weighting for dynamic illumination adaptation. Experimental results validate the framework’s superiority in perceptual quality and operational efficiency across benchmark datasets. Future work will extend the architecture to multispectral fusion and real-time video applications.
2025,
47(4):
1161-1171.
doi: 10.11999/JEIT240440
Abstract:
Objective Virtual Reality Motion Sickness (VRMS) hinders the development of virtual reality technology and affects users’ experience, potentially threatening their health. Accurately assessing VRMS levels is essential for studying its causes and treatment strategies. ElectroEncephaloGram (EEG) provides a non-invasive, low-cost method with high temporal resolution, reflecting real-time neural activity, making it suitable for VRMS assessment. This paper introduces and improves an end-to-end EEG regression model based on Convolutional Neural Networks (CNN) and functional brain networks, termed Brain Connection-based CNN (BCCNN), to quantitatively recognize VRMS in users within a VR environment. Methods The BCCNN utilizes a 1D-CNN to filter EEG signals and compute correlation coefficients among electrodes, forming functional brain networks. It then employs CNN and fully connected layers to extract network features and perform regression analysis (Figure 2 ). This study optimizes the kernel size of the 1D-CNN and proposes a novel lead attention module to enhance feature extraction. The attention module, inspired by the squeeze-and-excitation mechanism, computes the weights from the filtered EEG signals rather than the extracted features. Additionally, the attention weights are derived from and applied to the leads of the EEG (Figure 3 ). To induce VRMS, a virtual reality scene called “VRQ test” is used. The subject’s EEG signal and subjective VRMS level, recorded via the Simulator Sickness Questionnaire (SSQ), are collected. These data are then used to validate the model (Figure 2 ). The model’s performance is evaluated using Mean Squared Error (MSE), Mean Absolute Error (MAE), and the goodness of fit (R2), comparing the predicted VRMS levels with the real values. A comparison with reference methods is conducted to assess the effectiveness of the BCCNN and its optimizations. Results and Discussions The results show that the optimized kernel size of the 1D-CNN layer is 16, reducing the average MSE of the original BCCNN by 6.53 (Table 1 , Table 2 , Figure 4 ). Additionally, the lead attention module further improves the BCCNN, lowering the average MSE by 7.65 compared to the original model, outperforming the channel attention module (Table 2 ). The optimized BCCNN achieves an average MSE of 15.10 and an average R2 of 96.63% in 10-fold cross-validation, significantly exceeding the original BCCNN and ten state-of-the-art and baseline models (Table 2 ). Among the reference methods, the combination of difference entropy and Gaussian process regression yields the best performance. Furthermore, reference models using a filter bank outperform other reference models, indicating that handcrafted processing of the EEG data can enhance model performance (Table 2 ). Visualizing the functional connections and extracted features of the BCCNN reveals that functional connections are stronger at higher VRMS levels compared to lower VRMS levels. Conclusions This study introduces and optimizes the BCCNN for assessing VRMS using EEG. The main innovation of this work lies in optimizing the kernel size of the 1D-CNN and proposing a novel lead attention module. The results demonstrate that these optimizations enhance the accuracy of VRMS assessment, with the updated model offering a more precise evaluation. EEG is thus expected to become a standard method for assessing VRMS in VR products. The proposed approach enables VRMS assessment during and after a user’s experience in a VR scene.
2025,
47(4):
1172-1181.
doi: 10.11999/JEIT240855
Abstract:
Objective The rapid advancement of the Internet of Things and sensor technology has led to the accumulation of vast amounts of unlabeled time-series data, making deep time-series clustering a key analytical approach. However, existing deep clustering methods lack supervised constraint information and label guidance, making them susceptible to noise and outliers. Deep semi-supervised clustering methods rely on predefined Must-Link (ML) and Cannot-Link (CL) constraints, limiting improvements in clustering performance. Existing active clustering approaches sample only within clusters in the representation space, overlooking pairwise annotations from different clusters. This results in lower-quality ML and CL constraints and prevents further extrapolation from manually annotated pairs, increasing annotation costs. To address these limitations, this paper proposes Deep Active Time-series Clustering based on Constraint Transitivity (DATC-CT), which improves clustering performance while reducing annotation costs. Methods DATC-CT defines an Annotation Cluster Set (ACS) and an Auxiliary Annotation Set (AAS) and obtains the representation vector of time-series samples using a pre-trained autoencoder. In each clustering epoch, samples closest to cluster centers in the representation space are selected, labeled, and stored in ACS. This ensures that all samples within an ACS belong to the same category, while those in different ACSs belong to different categories. Next, a time-series sample is randomly chosen from the ACS with the fewest samples. Another sample, which does not belong to the same cluster but is nearest to the selected sample’s center, is then sampled, labeled, and stored in either AAS or ACS. Samples in ACS and AAS belong to different categories. ML and CL constraints are inferred from these samples. The encoder’s network parameters and cluster centers are updated using KL divergence between the cluster distribution (modeled by a t-distribution) and an auxiliary distribution generated from it. Additionally, a constraint loss function is applied to increase the distance between ML-constrained samples while reducing the distance between CL-constrained samples in the representation space. Results and Discussions Experimental results on 18 public datasets show that the proposed method improves the average Rand Index (RI) by more than 5% compared to existing deep time-series clustering methods (Table 2 ). With the same labeling budget, it achieves an RI improvement of over 7% compared to existing active clustering methods (Table 3 ). These findings confirm the effectiveness of the active sampling strategy and constraint reasoning mechanism. Additionally, the method infers a large number of ML and CL constraints from a small set of manually annotated constraints (Fig. 4 ), significantly reducing annotation costs. Conclusions This paper proposes a deep active time-series clustering model based on constraint transitivity, incorporating a two-phase active sampling strategy: exploration and consolidation. In the exploration phase, the model selects the sample closest to each cluster center in the representation space and stores it in ACS. During consolidation, a sample is randomly chosen from the ACS with the fewest samples. Another sample, which does not belong to the same cluster but is nearest to the selected sample’s center, is then sampled, labeled, and stored in either AAS or ACS. The number of ACS and AAS matches the number of clusters. ML and CL constraints are inferred from ACS and AAS samples. Experiments on public datasets demonstrate that inferring new clustering constraints reduces annotation costs and improves deep time-series clustering performance.
2025,
47(4):
1182-1191.
doi: 10.11999/JEIT240841
Abstract:
Objective Deep learning-based object detection typically requires a large volume of high-quality annotated samples, which limits its practical applicability. Few-Shot Object Detection (FSOD) has gained significant attention as a promising research area. FSOD leverages base classes with abundant labeled data to recognize novel classes with limited training samples. Several methods based on generative models have been proposed to address the challenge of limited annotated data in FSOD. However, some limitations remain. (1) Most generative models fail to sufficiently capture the relationships between base and novel classes, which hinders their ability to accurately estimate novel class distributions and degrades the quality of generated samples. (2) Existing methods often prioritize increasing sample diversity, neglecting the critical need for representativeness. Low representativeness can cause confusion between categories, potentially reducing detection performance. Since the quality of generated samples directly affects the performance of the object detection network, which trains using both original and generated samples, this issue must be addressed. To address these challenges, a novel framework for data generation in FSOD via additional high-Quality and Representative Samples (FQRS), is introduced. A conditional control module, incorporating both inter-class and intra-class dynamics, is introduced to improve the quality and representativeness of generated samples, ultimately enhancing the accuracy of FSOD. Methods The proposed model architecture consists of a fine-tuning-based object detector and a data generator. First, the object detector is trained using base class data. Then, the pre-trained detector is employed to extract Region of Interest (RoI) features, which are used as training data for the generator. The generator, once trained, generates new samples for the novel classes. The architecture of the data generator includes a diffusion model for sample generation and an inter-class and intra-class conditional control module to guide the diffusion process. For inter-class conditional control, a semantic relation embedding is introduced, using cosine similarity to represent the degree of correlation between different classes. This enables the data generator to learn inter-class relations effectively. The relations between base and novel classes assist the diffusion model in estimating novel class distributions, improving the quality of generated samples. For intra-class conditional control, Intersection Over Union (IOU) information is utilized to constrain the position of generated samples within the corresponding feature space. This ensures that generated samples cluster around their respective category centers, enhancing their representativeness and preserving important class characteristics. Finally, the object detector is fine-tuned using both the generated samples and the original training samples. A hyperparameter in the loss function is introduced to control the influence of generated samples on the object detector’s training process. Results and Discussions The effectiveness and robustness of the proposed network are validated on two public datasets: PASCAL VOC and MS COCO. Detection accuracy is evaluated using mAP and mAP50 metrics. Quantitative comparisons (Tables 1 and 2 ) show that the proposed network outperforms existing methods across both datasets. For example, on the MS COCO dataset under the 1-shot setting, the proposed method achieves a 16.9% improvement over the state-of-the-art DeFRCN approach. A cross-domain experiment (Table 3 ), where base and novel class data are sourced from different datasets, demonstrates the superior generalization capability of the proposed method. Visual comparisons (Fig. 5 ) highlight that the proposed method effectively addresses issues like missed detections and category confusion arising from limited training data, thus improving the performance of FSOD. Ablation studies (Tables 4 , 5 , and 6 ) confirm the efficacy of the proposed modules and reveal the impact of varying parameter configurations on detection performance. t-SNE visualization results (Fig. 6 ) show that the inter-class and intra-class conditional control module enhances feature aggregation within the same category, while improving discriminability between categories and reducing categorical confusion. Additionally, quantitative analysis (Table 7 ) examines the variations in model complexity introduced by the data generator, focusing on both parameter count and floating-point operations. Conclusions This paper presents a novel data-generation-based framework for obtaining additional samples in FSOD. The framework integrates a data generator, built on a conditional diffusion model, into a fine-tuning-based object detection network. The proposed data generator learns category features in conjunction with inter-class relations, capturing distinct category characteristics and improving generalization to novel classes. Additionally, the generator enhances sample representativeness by constraining generated samples to cluster around category centers. These high-quality, representative generated samples facilitate the object detector’s training, leading to improved FSOD accuracy. In various few-shot settings, the proposed model outperforms the state-of-the-art fine-tuning object detection model, Decoupled Faster Region-based Convolutional Neural net-work (DeFRCN), on both the PASCAL VOC and MS COCO datasets. Extensive experimental results validate the superiority of the proposed approach.
2025,
47(4):
1192-1201.
doi: 10.11999/JEIT240811
Abstract:
Objective With continuous advancements in Integrated Circuit (IC) process technology, chip integration levels have steadily increased, driving higher market demands for performance. In the era of intelligence and digitalization, an inherent challenge arises: as the number of logic gates increases, both main frequency and power consumption rise, imposing stricter requirements on digital IC designers. Although existing Electronic Design Automation (EDA) tools optimize timing using useful skew in clock trees, this technique has notable limitations. To address this issue, a novel adaptive full-flow clock tree timing violation correction method is proposed. This method corrects timing violations unresolved by conventional flows while reducing power consumption and improving performance, meeting the market’s dual demands for high-performance and low-power chips. Methods The ADaptive Full Flow (ADFF) clock tree optimization method is based on the RISC-V CPU architecture. As an open-source architecture, RISC-V offers openness and flexibility, making it widely used in high-performance, low-power processor design. The method exploits imbalances in key path logic depth to enhance optimization. Useful skew is introduced to adjust logic delay distribution, improving overall performance. Timing feedback is incorporated at multiple stages, ultimately forming a joint optimization strategy for power consumption and timing, which enhances clock tree quality and reduces chip load. The method integrates feedback optimization during both the Clock Tree Synthesis (CTS) and routing stages. In the CTS stage, timing paths are traversed to gather feedback, which is then returned to the pre-CTS stage for early intervention. Adaptive iteration accurately identifies critical paths and resolves setup time violations. In the routing stage, targeted strategies address hold time violations, and the merging method reduces power consumption while optimizing timing correction. This enables full-flow correction of clock tree timing violations while improving power efficiency. Results and Discussions The ADFF clock tree optimization strategy is implemented using Synopsys IC Compiler II for layout and routing, establishing an adaptive full-flow framework for correcting clock tree timing violations (Fig. 5 ). For setup time violations in the reg2reg group, a loop iteration algorithm dynamically adjusts path delays, updating CTS guidance files to iteratively optimize critical path timing (Fig. 6 ). Using the ADFF method, total timing violations are nearly eliminated, achieving a 55.6% efficiency improvement over the built-in auto useful skew function (Table 2 ). For hold time violations in the reg2mem group (Figs. 9 and 10 ), 125 buffers are inserted. Across 50 critical paths, the total timing margin improves significantly, reducing the worst slack from –362.2 ns to near zero (Table 3 ). To further optimize timing, when a clock signal is transmitted to two physically close registers, and buffers in the final clock path stage are used for timing correction, a merging plan consolidates multiple register buffers into a single delay unit (Fig. 11 ). Through a rigorous filtering mechanism in the script language, nearly 700 clock delay units are reduced while maintaining timing integrity. Additionally, clock network nets are reduced (Table 4 ), improving clock tree quality, achieving timing convergence, and enhancing overall design efficiency. Conclusions This paper proposes a novel ADFF clock tree optimization strategy that integrates loop iteration adjustment with IC Compiler II, leveraging useful skew for adaptive full-flow automatic correction of setup and hold time violations. The method extends the traditional concept and has demonstrated significant results on a high-performance RISC-V-based CPU, achieving a main clock frequency of 800 MHz. Compared to conventional timing optimization methods, this strategy resolves timing violations that standard layout and routing processes cannot address, significantly improving timing convergence. Through joint optimization of power consumption and timing, the method reduces the cost and power overhead associated with useful skew optimization and is applicable to CPU pipelines, providing a valuable reference for chip design. Future research may refine filtering conditions, optimize script traversal statements, and incorporate mainstream tool techniques to improve path selection efficiency while minimizing runtime overhead in large designs. Additionally, further refinement of the mathematical model could help identify more suitable targets for power optimization, improving overall performance.
2025,
47(4):
1202-1214.
doi: 10.11999/JEIT240823
Abstract:
Objective As demand for 4k and 8k Ultra High Definition (UHD) videos increases, the latest generation of video coding standards has been developed to meet the growing need for UHD video transmission. UHD video coding requires processing more pixels and details, resulting in significant increases in computational complexity and resource consumption. Optimizing algorithms and implementing hardware acceleration are essential for achieving real-time encoding and decoding of UHD videos. In Alliance for Open Media Video 1 (AV1), richer intra-prediction modes have been introduced, expanding the number of modes from 10 in VP9 to 61, thereby increasing computational complexity. To address the added complexity of these modes and enhance hardware processing throughput, a hardware design for AV1 Rough Mode Decision (RMD) based on a fully pipelined architecture is proposed. Methods At the algorithm level, a 4×4 block is used as the minimum processing unit. RMD is applied to various sizes of Prediction Units (PUs) within a 64×64 Coding Tree Unit (CTU) following Z-order scanning. This approach allows for efficient processing of large blocks by dividing them into smaller, manageable units. To reduce computational complexity, the SATD cost calculations for different PU sizes (e.g., 1:2, 1:4, 2:1, and 4:1) are performed using a cost accumulation approximation method based on the 1:1 PU. This method minimizes the need to recalculate costs for every possible configuration, thus improving efficiency and reducing computational load. At the hardware level, the architecture supports RMD for PUs of various sizes (4×4 to 32×32) within a 64×64 CTU. This architecture differs from traditional designs, which use separate circuits for each PU size. It optimizes logical resource use and minimizes downtime. The design incorporates a 28-stage pipeline that enables parallel processing of intra-prediction modes, ensuring RMD for at least 16 pixels per clock cycle and significantly enhancing throughput and encoding efficiency. Additionally, the design emphasizes circuit compatibility and reusability across various PU sizes, reducing redundancy and maximizing hardware resource utilization. Results and Discussions Software analysis shows that the proposed AV1 coarse mode decision algorithm reduces processing time by an average of 45.78% compared to the standard AV1 algorithm under the All-Intra (AI) configuration, while achieving a 1.94% improvement in BD-Rate. The testing platform is an Intel(R) Core(TM) i9-9900K CPU @ 3.60 GHz with 16.0 GB of DRAM. Compared to existing methods, the algorithm significantly reduces processing time while maintaining encoding efficiency. It offers an optimized trade-off, with a slight BD-Rate loss in exchange for substantial reductions in encoding time. Hardware analysis reveals that the proposed hardware architecture has a total circuit area of 0.556 mm² after synthesis, with a maximum operating frequency of 432.7 MHz, enabling real-time encoding of 8k@50.6fps video. Although the circuit area is slightly larger than in existing designs, the architecture demonstrates significant improvements in processing speed and video resolution capability, providing a balanced trade-off between hardware resource usage and throughput/area efficiency. These results further confirm the design’s superiority in terms of hardware resource efficiency and processing performance. Conclusions This paper presents a high-throughput hardware design for AV1 RMD, capable of processing all PU sizes with 56 directional and 5 non-directional prediction modes. The design employs a 28-stage pipeline for parallel intra-frame prediction mode processing, enabling RMD for at least 16 pixels per clock cycle and significantly improving encoding efficiency. Techniques such as false-reconstructed reference pixel, Z-order scanning, PMCM circuit structures, and circuit reuse address the increased hardware resource demands of parallel processing. Experimental results show that the proposed algorithm reduces processing time by an average of 45.78% and improves BD-Rate by 1.94% compared to the AV1 standard, ensuring high speed and encoding quality. Circuit synthesis confirms the architecture’s capability for real-time 8k@50.6fps video processing, meeting the demands of future UHD video encoding with exceptional performance and efficiency.
