With the widespread deployment of heterogeneous operating and computing devices in modern flexible manufacturing, stream processing applications are burdened by mismatches of computing capacity and demand, making it difficult to maintain service level objective (SLO) compliance in real-time manufacturing. To this end, we propose industry-EI, a cloud-edge collaborative scheduling algorithm inspired by embodied intelligence (EI). Industry-EI models the industrial system as a closed-loop embodied agent that continuously perceives the environment, makes scheduling decisions, and receives feedback on execution outcomes. To support real-time interaction with complex industrial environments, we design a perception-decision-control architecture that integrates deep reinforcement learning to adapt across heterogeneous production lines. To update the EI agent efficiently under dynamic workloads, we develop an iterative learning paradigm incorporating timestamp tracking and differential reward, which handles delayed and misaligned feedback phenomena and enables optimal decision-making. We implement a KubeEdge-based prototype and validate industry-EI in a real-world environment. Experiments show that industry-EI reduces the 95th percentile (P95) latency by 40.7% and achieves a SLO compliance rate of 93.7%.
Low earth orbit (LEO) satellite networks are entering a critical phase of large-scale global deployment. However, their high dynamics pose unprecedented challenges to traditional terrestrial mobile communication and networking protocols. Although artificial intelligence (AI) technology provides new solution paths for high-efficiency and low-complexity operations, it still faces severe technical bottlenecks in practical application and deployment. This paper systematically analyzes three key challenges during the intelligent evolution of LEO satellites: model generalization issue, constrained payload capabilities, and decision latency bottleneck. In response to these challenges, this paper explores potential enabling AI technologies, including the use of meta-learning and graph neural networks to enhance model generalization, the implementation of model compression and lightweight strategies, and the application of generative AI and delay-tolerant reinforcement learning to improve resource management efficiency and decision robustness. On this basis, a further outlook on typical AI-enabled application scenarios is provided: at the communication transmission level, the paper highlights generative AI-driven channel estimation, delay-tolerant beam management, and interference suppression techniques based on diffusion models; at the networking level, graph-based routing strategies and meta-learning-based handover management schemes are discussed. The deep integration of AI technology and satellite communication regimes will serve as a critical support for constructing autonomous and ubiquitously intelligent integrated space information systems in the 6G era.
Non-terrestrial networks (NTNs) are a cornerstone for 6G's global connectivity vision, but their hyper-dynamic nature presents unprecedented challenges in optimization, management, and security. This paper posits that the synergistic convergence of artificial intelligence (AI) and quantum technologies offers a transformative paradigm to address these issues. We conduct a comprehensive survey, presenting a novel taxonomy that classifies applications into three domains: quantum-enhanced AI, AI-powered quantum systems, and converged AI-quantum services. For each, we analyze formal problem formulations and hybrid solution architectures. To bridge theory and practice, we develop a detailed case study on handover optimization in a low earth orbit (LEO) satellite constellation under dynamic channel conditions. By formulating the problem as a quadratic unconstrained binary optimization (QUBO) model and solving it with a quantum-inspired annealer, our Monte Carlo simulations demonstrate a superior performance trade-off. The proposed approach reduces handovers by approximately 84% compared to a classical greedy strategy while maintaining high link quality and zero outage. Finally, we identify critical open research challenges and outline a future vision for autonomous, intelligent, and unconditionally secure global networks.
The hybrid satellite-terrestrial relay network (HSTRN) is a promising solution to enhance link reliability. However, its performance analysis is challenged by the geometric constraints introduced by relay coverage regions and elevation angle limitations, as well as terrestrial co-channel interference (CCI). This paper investigates the uplink successful transmission probability of HSTRN within a stochastic geometry framework. In our analysis, low earth orbit (LEO) satellite constellations are modeled as a binomial point process (BPP), while interfering terminals are described by a Poisson point process (PPP). Analytical expressions for the uplink successful transmission probability and the co-channel interference are derived, and their accuracy is validated through Monte Carlo simulations. Furthermore, key design insights are obtained, including the optimal relay coverage radius and the optimal satellite altitude, providing valuable guidelines for system design.
Spatially reconfigurable antenna arrays (SRAAs) have recently emerged as a promising paradigm for enhancing wireless system performance by treating antenna position and orientation as new spatial degrees of freedom (DoFs). Unlike conventional fixed-geometry antenna arrays, SRAAs enable geometry-aware adaptation of the physical aperture, thereby allowing wireless systems to actively exploit spatial channel variations beyond signal-domain processing. This capability is particularly attractive for future sixth-generation (6G) networks that operate in highly dynamic propagation environments and face stringent performance requirements. This review provides a comprehensive and system-oriented overview of SRAAs from both theoretical and practical perspectives. Firstly, we present a unified and geometry-aware channel modeling framework for spatial reconfiguration at different architectural granularities. Secondly, we analyze how position- and orientation-induced channel variations, along with their combined effects, and enable performance gains without relying solely on massive antenna scaling. Afterwards, we survey design and optimization methods for position-orientation reconfiguration, covering both model- and learning-based techniques. Practical considerations are also discussed through a systematic review of hardware implementation options and channel estimation techniques under spatial reconfiguration. To further illustrate the system-level benefits of SRAAs, representative applications are examined, including point-to-point and multiuser multiple-input multiple-output (MIMO), cell-free massive MIMO, as well as aerial and mobile communications. A dedicated case study on six-dimensional aerial rotatable antenna (6DARA)-enabled cell-free networks is provided to demonstrate how array-wise position and orientation control, combined with distributed optimization, can achieve substantial performance gains with manageable complexity. Finally, we outline key issues and future directions for the large-scale and practical deployment of SRAAs in 6G wireless networks.
While millimeter wave (mmWave) communications offer vast bandwidth, their coverage is highly susceptible to blockages. Reconfigurable intelligent surface (RIS) have emerged as a promising solution to enhance mmWave connectivity. In this study, we analyze the coverage probability of RIS-assisted mmWave communications, which for the first time incorporates rectangular blockages, a two-step association strategy, and a far-field path loss model for RIS. Specifically, utilizing tools from stochastic geometry, we establish the system model, which includes the base station (BS), the blockage, the RIS, and the user distribution models. Subsequently, we define the association criterion and derive the conditional coverage probabilities for three distinct scenarios: line of sight (LOS) association, non-line-of-sight (NLOS) association, and RIS-assisted association. Finally, we combine the three cases to obtain the exact expressions for the coverage probability. Simulation results validate the theoretical analysis, showing that RIS provides up to 25% coverage gain and exhibits deployment-specific behavior under realistic blockage and interference. Notably, the benefits of RIS are most pronounced in heavily obstructed environments. Furthermore, as RIS deployment density and surface size increase, the optimal signal to interference plus noise ratio (SINR) threshold for maximizing area spectral efficiency (ASE) scales upward, shifting the network's peak performance toward higher operating regimes.
To enable reliable and efficient communication under time-varying channel conditions, this paper proposes a feedback-based symbol rate adaptation scheme for ultraviolet (UV) communication. The receiver periodically reports channel state information, including the mean detected photon counts of signal and background components, to the transmitter after a fixed number of frames. Based on this feedback and predefined rate-switching thresholds obtained from off-line low-density parity-check (LDPC) performance simulations, the receiver selects appropriate symbol rates from a set of discrete levels and conveys the update decisions through a feedback channel. A lightweight decision logic is adopted to enable timely rate adjustments, and a feedback frame structure with frame indices is designed to ensure reliable symbol rate synchronization under frame loss. Indoor experiments with controlled link distance and noise variations verify the predictable adaptation behavior of the proposed scheme, while outdoor daylight experiments demonstrate stable communication and effective symbol rate tracking under time-varying channel conditions.
Satellite-ground semantic communication is an important component of the forthcoming 6G era. Due to the strict bandwidth limitations of low earth orbit (LEO) satellites, efficient transmission of massive satellite remote sensing images is difficult. To this end, we propose a channel-aware adaptive semantic communication method for LEO satellite remote image transmission. The proposed method includes a feature extraction module, an important feature enhancement module, a rate adaptive module, and the corresponding decoding modules. The feature extraction module can extract global context information via lightweight mamba. The important feature enhancement module can enhance useful features via the involution layer and the signal to noise radio (SNR) adaptive block according to channel state SNRs. The rate adaptive module can further adaptively adjust the size of transmission features according to the transmission rate via the rate adaptive block and the rate mask block. The extensive experimental results on the WHU-RS19 dataset demonstrate that our method obtains higher peak signal to noise radio (PSNR), multi-scale structural similarity index measure (MS-SSIM) and learned perceptual image patch similarity (LPIPS) than state-of-the-art methods under low SNR and limited bandwidth conditions.
Reconfigurable distributed antennas and reflecting surface (RDARS) has emerged as a transformative solution to address the stringent requirements of future wireless networks. By combining distributed active antennas with reconfigurable passive reflecting surfaces, RDARS integrates the advantages of both active transmission and passive wave control in a cost-effective and energy-efficient manner. This hybrid architecture enables enhanced coverage, improved spectral efficiency, and seamless support for integrated communication and sensing. In this article, we first introduce the fundamental architecture and working principles of RDARS, followed by practical benefits and comparisons with recently proposed intelligent surface variants. We then highlight the signal-to-noise ratio (SNR) gains in representative applications of RDARS-aided communication and sensing scenarios, where RDARS demonstrates clear advantages over conventional reconfigurable intelligent surfaces. Finally, we outline key challenges related to practical implementation and resource allocation, and discuss potential research directions. With its unique hybrid mode synergy, RDARS is envisioned to play a pivotal role in shaping the evolution of next-generation intelligent communication systems.
The advent of extremely large-scale arrays technology is the cornerstone of the sixth-generation wireless networks, promising significant advances in spectrum efficiency. However, compared with traditional far-field beam training methods, near-field beam training faces greater challenges as the spherical wavefront propagation characteristic in the near-field environments necessitates beam search in both the angle and distance dimensions. To reduce the training overhead of the two-dimensional search, we propose a data-driven on-grid beam training scheme that can simultaneously search the angle and distance domains to find the optimal codewords through real-time adaptive alignments. To further address the grid-based non-uniform sampling, we propose a datadriven off-grid adaptive optimization scheme to further improve near-field beam training accuracy with fast convergence. By establishing equivalent dynamic linearization data models, the proposed approach adaptively adjusts the angle-distance domain estimation based on real-time measurements. Numerical results show that the proposed approach can achieve enhanced beamforming performance with reduced training overhead.
Conventional transmissive metasurface design for Internet of things(IoT)communications relies heavily on expert knowledge and iterative full-wave simulations,resulting in high computational cost and limited efficiency.To address this challenge,we propose an end-to-end deep-learning-based design framework for sub-6 GHz transmissive communication metasurfaces,which enables automatic mapping from target transmis-sion responses to manufacturable physical structures.We first develop a prediction model,Img2S,to accurately estimate the S-parameters of metasurfaces,significantly reducing the need for full-wave simulations.Based on this model,two variational generative networks,strictly constrained-conditional variational autoencoder(SC-CVAE)and loosely constrained-conditional variational autoencoder(LC-CVAE),are proposed to synthesize physically realizable metasurface structures by incorpo-rating geometric priors and electromagnetic consistency constraints.Experimental results show that Img2S achieves a mean squared error(MSE)of 9.76 × 10-4 in predicting the simulated S-parameters of metasurfaces over the operating frequency band.Both simulation and measurement results confirm that the generated metasurfaces closely match the target electromagnetic responses,with single-state mean absolute errors(MAEs)below 0.16 in simulation and below 0.31 in measure-ment,respectively,outperforming conventional design approaches in terms of accuracy and frequency stability while significantly improving the overall design efficiency.
Unmanned aerial vehicle (UAV) edge computing effectively reduces task latency and mitigates computing pressure for ground terminals (GTs), particularly in scenarios lacking fixed terrestrial infrastructure. This paper constructs a novel framework for a multi-UAV edge computing system with cross-terminal dependent subtasks, in which the task offloading decision, communication bandwidth allocation, and UAV trajectory planning are jointly optimized. Unlike traditional task offloading schemes, the internal dependency relationships of subtasks impose complex temporal constraints on task offloading decision. Firstly, a directed acyclic graph (DAG) is employed to describe the structure of dependent subtasks. Accounting for computing timeliness requirements and UAV energy constraints, a system cost based on weighted delay and energy consumption is defined. Subsequently, a long-term optimization problem with the objective of minimizing system cost is formulated. In order to solve this complex non-convex mixed-integer programming problem, an algorithm combined with a pre-trained graph attention network (GAT) and the proximal policy optimization (PPO) is proposed. GAT utilizes its specialized graph-processing capabilities to extract high-level subtask features from the DAG. Then PPO integrates these high-dimensional features with environmental state information for global reasoning to obtain the task offloading decision and the UAV trajectory planning. Comprehensive simulations demonstrate that the proposed algorithm effectively reduces system cost under varying system parameters and successfully addresses the unique challenges of a multi-UAV edge computing system with dependent tasks.
Contactless human-computer interaction (HCI) based on radio frequency (RF) signals is critical for next-generation intelligent environments. However, this field lacks a standardized benchmark dataset for practical interaction commands and faces challenges in heterogeneous signal fusion. We address this by constructing and releasing WRF-G20, a novel Wi-Fi and radio frequency identification (RFID) multimodal gesture dataset. It encompasses 20 refined gestures for cutting-edge applications and adheres to the standardized XRF55 protocol for reproducibility. To address the fusion challenge, we propose the mixture of experts-contrastive learning network (MoE-CLNet) framework. It first employs cross-modal contrastive learning to map heterogeneous signals into a shared semantic space, then utilizes a Top-K sparse MoE architecture to perform adaptive dynamic weighted fusion based on the signal quality characteristics of each input sample. On the WRF-G20 benchmark, MoE-CLNet achieves a 92.78% F1-score, validating the dataset's high quality and learnability while outperforming existing baseline models. The framework also demonstrates robust generalization on the public XRF55 benchmark, attaining 91.81% accuracy. This work provides the heterogeneous RF sensing field with critical research infrastructure and a high-performance technical reference.
Frequency hopping (FH) plays a significant role in wireless ad hoc networks (WANETs) by enhancing communication security, reliability, and resistance to jamming. However, conventional frequency hopping schemes rely on pre-arranged hopping sequences and strict time synchronization among participating nodes, which limits their adaptability in dynamic and decentralized environments. Motivated by these limitations, we propose a random frequency hopping framework for multi-hop broadcasts in WANETs in combination with carrier sense multiple access with collision avoidance (CSMA/CA). To analyze the performance of the proposed framework, we leverage the analogy between information dissemination and epidemic spreading to develop an analytical model, which characterizes message propagation as an annular infection process, incorporating the effects of frequency matching probability, contention-based access control, and equivalent communication radius under Rician fading. Simulation results demonstrate that the proposed analytical model can effectively predict the process of message propagation, achieving a mean absolute percentage error (MAPE) of less than 5%. Across diverse network configurations, the analytical and simulation curves remain closely aligned, confirming the robustness and validity of the proposed model.
In the information age, individuals and organizations have accumulated massive volumes of digital data. As society enters the intelligence era, the primary demand has shifted toward effectively utilizing these data through intelligent models. However, prevailing large language models (LLMs) typically require data to be uploaded to centralized cloud servers for processing, making it difficult to ensure data privacy. Consequently, users increasingly seek small, locally deployable models that can operate directly on private data. To realize this expectation, this paper explores intelligent systems built upon small models rather than monolithic large models. On this basis, we first revisit the essence of intelligence as the capability to utilize knowledge to solve problems and accomplish tasks, and further propose the first principle of intelligence, which models intelligence as a closed-loop, goal-oriented process. Within this process, an agent, under external constraints, leverages knowledge to evaluate discrepancies between its internal state and task objectives, formulates strategies to reduce these discrepancies, and iteratively executes actions to converge toward goal completion. From this perspective, intelligence is viewed as a collaborative system of interdependent cognitive functions. Grounded in this theoretical principle, we propose the six-capability network, a knowledge-driven cognitive architecture that decomposes intelligence into six fundamental capabilities: observation, attention, understanding, discrimination, memory, and execution. These capabilities constitute the core cognitive model of the proposed framework and are each realized by lightweight, deployable small models operating over structured knowledge representations. Finally, to demonstrate the executability of the network, we consider a networked intelligence system. In this system, agents exchange semantic symbols that encode knowledge rather than raw data, enabling each agent to operate within its functional scope while acquiring missing knowledge from others when needed. This approach offers a new intelligence paradigm, enabling deployment in private, local, and resource-constrained environments.
This paper investigates the maximization problem of secrecy rate in underlay cognitive radio networks (CRNs) with the aid of reconfigurable intelligent surface (RIS). The secure communication method is studied, which takes into account the underlay spectrum access mode, the characteristics of RIS, and the threat of eavesdropping user to cognitive network communication. Using convex optimization theory, the secrecy rate of cognitive networks is maximized by jointly optimizing cognitive base station (CBS) beamforming and RIS phase shifts. Based on the proposed optimization and analyzing method, the impact of network parameters on the secrecy rate is analyzed in depth, and the security of cognitive networks is discussed. Theoretical analysis and simulation have shown that the proposed method can significantly improve the security of cognitive networks. With the aid of RIS, the secrecy rate is greatly improved compared to the case without RIS. Further, the secrecy rate is improved even more through optimizing RIS phase shifts and CBS beamforming.
With the advantages of low cost and easy deployment, ambient backscatter has attracted significant attention in the Internet of things (IoT) community and experienced rapid growth over the past decade. As the number of deployed backscatter devices increases, enabling concurrent transmission among multiple tags has become critical for achieving high throughput and large-scale network coverage. To this end, this article presents a comprehensive survey of concurrent transmission techniques. To provide a structured overview, we categorize existing studies on backscatter concurrency into five implementation-oriented domains, including time, frequency, spatial, code, and energy, according to their technical characteristics. For each domain, we comprehensively review representative works, emphasizing their underlying principles and key designs. Furthermore, we investigate the growing trend of multi-domain integration, demonstrating how cross-layer synergies enhance performance and facilitate system evolution. Finally, we outline open challenges and future research directions for concurrent backscatter communication.
The integration of wireless communication and state estimation has become a fundamental enabler for large-scale industrial Internet of things(IIoT)systems,where estimation,transmission,and control are tightly coupled across heterogeneous networks.This survey pro-vides an overview of recent advances in state estimation and communication co-design,highlighting the evolution from isolated subsystem optimization toward unified frameworks that jointly address estimation accuracy,communication efficiency,and scalability.We first review theoretical foundations that characterize how data rate and packet loss constrain estimation stability,introducing the key results such as data-rate theorem and critical loss rate for estimation convergence.Next,we discuss the scalability perspective,addressing the horizontal expansion through multihop relaying to overcome trans-mission distance limitations,and the vertical expansion through multi-sensor fusion to address the limited ob-servation range of individual sensors.At the network level,we discuss layered design methodologies across the application,transport,medium access control(MAC),and physical layers to ensure estimation performance.Subsequently,we examine control-oriented co-design paradigms,including separation principle-based designs,joint optimization without separation,and learning-based frameworks that integrate estimation,communication,and control in a unified manner.Finally,we discuss several inspiring future research directions.
The rapid proliferation of Internet of things (IoT) devices has led to an unprecedented increase in spectrum congestion within the unlicensed industrial, scientific, and medical (ISM) bands. Heterogeneous systems coexist and compete for the same frequency resources, resulting in severe structured interference and reducing communication reliability. Traditional interference suppression techniques rely on fixed modulation schemes, which fail to adapt to dynamic interference patterns in multi-protocol environments. This paper presents DyniaPHY, a neural network-based physical layer framework that can autonomously extract the available channel state feature of any interference-prone channel and generates the optimal modulation scheme to achieve anti-interference communication performance under various channel conditions. Experimental results show that DyniaPHY can dynamically adjust encoding strategies to avoid or counteract structured interference and compensate for carrier frequency offsets and sampling frequency offsets.
Artificial intelligence (AI) has become a key enabler for next-generation wireless communication systems, offering powerful tools to cope with the increasing complexity, dynamics, and heterogeneity of modern wireless environments. To illustrate the role and the impact of AI in wireless communications, this paper takes collaborative spectrum sensing (CSS) in cognitive and intelligent wireless networks as a representative application and surveys recent advances from an AI perspective. We first introduce the fundamentals of CSS, including general framework, classical detector design, fusion strategies, and evaluation metrics. Then, we present an overview of the state-of-the-art research on AI-driven CSS, classified into three categories according to learning paradigms: discriminative deep learning (DL), generative DL models, and deep reinforcement learning (DRL). Building on this, we explore AI-empowered semantic communication (SemCom) as a paradigm-shifting solution for CSS. By extracting and transmitting task-relevant features, Sem-Com upgrades CSS from a computation-centric approach to a highly efficient joint communication and computation framework. Both single-user and multi-user SemCom scenarios are elaborated in detail. Finally, we discuss limitations, open challenges, and future research directions at the intersection of AI and wireless communication.
