Satellite-to-ground communications are anticipated to play a crucial role in sixth-generation (6G) mobile systems by providing seamless coverage. Due to the long transmission distances, complex channel characteristics, and diverse terminal environments, satellite-to-ground channels exhibit unique features that distinguish them from terrestrial communication channels. Understanding and accurately modeling the channel is a premise for the design, optimization, and evaluation of satellite-to-ground communication systems. This paper provides a comprehensive review of the challenges and ongoing research in satellite-to-ground channel measurement, characterization, modeling, and standardization. Specifically, the paper discusses simulated and on-orbit satellite channel measurements, introduces large-scale and small-scale characteristics, and analyzes the challenges and progress in four satellite-to-ground channel models. In addition, the paper reviews the relevant standards efforts of key organizations and outlines potential future research directions.

In this work, we propose a multi-attempt successive cancellation list (MA-SCL) decoder for polar codes that achieves identical error-correction performance to standard SCL decoding while reducing average complexity. Unlike CRC-aided SCL, the proposed MA-SCL progressively restarts decoding with increasing list sizes and reuses information from previous attempts. This design eliminates the need for outer CRC codes. The decoder features dynamic search-space pruning and an early stopping criterion based on path metrics. Simulations show MA-SCL matches SCL performance with lower average complexity, particularly for short polar-like codes with reed-muller (RM) rate profiles and dynamic frozen constraints. Compared to existing adaptive decoders, MA-SCL offers implementation advantages by eliminating the need for stack/heap management while providing relatively stable latency bounds ($1\times$ to $|\Lambda|\times$ SCL latency).

This paper proposes a novel end-to-end learnable framework for semantic image transmission, pioneering joint optimization across the spatial, channel, and power domains. In this scheme, the transmitter employs semantic analysis with spatial-channel domain adaptation to extract and compress vital semantic features from image latent representations, enabling efficient compression by integrating spatial structures and retaining channel priority attributes. Subsequently, a dynamic power allocation strategy intelligently adjusts the transmission power of these features based on real-time noise conditions to mitigate channel impairments. At the receiver, a hierarchical reconstruction network subsequently decodes images through cross-feature analysis of semantic relationships from distorted features. Extensive experimental validation under Rayleigh fading channels demonstrates that the proposed framework achieves significantly superior bandwidth utilization and reconstruction quality compared to existing seep joint source channel coding (DJSCC) schemes. It exhibits robust performance across diverse channel conditions and compression ratios, thereby establishing a new benchmark for semantic communications (SemCom).

Escalating demands for HD linear streaming and seamless mobility challenge 6G cellular networks, where 5G-optimized resource strategies approach saturation. Broadcast networks, with inherent spectral efficiency and coverage advantages, offer complementary capacity. We propose a converged cellular-broadcast architecture integrating a novel 6G broadcast core, enabling dynamic cellular traffic offloading to broadcast networks while enhancing broadcast reception via cellular links. A User-perception optimization framework jointly addressing broadcast directionality, resource allocation, and power control is established and solved by an efficient multi-stage heuristic algorithm. Simulations confirm that the proposed broadcast core significantly alleviates cellular congestion and improves broadcast resource utilization by comparing with existing literature. Field trials in campus environments demonstrate consistent operational efficacy, validating the architecture’s practical feasibility for 6G media delivery.

In this paper, a large-scale field trial campaign is conducted for the passive reconfigurable intelligent surfaces~(RIS) in Chongqing, China, a mountainous megacity characterized by complex terrain, to validate scenario-adaptive coverage enhancement in the existing 5G commercial networks. Operating RIS at 2.6 GHz (512 elements) and 4.9 GHz (1,024 elements), the study systematically evaluates eight real-world scenarios, including mountain tunnels, river crossing bridges, riverside roads, iconic landmarks, business districts, residential areas, rural settlements, and scenic areas. These trials incorporate real-world uncertainties, extending the analysis from downlink coverage to comprehensively evaluating uplink / downlink channel quality, throughput, and regional interference metrics. The key results demonstrate significant performance gains, such as a maximum increase in received signal strength by 6.14 dB in residential areas and a maximum gain in channel quality 351% on riverside roads, along with a maximum throughput improvement of up to 89\% in uplink transmissions. The study further proposes a scenario applicability evaluation framework that includes five dimensions, considering deployment constraints and performance gains, to guide the deployment of RIS in future networks.

As the sixth generation (6G) networks aim to support increasingly diverse, dynamic and intelligent services, conventional radio access network (RAN) architectures face challenges in flexibility, scalability and service responsiveness. In this context, the concept of service-based RAN has emerged as a promising architectural evolution to achieve multi-dimensional coordination across function, service and resource. This paper presents an intelligent service-based RAN architecture using embedded Artificial Intelligence (AI) techniques, characterized by a dual-module design that integrates functional recomposition and intent-driven orchestration. Specifically, the functional decoupling module introduces a two-stage mechanism, including semantic decoupling and function recomposition. Leveraging large language models (LLMs) for semantic parsing of heterogeneous protocol documents, atomic functions are extracted and standardized. These functions are then aggregated to form RAN services for the control plane and user plane respectively. On the orchestration side, we develop an intent-driven approach in which LLMs parse high-level service requirements and translate them into service function chains and resource mappings. Simulation results validate the effectiveness of the proposed functional recomposition and orchestration, highlighting reliable guarantee for the intent of users. Finally, several key challenges are identified thatwill be critical to the future evolution of intelligent service-based RAN.

In unmanned aerial vehicle (UAV)-assisted millimeter-wave (mmWave) multi-input multi-output (MIMO) communications with malicious reactive jamming (RJ), signal leakage through antenna side lobes may expose frequency-band information to RJ, degrading secure transmission rate (STR). To address this issue, this paper proposes a deception-based secure communication scheme (Dec-SC) for UAV-assisted mmWave MIMO systems. The main idea is to send a deceptive signal in a different frequency band and exploit the UAV’s mobility to reduce the leakage of the legitimate signal to RJ, deceiving RJ to jam the wrong band and enhancing STR. To maximize the deception-based STR (DS-STR), the beamforming vectors, UAV position, and power allocation between legitimate and deceptive signals are optimized with a deception probability constraint. A low-complexity suboptimal algorithm, named signal leakage minimization-based three-phase optimization (SLM-TPO), is proposed. Simulation results demonstrate that the proposed Dec-SC scheme achieves near-optimal performance, with a gap of no more than 3% from the global optimum. Moreover, it can improve DS-STR by over 9% compared to direct-link transmission without UAV and by more than 100% compared to the no-deception scheme.

Herein, we develop a wireless temperature monitoring system to address the insufficient real-time, high-precision, three-dimensional temperature field monitoring data for large-scale wind power generators (LSWPGs) thermal design. Our system is based on SmartMesh IP wireless mesh networks. Our innovations include: 1) the development of high-temperature-resistant, centrifugal-force-resistant miniaturized wireless sensors, 2) implementation of the time-slotted channel hopping protocol to achieve sub-200 ms communication latency with a 99.999% packet delivery rate, 3) real-time temperature field reconstruction to achieve 95.9% hotspot identification precision and $\pm$1$^\circ$C measurement accuracy. Based on experimental data, we perform a coupled temperature field-flow field analysis to optimize LSWPG cooling performance. High-fidelity wireless sensing networks are combined with multiphysics simulation, forming a closed-loop, real-time monitoring system for the thermal design and optimization of LSWPGs. The system is deployed in 6--26 MW permanent magnet synchronous generators (Dongfang Electric Machinery Co., Ltd.), including the world's largest 26~MW offshore wind turbine (as of 2024).

Conventional path protection algorithms in all-optical data center networks (DCNs) need a relatively long calculation time but with low efficiency. In this article, we propose a preset path recovery algorithm by combining machine learning and priority execution method. The former is to predict areas of frequent optical path failures in data centers, and the latter is to ensure accurate protection in those areas while reducing computation time and improving performance. Firstly, we construct a dataset using rules composed of topological edge relationships, failure time, failure frequency, packet loss rate, latency and service type. Then, based on the above dataset, SVM (support vector machine) algorithm is used to predict frequent failure areas. Finally, based on different prediction frequencies, priority construction is carried out to cluster the entire topology, and p-cycle (preconfigured protection cycle) is applied to each topology. Experiments show that the combination of SVM prediction and classification protection can significantly reduce the protection range of potential failure areas to reduce computation time while ensuring high accuracy, as compared with traditional optical path protection algorithms.

In this paper, we propose a Transformer-OP co-optimized routing framework for UAV optical communication networks to address routing challenges. The framework features an encoder that employs multi-head attention mechanisms to extract spatiotemporal correlation features among nodes. The de-coder incorporates triple dynamic masking for path generation while dynamically evaluating link quality using Gamma-Gamma turbulence modeling. The REINFORCE reinforcement learning algorithm is applied to optimize reward functions, achieving an optimal trade-off between mission performance and turbulence-induced losses. Experimental results demonstrate that our method significantly improves routing stability in dynamic turbulence environments, enables millisecond-level routing decisions, and enhances overall system efficiency by more than 30%, providing valuable theoretical foundations and technical solutions for space-air-ground integrated communications.

The security of maritime communication for transmitting sea clutter signals is of critical importance from both information-theoretic and electromagnetic signal processing perspectives. Existing prediction models face significant challenges in accurately capturing the non-stationary and chaotic characteristics of sea clutter, a typical electromagnetic scattering signal influenced by complex maritime environments. To address this, we propose a novel maritime covert communication scheme based on information-theoretic secrecy metrics, leveraging communication relay unmanned aerial vehicles to minimize the monitor's detection probability. A hybrid neural network model is developed for sea clutter prediction by integrating convolutional neural networks (CNN), bidirectional long short-term memory (BiLSTM), and an attention mechanism. Initially, phase space reconstruction is applied to sea clutter signals measured by IPIX radar, exploiting their spatiotemporal correlation—a key property in electromagnetic signal analysis. The CNN extracts spatial features from the reconstructed signals, while the BiLSTM models temporal dependencies, effectively mitigating overfitting in long-sequence prediction. The attention mechanism further enhances performance by dynamically weighting salient features. Experimental results demonstrate superior prediction accuracy and robust target detection capability based on prediction errors. This work bridges deep learning-based electromagnetic signal processing and covert communication design, providing insights for the design of secure maritime information systems.

In extremely high-mobility scenarios, the severe wideband Doppler effect significantly deteriorates the performance of traditional acquisition methods and increases computational complexity. Using the generalized wideband ambiguity function (GWAF), this paper derives the analytical and approximate closed-form expressions of the GWAF for the linear frequency modulated (LFM) signal and further clarifies the mechanism of correlation acquisition performance degradation for LFM signals under extremely high-mobility conditions. To overcome the limitation of performance degradation, an optimization method for the superposed LFM signal, based on its GWAF, using an alternating iterative genetic algorithm (GA) is proposed to improve the time-domain correlation performance of the superposed LFM signal, satisfying the requirements for time-domain correlation acquisition in low SNR and extreme high mobility scenarios. Based on the optimized signal, a low-complexity iterative superposition matched filtering (ISMF) acquisition method is designed for extremely high-mobility signals. Simulation results demonstrate the validity of the theoretical expressions of the GWAF and the proposed ISMF acquisition method based on optimized waveforms, achieving performance comparable to traditional methods while significantly reducing computational complexity.

The satellite-ground integrated Networks (SGIN) emerge as a promising paradigm to extend the coverage and resilience of terrestrial networks. However, the high mobility and intermittent connectivity of satellites lead to inevitable ground-satellite handovers. Existing handover algorithms often overlook the inherent interdependence between ground-satellite handover and inter-satellite routing, resulting in suboptimal performance and degraded quality of service (QoS). To address these issues, we propose a heterogeneous graph neural networks-enhanced deep reinforcement learning (HGRL) algorithm for joint handover and routing optimization. First, we propose the semantic-based heterogeneous graph neural networks (SHGNN) to model SGIN as a heterogeneous graph, capturing the intricate relationships between handover and routing through diverse representations of nodes and edges. Then, we embed the SHGNN into a deep reinforcement learning (DRL) framework, enabling QoS-aware decisions for both ground-satellite handover and inter-satellite routing. Additionally, a non-dominated crowding sorting (NCS) mechanism is proposed to prune alternative paths while balancing multiple QoS objectives. Finally, extensive simulations in NS3 show that HGRL outperforms state-of-the-art algorithms, reducing the handover times and average delay by 63.63% and 36.85%, and improving the average throughput by 26.53%.

This paper proposes a two-user cooperative physical layer security (PLS) satellite transmission scheme based on the weighted fractional fourier transform (WFRFT). This approach achieves energy-efficient secure transmission in two-way satellite relay by superimposed WFRFT signals with different orders between cooperative users. The Gaussian-like characteristics of the superimposed WFRFT signals with order differences make them inseparable, preventing eavesdroppers from obtaining the correct transform order for computational attacks, thereby enhancing the resistance to computational analysis. Meanwhile, the signals from the two cooperative users form equivalent strong artificial noise (AN) to each other, reducing eavesdropping channel capacity, and WFRFT parameter errors further reduce the eavesdropping channel capacity. As a result, the proposed scheme improves secrecy capacity and secrecy energy efficiency (SEE) without consuming additional interference power. Based on the Gaussianity of the superimposed signal, an order selection method for the cooperative users is provided. The PLS performance of the proposed scheme is analyzed, and expressions for the secrecy capacity and SEE are derived. Simulation results are provided to validate the performance of the proposed method.

To address the critical challenges of non-uniform resource sensing and high dynamism within terminal-side computing power networks, this paper proposes a novel and efficient hierarchical resource scheduling mechanism. Firstly, architect a collaborative network architecture integrating a terminal layer and a cloud layer. Subsequently, a multi-dimensional model for computing power sensing and standardized measurement is established. Furthermore, investigate a hierarchical scheduling mechanism based on federated learning, which facilitates the effective management and intelligent scheduling of heterogeneous, dynamic resources. Experimental results demonstrate that this mechanism significantly reduces service latency in near-field computing, terminal-cloud collaboration, and ubiquitous computing scenarios.

The limited onboard cache and computing resources significantly constrain the computational service capabilities of individual edge satellites in hotspot regions. To address this challenge, we propose a two-tier cloud-edge computing architecture that organizes edge satellites and their associated ground clouds into multiple collaborative domains. Within each domain, we formulate a joint optimization problem for computation offloading and service downloading under the constraints of edge satellites' service deployment and caching space, aiming to minimize the sum of weighted energy consumption and latency. The originally non-convex problem is transformed into a more tractable convex optimization formulation through variable relaxation. Subsequently, we develop an alternating direction method of multipliers (ADMM)-based distributed optimization framework that enables cooperative decision-making among domain satellites for the optimization of computational offloading, service downloading, and service deleting variables. Additionally, we propose an innovative binary variable recovery algorithm that ensures feasible conversion from continuous solutions to discrete decision variables while preserving constraint satisfaction. Extensive simulations demonstrate that our approach achieves lower task execution cost and packet loss rate compared with benchmarks.

Low earth orbit (LEO) satellites have recently been applied in various Earth observation applications, generating massive observation data. Promptly transmitting such observation data to the ground is hindered by satellite-ground transmission links with limited data rates. To address this challenge, effective transmission and orbital edge computing schemes are explored to enable continuous transmission and computation offloading via inter-satellite links (ISLs). However, most existing works primarily emphasize the joint optimization of computation and communication resources while ignoring the distribution of observation resources and the coexistence of heterogeneous tasks, thus resulting in degraded performance due to the mismatch between satellite resources and heterogeneous task requirements. This paper proposes cooperatively scheduling satellites for observation, relay, and computation to maximize the number of completed observation tasks with diverse requirements, which is formulated as a mixed-integer linear programming problem constrained by satellites' observation, transmission, and computing resources. An iterative resource-aware scheduling algorithm, JORCA, is designed to find the solution by decomposing the original problem into two sub-problems. Extensive experimental results reveal that JORCA can improve the number of completed observation tasks by up to 40.6% compared to the benchmark policies.

6G is driving the evolution of Internet of Things from Internet of Everything to Intelligent Internet of Everything, requiring globally seamless communication coverage and efficient computing capabilities at network nodes. To this end, this paper proposes a low earth orbit (LEO) satellite and high altitude platform (HAP) cooperative edge computing network architecture, where ground users can either compute tasks locally or offload them to the HAP or access satellites. Particularly, when local computation resources are insufficient, access satellites can schedule available resources from other satellites via inter-satellite links. Specifically, access satellites incrementally expand a candidate set hop-by-hop until accumulated computation resources meet task requirements and then coordinate with satellites in the set. A weighted cost minimization problem of average energy consumption, average computation resource price, and task overflow rate is formulated under the latency constraint. Due to its mixed-integer nonlinearity, the problem is decomposed into three subproblems. An alternating iteration resource allocation algorithm (AIRAA) is designed to jointly optimize communication and computation resource allocation, while a heuristic algorithm is proposed for task offloading decision optimization. Simulation results demonstrate that the proposed algorithm exhibits good convergence and achieves weighted cost reductions of 45.6%, 18.4%, 33.7%, and 37.5% compared with four baseline schemes when the user-satellite link bandwidth is set to 150 MHz.

Queue management and resource allocation play a critical role in enabling cooperative status awareness in vehicular networks. This paper investigates the problem of age of information (AoI)-aware status updates in vehicle-to-vehicle (V2V) communication, where each vehicle's status is represented by multiple interdependent packets. To enable fine-grained queue management at the packet level under resource constraints, we formulate a joint optimization problem that simultaneously learns active packet dropping and transmit power control strategies. A hybrid action space is designed to support both discrete dropping decisions and continuous power control. To exploit the graph-structured interference inherent in V2V topology, a graph neural network (GNN) is introduced to aggregate slowly varying large-scale fading, allowing agents to capture topological dependencies implicitly without frequent message exchange. The overall framework is built upon multi-agent proximal policy optimization (MAPPO), with centralized training and decentralized execution (CTDE). Simulations demonstrate that the proposed method significantly reduces average AoI across a wide range of network densities, channel conditions, and traffic loads, consistently outperforming several baselines.

Multiple-input multiple-output (MIMO) systems are essential for improving capacity and reliability in semantic communications. Existing methods mainly design the channel-aware neural networks but neglect the underlying signal distribution. In this paper, we develop a denoising diffusion null-space model-based module over MIMO channels (DDNM-MIMO), which is a plug-in module deployed at the receiver. By modeling the MIMO channel, precoding, and equalization as a linear transformation with additive noise, we design corresponding linear and scaling matrices to construct a sampling process for denoising the received signal. The DDNM-MIMO integrates channel state information (CSI) embedding, supporting both closed-loop MIMO with CSI at the transmitter and open-loop MIMO with CSI at the receiver, thereby improving channel adaptability across various noise levels. As a plug-in, the DDNM-MIMO module operates independently of the joint source-channel coding (JSCC) coder structure, offering flexible integration into diverse systems. Experimental results show that DDNM-MIMO effectively reduces the mean square errors (MSE) between the encoded and equalized signals. Consequently, the proposed DDNM-MIMO semantic communication system achieves superior image reconstruction performance compared to existing JSCC-based semantic communication method.

In recent years, with the advancement of computational hardware performance, machine learning algorithms have achieved significant development and widespread application across various fields, and have become deeply embedded in smart grids and communication systems. However, it is important to note that despite the widespread deployment of smart meters in the power system, the lack of reliable intelligent diagnostic, a large number of such electricity meters experiencing communication failures caused by internal topological defects every year. To address this issue, we propose a machine learning-based monitoring and early warning model using multidimensional feature fusion. By integrating more than twenty key features in four categories, including attribute features, operational load features, communication behavior features, and derived combined features—an XGBoost classification algorithm framework is constructed to implement risk early warning for electricity meter communication faults. Validated with data from millions of users, the proposed model achieves an accuracy of approximately 90%, the annual average reduction in power outages caused by communication faults is more than 10,000 hours, and significantly enhances the grid’s safety and operational stability.

Causal inference refers to the discovery and acquisition of causal relationships from observational data, which is an important way for Internet of Things (IoT) systems to realize from perception to cognition. However, most existing causal inference methods assume that the input data is structured. For irregular time series with random missing items, the absence of important time points often leads to serious degradation of causal inference performance, which hinders practical applications. To this end, a GNN-based time-frequency cooperated causal inference (TFCI-GNN) method for irregular time series is proposed. The temporal features are initially extracted using a temporal encoder, and the frequency domain encoder adaptively models the frequency interdependence between channels using discrete cosine transform (DCT), forming attention coefficients that act on the temporal features. By utilizing the estimated adjacency matrix, feature aggregation is performed using a GNN. Finally, the causal graph is decoded and updated using a unique self-supervised approach, which mutually promotes the process of interpolation and causal inference. Experimental results on synthetic and real datasets show that the proposed TFCI-GNN method outperforms the baseline algorithms in inference performance.

To tackle the formidable challenges from future AI applications in communication and computing power, an urgent need arises for a scalable distributed learning framework, which should adeptly orchestrate the varied resources of widely dispersed nodes within 6G networks, offering flexible connectivity and computing services. For distributed AI training in 6G computing-power network (CPN), this paper proposes an adaptive split federated learning (SFL) framework. Considering the terminal computing power heterogeneity, it introduces three train modes: local-only, single base station (BS) collaboration, and dual BSs collaboration. Given the varying channel qualities, two transmission modes are used in the model aggregation phase: direct uploading to the BS and uploading via Device-to-Device (D2D) relays. Accordingly, we formulate a joint optimization problem to minimize overall task latency, which involves model splitting method, cooperative node selection, and multi-domain resource allocation, and then decompose it into two subproblems. First, a shortest path search algorithm is devised to solve the optimal model splitting method and cooperative node selection. Second, convex optimization is employed to derive the optimal multi-domain resource allocation. Simulation results show that the proposed framework attains lower total training latency while preserving high model accuracy.