Uplink integrated sensing and communication (ISAC) systems enable base stations to jointly estimate radar target parameters and detect communication signals from multiple users using shared time-frequency resources. However, the mutual interference between radar echoes and communication signals poses significant challenges for signal processing, particularly under practical conditions such as comparable power levels and uncertain noise variance. To address these challenges, we propose two variational Bayesian message passing receivers based on mean-field and Bethe approximations. The mean-field method derives a closed-form expression for the variational free energy and incorporates a noise variance learning mechanism to improve convergence. The Bethe-based method further refines the posterior approximation by analyzing the stationary points of the Bethe free energy and includes a fixed-point update for noise variance. We also develop a state evolution (SE) analysis to characterize the convergence behavior and estimation accuracy of the Bethe method. Simulation results demonstrate that both methods outperform baseline schemes in terms of mean squared error (MSE) and bit error rate (BER), while maintaining similar computational complexity.

The high dimensionality of antenna arrays in extremely large-scale multiple-input-multiple-output (XL-MIMO) systems leads to prohibitive computational complexity in uplink linear detection. To exploit the intrinsic low-rank structure of near-field channels and enable low-complexity linear detection, the global lowrank matrix fitting (LMaFit) approach with a rank-increasing strategy is employed to facilitate the matrix computation, where the rank adaptation leverages the underlying propagation structure. Building upon it, a block-wise LMaFit scheme is further proposed to utilize localized low-rank structures within subarrays, thereby enabling substantial complexity reduction in linear detection. Finally, the convergence behaviors of both the global and block-wise LMaFit algorithms are analyzed.

State-of-the-art schemes for the performance analysis and optimization of multiple-input multiple-output (MIMO) communications generally suffer from degradation or even become ineffective in highly dynamic and complex environments with unknown interference and uncertain channel state information (CSI). To address these challenges and enhance network self-optimization, we propose a learnable model-driven regularized zero-forcing precoding scheme, and design a light-weight neural network for refined prediction of sum rate and detection error, by leveraging coarse model-driven approximations. Then, we estimate the CSI uncertainty based on the learned predictor in an iterative manner and, in turn, optimize both the transmit regularization term and subsequent receive power scaling factors. To achieve a favorable trade-off between convergence speed and robustness, we further propose a deep-unfolded projected gradient descent algorithm for power scaling.

The increasing demand for ultra-reliable and low-latency communications (URLLC) has motivated research on transmission strategies under delay constraints. The access delay, i.e., the time from a codeword becoming the head-of-line to being successfully transmitted, constitutes a major portion of the overall transmission latency. This paper investigates a multiuser multiple-input single-output (MU-MISO) downlink system with mean access delay. We first derive analytical expressions for the system information output rate (IO-Rate) as explicit functions of the mean access delay threshold and per-user average power constraint, under heterogeneous large-scale path loss. The analysis considers two representative power allocation schemes: truncated channel inversion and waterfilling. To gain further insight, we then analyze a special case with identical path loss among users, which reveals the fundamental rate-delay trade-off and its scaling behavior with respect to the degrees of freedom of the effective channel and transmit power. Furthermore, an optimal average power allocation strategy is formulated to maximize the IORate of the system under a total average power constraint.Simulations validate the theoretical analysis, showing that: 1) The IO-Rate saturates under loose delay constraints and increases linearly with transmit power (in dB) in the high signal-to-noise ratio regime. 2) When the number of base station antennas scales proportionally with the number of users, the IO-Rate of the system scales linearly with user count. 3) The proposed delay-aware power allocation achieves 25% higher IO-Rate than equal power allocation and outperforms the ergodic capacity maximizing allocation under stringent delay constraints while the gap increases as the constraint tightens.

Expectation propagation with successive updating (EP-SU) offers a symbol-wise, layering-like update, enhancing the detection performance for massive MIMO. Nevertheless, this layered potential has not yet been exploited in iterative detection and decoding (IDD). To address this gap, we propose a jointly-layered IDD algorithm for LDPC-coded massive MIMO systems, termed EP-SU-JL. By restructuring EP-SU in a layered manner and embedding it within a jointly-layered IDD framework, the proposed approach facilitates more frequent and efficient iterative information exchange between the detector and decoder, accelerating convergence and enhancing performance. Additionally, a sorting and grouping scheme is introduced to optimize the symbol update process. Numerical simulations demonstrate that the proposed EP-SU-JL consistently outperforms state-of-the-art (SOA) EP-based IDD algorithms across various MIMO scenarios while maintaining comparable complexity.

The upcoming 6G necessitates more stringent performance requirements than those in 5G. Polar-coded MIMO systems, benefiting from polar codes' capacity-achieving properties, offer high-reliability potential. To fulfill these stringent reliability requirements, tree search receivers demonstrated satisfactory error rates. However, this enhanced performance raises complexity, hindering latency and energy efficiency goals. Identifying metric calculation and path sorting as the main contributors to complexity, this paper aims to reduce it without impacting performance. The former can be reduced by fine-tuning the list sizes, and the latter by omitting sorting in specific layers - a new insight. This paper delineates the relationship between parameters and complexities, and then formulates these relationships into integer programming (IP), highlighting their intrinsic link. To address these problems, we present a multi-order greedy method-based solver and propose a multi-order greedy method-aided tree search (MOGATE) receiver. Using multi-order greedy techniques reduces the complexity of solving problems by 36% compared to genetic algorithms. The effectiveness of the MOGATE has been proven in both separated detection and decoding (SDD) and joint detection and decoding (JDD) scenarios. Compared to its genetic counterparts, MOGATE achieves a notable reduction in complexity - up to 33% less in metric calculation and 10% less in path sorting - without compromising on error rate.

Extremely large-scale MIMO (XL-MIMO) is regarded as a pivotal enabler for achieving ultra-high spectral efficiency in 6G communications. Near-field channel models, which integrate both line-of-sight (LoS) and non-line-of-sight (NLoS) components, provide accurate characterizations of near-field XL-MIMO channels. However, existing channel estimation schemes encounter severe performance bottlenecks due to the high-dimensional nature of near-field XL-MIMO channels and their structured angular sparsity compared to far-field MIMO systems. To address these challenges, we propose a variance exploding stochastic differential equation (VE-SDE) generator based on an improved diffusion transformer (IDiT) network. The VE-SDE progressively maps the complex XL-MIMO channel distribution to a tractable prior distribution by gradually injecting noise. We utilize the patchify technique to decompose the perturbed angular domain channels into token sequences, which are then processed with diffusion transformer (DiT) blocks, substantially reducing floating-point operations (FLOPs). Additionally, a sparse self-attention mechanism is employed to enhance structured sparsity characterization learning, thereby improving estimation accuracy. Theoretical analysis and numerical experiments show that the VE-SDE generator exhibits strong generalizability and robustness across diverse channel distributions without requiring retraining. Simulation results reveal that the proposed method outperforms state-of-the-art estimation approaches, achieving high-fidelity channel estimation with only 20% pilot density.

As a paradigm shift towards pervasive intelligence, semantic communication (SemCom) has shown great potentials to improve communication efficiency and provide user-centric services by delivering task-oriented semantic meanings. However, the exponential growth in connected devices, data volumes, and communication demands presents significant challenges for practical SemCom design, particularly in resource-constrained wireless networks. In this work, we propose a task-agnostic semantic communication (TASC) framework capable of supporting multimodal data across diverse tasks. To investigate the interplay between communication and intelligent tasks from an information-theoretic perspective, we introduce a distributed multimodal information bottleneck (DMIB) principle, which enables the extraction of minimal sufficient unimodal and multimodal representations by eliminating redundant information while preserving task-relevant semantics. To further reduce the communication overhead, we develop an adaptive semantic feature transmission method under dynamic channel conditions. Then, TASC is trained based on federated meta-learning (FML) to learn a well-initialized model for rapid adaptation and generalization. To gain deep insights, we conduct theoretical analysis and devise resource management to accelerate convergence while minimizing the training latency and energy cost. Moreover, we develop a joint user selection and resource allocation algorithm to address the non-convex problem with theoretical guarantees. Extensive simulation results validate the effectiveness and superiority of the proposed TASC compared to baselines.

In the sixth-generation (6G) mobile communication networks, low earth orbit (LEO) satellite system is the key enabler for global coverage and ubiquitous wireless access. With the dense deployment of LEO satellite constellations, a large number of satellite beams are expected to be collaboratively scheduled to improve the system throughput and guarantee the user requirements. However, the scarce spectrum resources, the unevenly distributed traffic demands and the highly dynamic topology put forward higher requests to the beam scheduling, interference avoidance and mobility management. To overcome these technical challenges, various solutions for intelligent beam management have been investigated in both industry and academia. In this paper, we provide a comprehensive review on the key technologies for beam management in the LEO satellite systems, including the spectrum sharing schemes with interference avoidance, the beam-hopping-based scheduling methods and the beam handover mechanisms. Moreover, we identify the future trends and the open issues for further exploration.

The communication performance of unmanned aerial vehicles (UAVs) in millimeter-wave (mmWave) cell-free massive multiple-input multiple-output systems (CF-mMIMO) is severely limited by beam misalignment caused by high mobility. Conventional approaches rely either on error-prone navigation information (NaInfo) or on beam sweeping with prohibitive overhead, making accurate beam tracking difficult to achieve. In this letter, we propose a dual-fingerprint-based intelligent beam tracking framework that integrates model-driven Kalman-filter-based NaInfo state prediction with data-driven real-time wide-beam sensing. By incorporating a gating network and an attention-based fusion neural network, the proposed approach adaptively balances NaInfo-driven predictive beam steering and wide-beam-based real-time correction. Simulation results demonstrate that, for a UAV equipped with 64 antennas, the proposed scheme achieves an approximately eightfold reduction in training overhead compared with narrow-beam training, at the cost of only a 7.5% degradation in the achievable rate. In addition, compared with full wide-beam training, the incorporation of the gating network reduces the training overhead by 75%, while incurring only a 2.8% degradation in performance. Furthermore, the proposed scheme consistently outperforms benchmark schemes under large attitude errors and low-signal-to-noise ratio (SNR) conditions, demonstrating its robustness in high-mobility UAV communications.

Network slicing has emerged as a promising solution for end-to-end (E2E) resource management and orchestration, enabled by software-defined networking (SDN) and network function virtualization (NFV) technologies. In this paper, we investigate the dynamic E2E optical-wireless network slicing mapping problem in converged optical-wireless access networks. To address user data rate requirements in wireless networks and radio access network (RAN) slicing scheduling in optical networks, we first formulate an E2E optical-wireless network slicing mapping model with its associated constraints. Subsequently, to provide feasible solutions for real-world applications, we propose a dynamic E2E optical-wireless network slicing mapping (D-E2E-OW-NSM) algorithm based on deep reinforcement learning (DRL). To facilitate the decision-making process of the DRL agent, we decompose the intricate E2E optical-wireless network slicing request into several sub-requests, solving them one by one in turn. Simulation results demonstrate that our proposed method reduces the request blocking probability by up to 18.2% in a small-scale network and 11.3% in a large-scale network compared to baseline methods. Our analyses provide valuable insights into the modeling and design of efficient converged optical-wireless access networks for 5G and beyond.

Feedback-free user-specific channel acquisition is essential for wireless applications such as network digital twins (DTs) and channel prediction, particularly for positions of interest (POIs). However, the increasing dimensionality of wireless channels poses significant challenges to efficient channel acquisition. This paper proposes a user-specific channel generative model, termed Position-to-Channel Generative Network (P2CGNet), for multi-input multi-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. By integrating classifier-free guided diffusion models with transformer architectures, P2CGNet generates high-fidelity channels conditioned on user positions and frequency offsets. To improve computational efficiency while preserving generation fidelity, a patch-based subchannel decomposition strategy and an implicit denoising mechanism are further developed. Theoretical analysis demonstrates that P2CGNet can learn the underlying channel distribution and accelerate inference. As a practical validation, P2CGNet is integrated into a channel DT for joint spatial-temporal-frequency resource optimization in multi-user MIMO-OFDM systems with sparse probing beam measurements. To this end, an intelligent Resource Block assignment and Beam Alignment (R2BA) model is proposed, which leverages the channel DT and prior communication model to acquire intent-oriented measurements and employs a hierarchical graph neural network to extract effective representations. Numerical results show that P2CGNet overall outperforms vanilla diffusion models and generative adversarial network-based methods in channel generation quality, data efficiency, antenna-size scalability, and computational overhead. Moreover, P2CGNet exhibits better staleness robustness and substantially improves the spectrum efficiency of the downstream task by generating channels for given POIs.

Satisfying stringent delay requirements while sustaining acceptable throughput remains a critical challenge for wireless communication systems. To address this issue, this paper proposes a generative learning–enabled delay–bounded scheduling (GLDBS) framework that integrates a conditional variational autoencoder–generalized pareto distribution (CVAE–GPD) model with parameterized resource scheduling. The CVAE–GPD component captures both the main and tail behaviors of packet delay distributions, enabling precise characterization of stochastic delay dynamics. The predicted statistical parameters are subsequently incorporated into a reinforcement learning–based scheduler that adaptively adjusts interpretable control parameters for proactive and risk–aware resource allocation. Simulation results demonstrate that the proposed GLDBS framework effectively reduces delay violation probability while maintaining system throughput, offering a promising solution for wireless services.

Joint sensing and communication channel estimation (JSCCE) is a critical problem in integrated sensing and communication (ISAC) systems, where two types of channels inherently share similar physical propagation mechanisms while exhibit distinct task-specific characteristics. Leveraging this property, this work proposes a multi-objective unified framework that combines the feature extraction capacity of deep learning with the principled exploitation of ISAC channel commonalities and differences. Specifically, we design a novel dual-task unified estimation Transformer network (DUET-Net) comprising a shared encoder backbone for capturing correlated sparse channel features, alongside dual-head decoders with task-specific branches for accommodating the individual statistical characteristics and structural properties of the two channels. To further address the limited angular resolution in communication channels, we incorporate an angle-aware attention (AAA) module that selectively emphasizes informative angular components. Moreover, an uncertainty-weighted multi-task loss is introduced to adaptively balance the training process of both tasks in accordance to their different learning dynamics, thereby improving convergence stability. Extensive numerical simulations demonstrate that the proposed DUET-Net consistently outperforms existing methods over a wide range of transmit power levels and across various scenarios, with an efficient performance-complexity trade-off. The effectiveness of the designed modules is further confirmed by ablation studies.

In millimeter-wave (mmWave) multi-subarray systems, the beamforming space is limited by hardware constraints, and the anti-jamming performance degrades due to direction-of-arrival (DoA) estimation errors. To tackle these challenges, this paper proposes two robust hybrid beamforming (HBF) approaches, namely interior-point (IP)-based HBF and deep unfolding network (DUN)-based HBF. The IP-based HBF leverages Riemannian manifold optimization to design the analog beamforming matrix under hardware constraints, and employs the interior-point algorithm to maximize the signal-to-interference-plus-noise ratio (SINR) under strict jamming suppression constraints. The DUN-based HBF unfolds the gradient ascent algorithm into a deep network, significantly reducing computational complexity with slight performance degradation. Simulation results demonstrate that the IP-based HBF achieves superior anti-jamming performance, while the DUN-based HBF achieves a favorable trade-off between anti-jamming performance and computational complexity. Both proposed methods outperform the baseline schemes, providing flexible and effective anti-jamming solutions for practical mmWave multi-subarray systems.

The deployment of extremely large-scale antenna array (ELAA) in sixth-generation (6G) communication systems introduces unique challenges for efficient near-field channel estimation. To tackle these issues, this paper presents a theory-guided approach that incorporates angular information into an attention-based estimation framework. A piecewise Fourier representation is proposed to implicitly encode the near-field channel's inherent nonlinearity, enabling the entire channel to be segmented into multiple subchannels, each mapped to the angular domain via the discrete Fourier transform (DFT). Then, we develop a joint subchannel-spatial-attention network (JSSAnet) to extract the spatial features of both intra- and inter-subchannels. To guide theoretically the design of the joint attention mechanism, we derive upper and lower bounds based on approximation criteria and DFT quantization loss mitigation, respectively. Following by both bounds, a JSSA layer of an attention block is constructed to assign independent and adaptive spatial attention weights to each subchannel in parallel. Subsequently, a feed-forward network (FFN) of an attention block further captures and refines the residual nonlinear dependencies across subchannels. Moreover, the proposed JSSA map is linearly computed via element-wise product combining large-kernel convolutions (DLKC), maintaining strong contextual learning capability. Numerical results verify the effectiveness of embedding sparsity information into the attention network and demonstrate JSSAnet achieves superior estimation performance compared with existing methods.

This paper investigates an intelligent reflecting surface (IRS)-aided wireless-powered communication network (WPCN) for maximizing the weighted sum rate (WSR). To reduce the complexity of traditional model-driven algorithms and improve convergence in data-driven deep learning approaches, a novel hybrid block coordinate descent (BCD) algorithm motivated by the dilation extraction and context attention (DECA) neural network (NN) is proposed. Specifically, the WSR maximization problem is firstly reformulated as a more tractable form, enabling the BCD algorithm to efficiently optimize the decoupled variables within the constraints. Meanwhile, at each BCD iteration, the DECA NN accelerates IRS phase shift optimization by facilitating the majorization-minimization (MM) algorithm to solve the computationally intensive fractional programming problem. Moreover, by leveraging dilation convolution and high-speed attention mechanisms, the DECA NN significantly outperforms existing deep learning benchmarks in both precision and convergence speed. Numerical results show that the proposed hybrid framework delivers performance comparable to the traditional BCD algorithm with dramatically reduced time consumption, while consistently maintaining robust performance under imperfect CSI and exhibiting strong transferability across diverse communication scenarios.

Integrated sensing and communication (ISAC) is a key technology for future networks, where predictive beamforming is crucial for high-mobility targets. To overcome the limitations of open-loop predictive schemes, where prediction errors are propagated without a real-time correction mechanism, this letter proposes a closed-loop predictive beamforming framework. The core of this framework is a weighted sum maximization problem, where the feedback mechanism is embedded by linking the radar estimation rate to the extended Kalman filter (EKF)’s measurement noise. To solve this problem, we design an adaptive dual-expert network (ADE-Net) that employs separate GRUs and a weighting factor to steer feature fusion. Simulations demonstrate that our scheme achieves significant performance advantages and facilitates a more intelligent trade-off between sensing and communication.

Direct handset-to-satellite (DHTS) systems represent a key enabler of nonterrestrial networks (NTNs), allowing mass-market handheld devices to directly access low-Earth-orbit (LEO) satellites. This article provides a tutorial perspective on the architecture-aware roles and system-level challenges of signal detection in DHTS systems. We show that regenerative payload configuration, multisatellite cooperation, and control plane (CP)/user plane (UP) separation fundamentally reshape where and how detection is performed. Meanwhile, severe Doppler shifts, asynchronous multilink superposition and onboard processing constraints invalidate many terrestrial assumptions, motivating waveform-aware, structure-exploiting, and cooperative detection strategies. These challenges also drive the adoption of soft-output detection schemes to suppress error propagation in iterative decoding. Such emerging detection requirements motivate tighter integration with access control mechanisms and the development of onboard intelligence. Overall, the evolving role of detection highlights the need for scalable, reliable, and resource-efficient solutions in future DHTS deployments.

Against the backdrop of the commercialization of 5G-Advanced (5G-A) and the conceptualization of 6G, the Mobile Information Network (MINet)-a futuristic network paradigm centered on extreme connectivity and multi-dimensional integration-has emerged to transcend the limitations of traditional mobile networks, which focus solely on isolated communication services. MINet is designed to support the full lifecycle of information services, integrating communication with advanced computing, artificial intelligence (AI), and big data to achieve ubiquitous extreme connectivity and deliver integrated information services encompassing different elements. At its core, MINet's competitive edge lies in the "One Extreme and Three Integrations" framework: this framework takes "ubiquitous extreme connectivity" as the foundational performance pillar and relies on three functional directions: "communication-sensing-computing-intelligence convergence," "space-terrestrial integration," and "digital-physical integration". To operationalize this framework, DOICT (Data, Operation, Intelligence, Computing, and Telecommunication) convergence serves as a critical enabler: it underpins the design of MINet's scalable core networks and edge intelligence systems, while supporting key technological innovations. This paper reviews the foundational theories, network architectures, key technologies, and devices driving the development of MINet, and also highlights critical challenges in MINet's development, including design considerations, application requirements, sustainable and low-carbon. Ultimately, as MINet evolves from 5G-A to 6G, the synergy of its core paradigm (MINet), technical framework ("One Extreme and Three Integrations"), and enabling mechanism (DOICT convergence) will bridge the physical and digital worlds, fostering transformative applications in immersive experiences, industrial automation, and smart society.