Money laundering is a financial crime that obscures the origin of illicit funds, necessitating the development and enforcement of anti-money laundering (AML) policies by governments and organizations. The proliferation of mobile payment platforms and smart IoT devices has significantly complicated AML investigations. As payment networks become more interconnected, there is an increasing need for efficient real-time detection to process large volumes of transaction data on heterogeneous payment systems by different operators such as digital currencies, cryptocurrencies and account-based payments. Most of these mobile payment networks are supported by connected devices, many of which are considered loT devices in the FinTech space that constantly generate data. Furthermore, the growing complexity and unpredictability of transaction patterns across these networks contribute to a higher incidence of false positives. While machine learning solutions have the potential to enhance detection efficiency, their application in AML faces unique challenges, such as addressing privacy concerns tied to sensitive financial data and managing the real-world constraint of limited data availability due to data regulations. Existing surveys in the AML literature broadly review machine learning approaches for money laundering detection, but they often lack an in-depth exploration of advanced deep learning techniques—an emerging field with significant potential. To address this gap, this paper conducts a comprehensive review of deep learning solutions and the challenges associated with their use in AML. Additionally, we propose a novel framework that applies the least-privilege principle by integrating machine learning techniques, codifying AML red flags, and employing account profiling to provide context for predictions and enable effective fraud detection under limited data availability. Specifically, our approach defines AML-relevant financial profile characteristics and risk indicators to contextualize transactions and assess their associated risks. The proposed context-risk-predict AML (CRP-AML) model demonstrates notable success, achieving an F1 score of 82.51% on the minority class and nearly doubling the performance of other pattern detection models when the proportion of money laundering records in the dataset drops as low as 0.0005.

Resource management is one of the challenges in satellite networks due to their high mobility, wide coverage, long propagation distances, and stringent constraints on energy, communication, and computation resources. Traditional resource allocation approaches rely only on hard and rigid system performance metrics. Meanwhile, incentive mechanisms, which are based on game theory and auction theory, investigate systems from the "economic" perspective in addition to the "system" perspective. Particularly, incentive mechanisms are able to take into account rationality and other behavior of human users into account, which guarantees benefits/utility of all system entities, thereby improving the scalability, adaptability, and fairness in resource allocation. This paper presents a comprehensive survey of incentive mechanism design for resource management in satellite networks. The paper covers key issues in the satellite networks, such as communication resource allocation, computation offloading, privacy and security, and coordination. We conclude with future research directions including learning-based mechanism design for satellite networks.

The energy optimization and demand side management (DSM) of Internet of Things (IoT)-enabled microgrids are being transformed by generative artificial intelligence, such as large language models (LLMs). This paper explores an integration of LLMs into energy management, and emphasizes their roles in automating the optimization of DSM strategies with Internet of Electric Vehicles (IoEV) as a representative example of the Internet of Vehicles (IoV). We investigate challenges and solutions associated with DSM and explore new opportunities presented by leveraging LLMs. Then, we propose an innovative solution that enhances LLMs with retrieval-augmented generation for automatic problem formulation, code generation, and customizing optimization. The results demonstrate the effectiveness of our proposed solution in charging scheduling and optimization for electric vehicles, and highlight our solution’s significant advancements in energy efficiency and user adaptability. This work shows LLMs’ potential in energy optimization of the IoT-enabled microgrids and promotes intelligent DSM solutions.

The recent advancements in large language models (LLMs) have significantly enhanced language understanding and content generation capabilities. However, the deployment of LLMs on resource-constrained Internet of Things (IoT) devices remains challenging due to their substantial computational and storage requirements. To address this issue, we propose a novel LLM pruning method, termed structurally-aware adaptive pruning (SAAP), to reduce computational and storage costs for LLMs while maintaining model performance. Specifically, SAAP first leverages maximum likelihood estimation to calibrate traditional structural importance metrics for LLM pruning. Next, it employs a Bayesian fusion approach to address the predictive uncertainty in multi-granularity metrics, enabling accurate assessments of structural importance for LLMs. Then, SAAP introduces a cross-layer importance alignment mechanism based on quantile mapping, which normalizes layer-wise importance scores to ensure consistent pruning from a global perspective. Furthermore, SAAP develops an efficient block-wise fine-tuning strategy for enhancing the performance of the LLM after pruning. To validate the effectiveness of SAAP, we conduct extensive experiments on nine open-source LLMs across two representative tasks—language modeling and zero-shot classification. Experimental results show that SAAP consistently outperforms several baseline methods, achieving accuracy improvements of 2.5%, 2.63%, and 2.44% on LLaMA-7B, Vicuna-7B, and LLaMA-13B when the pruning ratio is 50%. Finally, SAAP is implemented on a testbed—NVIDIA Jetson AGX Orin 32GB Developer Kit. Test results demonstrate that compared to the foundation LLM, SAAP enhances the inference speed by 86.86% at a pruning ratio of 50%, highlighting its potential for practical deployment on resource-constrained IoT devices.

Despite significant advancements in terrestrial networks, inherent limitations persist in providing reliable coverage to remote areas and maintaining resilience during natural disasters. Multi-tier networks with low Earth orbit (LEO) satellites and high-altitude platforms (HAPs) offer promising solutions, but face challenges from high mobility and dynamic channel conditions that cause unstable connections and frequent handovers. In this paper, we design a three-tier network architecture that integrates LEO satellites, HAPs, and ground terminals with hybrid free-space optical (FSO) and radio frequency (RF) links to maximize coverage while maintaining connectivity reliability. This hybrid approach leverages the high bandwidth of FSO for satellite-to-HAP links and the weather resilience of RF for HAP-to-ground links. We formulate a joint optimization problem to simultaneously balance downlink transmission rate and handover frequency by optimizing network configuration and satellite handover decisions. The problem is highly dynamic and non-convex with time-coupled constraints. To address these challenges, we propose a novel large language model (LLM)-guided truncated quantile critics algorithm with dynamic action masking (LTQC-DAM) that utilizes dynamic action masking to eliminate unnecessary exploration and employs LLMs to adaptively tune hyperparameters. Simulation results demonstrate that the proposed LTQC-DAM algorithm outperforms baseline algorithms in terms of convergence, downlink transmission rate, and handover frequency. We also reveal that compared to other state-of-the-art LLMs, DeepSeek delivers the best performance through gradual, contextually-aware parameter adjustments.

Recent advances in large language models (LLMs) have driven remarkable progress across diverse NLP benchmarks. However, the application of these models to sophisticated domains such as wireless communications raises new challenges. This paper addresses the evaluation gap for LLMs in telecommunications by introducing Last Dance of Telecommunications (LDOT) – a comprehensive benchmark suite for telecom-related tasks. LDOT encompasses a broad range of problem categories, including conceptual telecom questions, mathematical and logical reasoning problems, and complex network optimization scenarios, specifically designed to challenging LLMs’ high-level reasoning and domain-specific knowledge. Using LDOT, we rigorously assess state-of-the-art (SOTA) LLMs (both closed-source and open-source) on domain-specific tasks. Our results reveal that while general-purpose LLMs exhibit strong performance on basic telecom knowledge questions, they struggle with reasoning-intensive wireless problems. Notably, certain multi-step optimization and planning tasks in LDOT remain unsolved by even the best models, exposing performance gaps that are not apparent from existing saturated benchmarks. We provide a detailed failure analysis to pinpoint whether these limitations arise from insufficient telecom-specific knowledge or from inadequate reasoning capabilities. The LDOT dataset and our evaluation findings aim to facilitate the development of more robust domain-adapted LLMs for next-generation wireless communications.

Nowadays, Generative AI (GenAI) reshapes numerous domains by enabling machines to create content across modalities. As GenAI evolves into autonomous agents capable of reasoning, collaboration, and interaction, they are increasingly deployed on network infrastructures to serve humans automatically. This emerging paradigm, known as the agentic network, presents new optimization challenges due to the demand to incorporate subjective intents of human users expressed in natural language. Traditional generic Deep Reinforcement Learning (DRL) struggles to capture intent semantics and adjust policies dynamically, thus leading to suboptimality. In this paper, we present LAMeTA, a Large AI Model (LAM)-empowered Two-stage Approach for intent-aware agentic network optimization. First, we propose Intent-oriented Knowledge Distillation (IoKD), which efficiently distills intent-understanding capabilities from resource-intensive LAMs to lightweight edge LAMs (E-LAMs) to serve end users. Second, we develop Symbiotic Reinforcement Learning (SRL), integrating E-LAMs with a policy-based DRL framework. In SRL, E-LAMs translate natural language user intents into structured preference vectors that guide both state representation and reward design. The DRL, in turn, optimizes the generative service function chain composition and E-LAM selection based on real-time network conditions, thus optimizing the subjective Quality-of-Experience (QoE). Extensive experiments conducted in an agentic network with 81 agents demonstrate that IoKD reduces mean squared error in intent prediction by up to 22.5%, while SRL outperforms conventional generic DRL by up to 23.5% in maximizing intent-aware QoE.

This paper investigates covert prompt transmission for secure and efficient large language model (LLM) services over wireless networks. We formulate a latency minimization problem under fidelity and detectability constraints to ensure confidential and covert communication by jointly optimizing the transmit power and prompt compression ratio. To solve this problem, we first propose a prompt compression and encryption (PCAE) framework, performing surprisal-guided compression followed by lightweight permutation-based encryption. Specifically, PCAE employs a locally deployed small language model (SLM) to estimate token-level surprisal scores, selectively retaining semantically critical tokens while discarding redundant ones. This significantly reduces computational overhead and transmission duration. To further enhance covert wireless transmission, we then develop a group-based proximal policy optimization (GPPO) method that samples multiple candidate actions for each state, selecting the optimal one within each group and incorporating a Kullback-Leibler (KL) divergence penalty to improve policy stability and exploration. Simulation results show that PCAE achieves comparable LLM response fidelity to baseline methods while reducing preprocessing latency by over five orders of magnitude, enabling real-time edge deployment. We further validate PCAE effectiveness across diverse LLM backbones, including DeepSeek-32B, Qwen-32B, and their smaller variants. Moreover, GPPO reduces covert transmission latency by up to 38.6% compared to existing reinforcement learning strategies, with further analysis showing that increased transmit power provides additional latency benefits.

Deep joint source-channel coding (JSCC) has emerged as a promising paradigm for semantic communication, delivering significant performance gains over conventional separate coding schemes. However, existing JSCC frameworks remain vulnerable to physical-layer adversarial threats, such as pilot spoofing and subcarrier jamming, compromising semantic fidelity. In this paper, we propose SecDiff, a plug-and-play, diffusion-aided decoding framework that significantly enhances the security and robustness of deep JSCC under adversarial wireless environments. Different from prior diffusion-guided JSCC methods that suffer from high inference latency, SecDiff employs pseudoinverse-guided sampling and adaptive guidance weighting, enabling flexible step-size control and efficient semantic reconstruction. To counter jamming attacks, we introduce a power-based subcarrier masking strategy and recast recovery as a masked inpainting problem, solved via diffusion guidance. For pilot spoofing, we formulate channel estimation as a blind inverse problem and develop an expectation-minimization (EM)-driven reconstruction algorithm, guided jointly by reconstruction loss and a channel operator. Notably, our method alternates between pilot recovery and channel estimation, enabling joint refinement of both variables throughout the diffusion process. Extensive experiments over orthogonal frequency-division multiplexing (OFDM) channels under adversarial conditions show that SecDiff outperforms existing secure and generative JSCC baselines by achieving a favorable trade-off between reconstruction quality and computational cost. This balance makes SecDiff a promising step toward practical, low-latency, and attack-resilient semantic communications.

Public Blockchain-based covert transmission (CT) can address the limitations of traditional CT methods. Monero is a blockchain-based cryptocurrency with strong privacy protection techniques. However, existing Monero-based CT methods are limited to unicast scenarios. If applied directly to group CT scenarios, they would lead to a significant increase in transaction volume as the number of receivers increases. Meanwhile, existing Bitcoin/Ethereum-based group CT methods at least face three challenges, including susceptibility to key and identity inference attacks, information leakage during off-chain negotiations, and exposure of communication channels.This paper proposes a Monero-Based Group CT approach (MBGCT), which enables on-chain group key (used for receivers to filter covert transactions and extract messages) issuance, fine-grained access control of messages, and covert transaction identification and decryption isolation. MBGCT can ensure confidentiality of group keys, unforgeability of messages, integrity of each transmitted message, obscurity of covert channels, isolation of key generation from key management, and enhanced anonymity. As a result, MBGCT can not only prevent information leakage and channel exposure, but also resist the attacks of entity impersonation, data tampering, key and identity inference. We implemented MBGCT in Monero client v0.18.1.0, and validated its capability of high embedding rates, low transaction fees, and high execution efficiency on the Monero public chain Stagenet.

The rise of latency-sensitive and bandwidth-intensive services has driven hardware accelerator adoption in cloud networks. Programmable data plane (PDP) switches achieve significant performance gains in high-performance cloud computing but face fixed pipeline constraints that limit flexibility in multi-tenant service function chaining (SFC) with dynamic function compositions. Existing approaches either exhaust resources through redundant embeddings or degrade performance via packet recirculation. This paper presents PBox, a framework enabling flexible SFC orchestration across multiple PDP switches by optimizing network function (NF) execution orders to minimize pipeline traversals-the dominant end-to-end processing delay source. This requires co-designing NF embedding with routing strategies. PBox contributes: (i) an optimized Network Service Header design supporting one-pass matching of multiple NFs through per-bit activation, reducing packet matching overhead by 45-60%; (ii) a nested optimization formulation capturing interdependence between long-term pipeline orchestration and short-term flow routing decisions; and (iii) a sampling-based genetic algorithm with statistical robustness guarantees, achieving fast switch configuration while adapting to dynamic service patterns. Evaluations on BMv2 and Intel Tofino demonstrate 33-79% completion time reduction, 46% capacity increase, and 78% line-rate efficiency, validating cloud-scale deployment potential.

The proliferation of mobile edge computing (MEC) enhances its capability to deliver low-latency services, while the geographical distribution and limited resources of edge servers simultaneously make them increasingly susceptible to distributed denial-of-service (DDoS) attacks. Existing edge DDoS mitigation strategies primarily focus on improving serviceability, that is, the number of mitigated requests, or reducing latency, yet often overlook systematic modeling of load fairness among edge servers. This omission can lead to resource imbalance, local overload, and ultimately degrade the system's long-term service performance and responsiveness to attacks. To address this issue, we propose the Fairness-aware Edge DDoS Mitigation (FairEDM) problem, which explicitly incorporates load fairness into the mitigation objective. Specifically, we first formulate FairEDM as a constrained optimization problem and prove its NP-hardness. To obtain a sub-optimal solution, we further model it as a game named FEDMGame, and design a distributed strategy formulation (FEDM-DSF) algorithm, based on multi-player best response dynamics. The resulting Nash equilibrium serves as a feasible mitigation strategy. Both theoretical analysis and experimental results validate the effectiveness and efficiency of the proposed FEDMGame framework.

In Multi-access Edge Computing (MEC) environments, Non-Orthogonal Multiple Access (NOMA) technology significantly improves spectrum efficiency and system access capacity through power-domain multiplexing. However, it also exposes edge servers to more complex resource contention and security risks. In particular, during Distributed Denial-of-Service (DDoS) attacks, conventional mitigation mechanisms often overlook users' quality of experience (QoE), which can lead to prolonged service degradation. To address this challenge, this paper investigates a QoE-guaranteed Edge DDoS Mitigation problem (QEDM) in NOMA-assisted MEC systems by jointly considering request scheduling and power allocation. First, the QEDM problem is modeled as a finite-strategy potential game, and a distributed game-theoretic algorithm, QEDMGame, is proposed. Then, through a multi-player best-response mechanism, the system converges to a Nash equilibrium, thereby achieving joint optimization of service rate and QoE. Next, experimental results on real-world base station deployment data demonstrate that QEDMGame outperforms several representative baseline methods. Additionally, theoretical analysis shows that QEDMGame exhibits strong convergence properties and performance guarantees.

Recommendation systems based on graph neural networks (GNNs) have emerged as a promising paradigm due to their ability to capture high-order interactions between users and items. However, in federated scenarios, this advantage is compromised, as each user can access only a first-order subgraph composed of its directly interacted items. To address this issue, most existing solutions introduce a trusted server to assist users in expanding their local subgraphs. However, the server in reality is often untrusted and may deviate from the protocol for its own improper benefit. Furthermore, these solutions primarily focus on the privacy of items while neglecting the privacy of potential relationships between users. To this end, we propose Garland, a GNN-based federated recommendation scheme with malicious security. Garland departs from existing work by ensuring both item and relationship privacy while supporting integrity checks to defend against malicious servers. Specifically, we employ a trending cryptographic primitive of secret-shared shuffle to expand subgraphs in a privacy-preserving and verifiable manner. We also design a pre-shuffle triple-salt encryption mechanism and a post-shuffle user-governed expansion mechanism to reduce communication costs and achieve secure distribution of neighbor information, respectively. Moreover, we develop a secret-shared aggregation mechanism to enable privacy-preserving and verifiable federated training. Theoretical analysis demonstrates the privacy and integrity of Garland. Extensive experimental evaluations on four datasets show that Garland outperforms state-of-the-art solutions.

Low-altitude economy networking (LAENet) aims to deploy various aerial vehicles to support diverse services, where data collection from edge devices via unmanned aerial vehicles (UAVs) is a critical task. The key challenge lies in jointly optimizing energy consumption and data freshness in spectrum-constrained and eavesdropping-prone low-altitude environments during the data collection process. Although deep reinforcement learning (DRL) has become a viable solution for UAV-assisted data collection, the RL agent still has limited ability to obtain and utilize informative feedback from complex low-altitude environments. In this paper, we propose a large language model (LLM)-enhanced DRL framework for secure data collection in the LAENet, where we leverage an LLM to process environmental feedback for the RL agent. Specifically, we employ the LLM as (i) a state processor to transform basic environmental observations into task-aligned representations, (ii) a reward designer to generate enriched reward signals that guide the agent's actions toward the optimization objective, and (iii) a simulator to construct a virtual LAENet environment for evaluating enhanced state-reward pairs before policy training. Theoretical analysis and numerical results demonstrate that the proposed LLM-enhanced DRL framework achieves faster convergence, improved training stability, and superior performance compared with state-of-the-art baselines.

The rapid growth of the low-altitude economy has driven the widespread adoption of unmanned aerial vehicles (UAVs). This growing deployment presents new challenges for UAV trajectory planning in complex urban environments. However, existing studies often overlook key factors, such as urban airspace constraints and economic efficiency, which are essential in low-altitude economy contexts. Deep reinforcement learning (DRL) is regarded as a promising solution to these issues, while its practical adoption remains limited by low learning efficiency. To overcome this limitation, we propose a novel UAV trajectory planning algorithm that integrates DRL with the large language model (LLM) reasoning to enable safe, compliant, and economically viable trajectory planning. Specifically, we model the trajectory planning task as a partially observable Markov decision process, explicitly incorporating obstacle avoidance, regulation awareness, and energy constraints. We design a hybrid optimization algorithm based on the soft actor-critic algorithm and LLM reasoning to enable adaptive decision-making in uncertain and dynamic environments. Experimental results demonstrate that our algorithm achieves the best overall performance, with the highest data collection rate (99.50%), almost zero collision avoidance rate and regulation violation rate, a successful landing rate of nearly 100%, and the lowest energy consumption rate (76.95%). These results validate the effectiveness of our algorithm in addressing UAV trajectory planning key challenges under constraints of the low-altitude economy networking.

This paper proposes a graph neural network (GNN)-enabled coordinated beamforming design, termed HSTGNN, for high speed railway (HSR) communication systems, where a high-speed train (HST) and low-speed users (LUs) coexist in multi-cell scenarios. Two transmission schemes with the goal of maximizing quality-of-service (QoS)-constrained sum rate and rate of the HST are formulated and then reformulated using a hybrid maximum ratio transmission and zero-forcing strategy. The HSR communication system is abstracted into a heterogeneous graph, and HSTGNN consists of complex heterogeneous graph attention layers and fully-connected layers. To meet QoS requirements and power budget constraints, we employ constraint-based penalty terms and a numerical scaling operation. HSTGNN is trained via unsupervised learning to solve the two schemes in a unified framework. Numerical results demonstrate that HSTGNN achieves millisecond-level inference speed with an average optimality gap of only 4% relative to traditional optimization algorithm across various scenarios. Moreover, HSTGNN exhibits strong scalability to unseen configurations of both cells and LUs.

Due to the massive computational demands of large generative models, AI-Generated Content (AIGC) can organize collaborative Mobile AIGC Service Providers (MASPs) at network edges to provide ubiquitous and customized content generation for resource-constrained users. However, such a paradigm faces two significant challenges: i) raw prompts (i.e., the task description from users) often lead to poor generation quality due to users' lack of experience with specific AIGC models, and ii) static service provisioning fails to efficiently utilize computational and communication resources given the heterogeneity of AIGC tasks. To address these challenges, we propose an intelligent mobile AIGC service scheme. Firstly, we develop an interactive prompt engineering mechanism that leverages a Large Language Model (LLM) to generate customized prompt corpora and employs Inverse Reinforcement Learning (IRL) for policy imitation through small-scale expert demonstrations. Secondly, we formulate a dynamic mobile AIGC service provisioning problem that jointly optimizes the number of inference trials and transmission power allocation. Then, we propose the Diffusion Enhanced Deep Deterministic Policy Gradient (D3PG) algorithm to solve the problem. By incorporating the diffusion process into Deep Reinforcement Learning (DRL) architecture, the environment exploration capability can be improved, thus adapting to varying mobile AIGC scenarios. Extensive experimental results demonstrate that our prompt engineering approach improves single-round generation success probability by 6.3×, while D3PG increases the user service experience by 50.3% compared to baseline DRL approaches.

This paper introduces a two-stage generative AI (GenAI) framework tailored for temporal spectrum cartography in low-altitude economy networks (LAENets). LAENets, characterized by diverse aerial devices such as UAVs, rely heavily on wireless communication technologies while facing challenges, including spectrum congestion and dynamic environmental interference. Traditional spectrum cartography methods have limitations in handling the temporal and spatial complexities inherent to these networks. Addressing these challenges, the proposed framework first employs a Reconstructive Masked Autoencoder (RecMAE) capable of accurately reconstructing spectrum maps from sparse and temporally varying sensor data using a novel dual-mask mechanism. This approach significantly enhances the precision of reconstructed radio frequency (RF) power maps. In the second stage, the Multi-agent Diffusion Policy (MADP) method integrates diffusion-based reinforcement learning to optimize the trajectories of dynamic UAV sensors. By leveraging temporal-attention encoding, this method effectively manages spatial exploration and exploitation to minimize cumulative reconstruction errors. Extensive numerical experiments show that this integrated GenAI framework consistently surpasses traditional interpolation and deep learning methods, especially under sparse sensing conditions. The proposed trajectory planner substantially improves spectrum map accuracy, reconstruction stability, and sensor deployment efficiency in dynamically evolving low-altitude environments.

The rapid proliferation of the Internet of Things has given rise to a multitude of real-time applications, which pose significant computing challenges for resource-constrained users. Air-ground collaborative mobile edge computing (AGMEC) emerges as an innovative solution, integrating aerial and terrestrial computing paradigms to provide flexible, efficient services that significantly enhance data processing capabilities. This paper focuses on the freshness of task data in AGMEC networks, characterized by the emerging metric of age of information (AoI). Due to limited spectrum resources and network coverage gaps, we introduce non-orthogonal multiple access (NOMA) and active reconfigurable intelligent surface (RIS) technologies to facilitate efficient task offloading. We formulate a joint optimization problem of uncrewed aerial vehicle trajectory, active RIS beamforming, and task offloading strategy to minimize the network’s average AoI under multidimensional constraints. Considering the non-convex nature and the dynamic characteristics of the AGMEC environment, we develop an action adjuster-based deep deterministic policy gradient (AADDPG) algorithm. The innovative design of the action adjuster enables the algorithm to not only achieve efficient processing of hybrid action spaces but also effectively protect UAV battery performance. Simulation results demonstrate that the proposed AADDPG algorithm significantly improves AoI performance compared to other benchmark algorithms. Additionally, the results corroborate the efficacy of both NOMA and active RIS in minimizing AoI for AGMEC networks.

Accurate Channel State Information is prerequisite for intelligent sensing and ubiquitous connectivity. However, the diversity of channel conditions—from sparse to dense and static to fast-varying—fundamentally challenges traditional single and fixed estimation algorithms. To address this issue, this paper proposes a Physics-Prioritized Mixture of Experts (PP-MoE) scheme, leveraging the MoE paradigm’s ability to allocate resources to specialized experts tailored for distinct physical environments. The proposed scheme features an innovative heterogeneous expert library, where the architecture of each expert is customized with embedded physical priors to match its specific propagation environment. To enable intelligent scheduling, we design a hybrid decision gating network that collaboratively leverages physical formula computation and data-driven deep learning to achieve accurate channel environment identification and expert routing. Furthermore, to overcome the expert collapse problem, we propose a three-stage training strategy—pretraining, freezing, and fine-tuning to ensure training stability specialization. Extensive simulations demonstrate that PP-MoE significantly outperforms traditional and deep learning baselines. Notably, in the low-SNR region (0–15 dB), it achieves an NMSE nearly an order of magnitude lower than LMMSE. Additionally, PP-MoE maintains high efficiency with only 0.0256 GFLOPs. This work provides an effective paradigm for designing adaptive and physically reliable wireless physical layers.

The integration of movable antenna (MA) technique into unmanned aerial vehicle (UAV) communications offers a promising solution to reliable and energy-efficient non-terrestrial networking for low-altitude economy. To characterize multi-MA-assisted UAV-to-UAV wideband fading channels, we propose a three-dimensional arbitrary-elevation two-concentric-cylinders reference model. Based on this model, we derive the space-time-frequency correlation function (STF-CF) in closed form. From the STF-CF, we also obtain the space-Doppler power spectral density (SD-PSD) and the power space-delay spectrum (PSDS). The excellent agreement between the theoretical PSDS and some previously reported measurement data demonstrates the utility of the reference model. We then establish corresponding simulation models, which produce consistent results with the derived expressions. By leveraging the closed-form correlation function, the gradient of the log-determinant of spatial correlation matrix with respect to the MA positions can be conveniently obtained, which may serve as a method to maximize the ergodic capacity of the multi-MA assisted wideband UAV-to-UAV channels.

Prior studies on mixed near-field and far-field communications have focused exclusively onsingle-cellscenarios, where both near-field and far-field users are served by the same base station (BS), leading tointra-cellmixed-field interference. In this paper, we consider a more general and practicalmulti-cell mixed-fieldscenario consisting of multiple cells, each serving multiple users, thus resulting in more complexinter-cellmixed-field interference. To address this new challenge, we propose leveragingrotatable antenna(RA) technology to enhance multi-cell mixed-field communication performance by exploiting the additional spatial degree-of-freedom (DoF) introduced by RA rotation to mitigate interference in an efficient way. Specifically, we study an RA-enabled multi-cell mixed-field communication system in which each BS is equipped with an RA array to serve its associated users. We formulate a network-wide sum-rate maximization problem that jointly optimizes the transmit beamforming and the rotation angles of the RA arrays, subject to per-BS power constraints and admissible array rotation limits. To gain useful insights into the role of RAs in multi-cell mixed-field communications, we first analyze a special case with a single user per cell. For this case, we obtain a closed-form expression for therotation-awareinter-cell mixed-field interference using the Fresnel integrals and analytically show that RA rotation can effectively mitigate such interference, thereby substantially improving system performance. For the general case with multiple users per cell, we develop an efficientdouble-layeralgorithm: the inner layer optimizes the transmit beamforming at each BS via semidefinite relaxation (SDR) and successive convex approximation (SCA); while the outer layer determines the rotation angles of the RA arrays using particle swarm optimization (PSO). Numerical results demonstrate that RA-enabled multi-cell systems achieve significant performance gains over conventional fixed-antenna systems, and the proposed joint design consistently outperforms various benchmark schemes.

Integrated sensing and communication (ISAC) is expected to enable the fifth-generation (5G) networks to provide ubiquitous communication and sensing. However, some high-dynamic scenarios hinder applications of conventional ISAC schemes owing to the high overhead and poor real-time performance. In this paper, we design a novel ISAC architecture and propose a large language model (LLM) based two-stage beamforming prediction scheme. Specifically, in the first stage, we develop an LLM-based approach to predict the future channel state information (CSI) according to the history echoes. Via the data preprocessing and supervised fine-tuning, the LLM can achieve effective channel prediction task with unstructured data. In the second stage, according to the predicted/estimated CSI, we formulate a beamforming optimization problem to maximize the achievable sum rate while satisfying the quality of service (QoS). Then, we propose a Primary-dual network with the unsupervised adversarial learning to handle it, facilitating the on-line beamforming. Simulation results verify that, compared with the benchmarks, our proposed beamforming prediction scheme not only enjoys a higher channel prediction accuracy but also achieves a better balance between the performance and computational complexity.

Semantic communication (SemCom) has recently emerged as a promising paradigm for next-generation wireless systems. Empowered by advanced artificial intelligence (AI) technologies, SemCom has achieved significant improvements in transmission quality and efficiency. However, existing SemCom systems either rely on training over large datasets and specific channel conditions or suffer from performance degradation under channel noise when operating in a training-free manner. To address these issues, we explore the use of generative diffusion models (GDMs) as training-free SemCom systems. Specifically, we design a semantic encoding and decoding method based on the inversion and sampling process of the denoising diffusion implicit model (DDIM), which introduces a two-stage forward diffusion process, split between the transmitter and receiver to enhance robustness against channel noise. Moreover, we optimize sampling steps to compensate for the increased noise level caused by channel noise. We also conduct a brief analysis to provide insights about this design. Simulations on the Kodak dataset validate that the proposed system outperforms the existing baseline SemCom systems across various metrics.

This paper studies the joint beamforming and power allocation optimization in Low Earth Orbit (LEO) satellite networks with Rate-Splitting Multiple Access (RSMA), where dynamic channels and limited channel state information significantly degrade the performance of conventional optimization methods. Specifically, we formulate a sum-rate maximization problem under RSMA constraints. The decision variables include the transmit power allocated to the common and private streams, which are subject to total power and minimum user rate constraints. To solve this challenging problem, we propose STELLAR, a novel framework that employs a Large Language Model (LLM) as an intelligent decision-maker to directly generate feasible transmission strategies without requiring repeated model training. Specifically, STELLAR combines model-driven beamforming initialization with prompt-based evolutionary refinement and population updates, enabling rapid adaptation to varying channel conditions. Simulation results show that STELLAR outperforms baseline approaches, achieving superior spectral efficiency and converging within 30 iterations in a system with a 16-antenna LEO satellite and four ground stations.

Generative Artificial Intelligence (GenAI) models, with their powerful feature learning capabilities, have been applied in many fields. In mobile wireless communications, GenAI can dynamically optimize the network to enhance the user experience. Especially in signal detection and channel estimation tasks, due to digital signals following a certain random distribution, GenAI models can fully utilize their distribution learning characteristics. For example, diffusion models (DMs) and normalized flow models have been applied to related tasks. However, since the DM cannot guarantee that the generated results are the maximum-likelihood estimation points of the distribution during the data generation process, the successful task completion rate is reduced. Based on this, this paper proposes a Maximum-Likelihood Estimation Inference (MLEI) framework. The framework uses the loss function in the forward diffusion process of the DM to infer the maximum-likelihood estimation points in the discrete space. Then, we present a signal detection task in near-field communication scenarios with unknown noise characteristics. In experiments, numerical results demonstrate that the proposed framework has better performance than state-of-the-art signal estimators.

Integrated sensing and communication (ISAC) is emerging as a crucial technology for 6G networks, with channel state information (CSI) based ISAC playing a vital role. These systems utilize various AI models to process and analyze the CSI extracted from wireless communication signals, thereby enabling monitoring of physical spaces and human activities. However, due to the costs and privacy issues, collecting sufficient training CSI data is challenging. In response, this paper proposes a data augmentation system based on the diffusion model. Specifically, we first use the limited samples collected from real-world ISAC scenarios to train a conditional diffusion model, which then generates new samples to enhance sample quantity. Subsequently, we train another diffusion model with noise-free data to reduce noise in these generated samples, thereby further enhancing the sample quality. The evaluation based on the real-world CSI data validates that our approach can effectively enhance the data from both quantity and quality perspectives, thereby supporting the model training in ISAC networks.

Maintaining stable and efficient communication links for high-speed unmanned aerial vehicles (UAVs) in millimeter-wave (mmWave) communication systems remains a challenge due to beam misalignment caused by UAV mobility. Existing beam-tracking methods often struggle to provide accurate tracking under dynamic conditions, leading to frequent communication disruptions. To address this issue, we propose an enhanced Interactive Multiple Model (IMM) algorithm integrated with a Random Forest Model (RF-EIMM) to improve the accuracy and robustness of UAV trajectory predictions across diverse motion patterns. Furthermore, we propose an adaptive beamwidth optimization strategy that dynamically adjusts the beamwidth in real time, reducing the beam switching frequency, and minimizing the power consumption of the antenna array. Experimental results demonstrate that our approach significantly improves beam alignment accuracy, mitigates misalignment caused by UAV mobility, and outperforms existing methods in terms of spectral efficiency and beamforming gain.

Distributed Artificial Intelligence-Generated Content (AIGC) has attracted significant attention, but two key challenges remain: maximizing subjective Quality of Experience (QoE) and improving energy efficiency, which are particularly pronounced in widely adopted Generative Diffusion Model (GDM)-based image generation services. In this paper, we propose a novel user-centric Interactive AI (IAI) approach for service management, with a distributed GDM-based AIGC framework that emphasizes efficient and cooperative deployment. The proposed method restructures the GDM inference process by allowing users with semantically similar prompts to share parts of the denoising chain. Furthermore, to maximize the users' subjective QoE, we propose an IAI approach, i.e., Reinforcement Learning With Large Language Models Interaction (RLLI), which utilizes Large Language Model (LLM)-empowered generative agents to replicate users interactions, providing real-time and subjective QoE feedback aligned with diverse user personalities. Lastly, we present the GDM-based Deep Deterministic Policy Gradient (G-DDPG) algorithm, adapted to the proposed RLLI framework, to allocate communication and computing resources effectively while accounting for subjective user traits and dynamic wireless conditions. Simulation results demonstrate that G-DDPG improves total QoE by 15% compared with the standard DDPG algorithm.

As an important part of underwater optical technology, underwater imaging plays a crucial role in accurately capturing underwater targets and environmental features. Facing the challenges of complex ocean environments and severe photon attenuation, quantum imaging breaks through the limitations of traditional optical imaging technology by utilizing the characteristics of two-photon entanglement and time-space correlation, thus offering a new perspective on ocean turbulence. To this end, we propose a new underwater entangled-photon quantum imaging system. Specifically, we first construct an entangled photon quantum imaging physical model through ocean long-exposure turbulence and then exploit an entangled light coincidence imaging reconstruction method to image the target object. Furthermore, in response to the problem that ocean environment has a great impact on photons, we develop a photon capture probability method based on entangled photon pairs to reduce the impact of the ocean turbulence noise on imaging and improve the underwater target imaging resolution. We experimentally demonstrate the effectiveness of our method by showing that even in harsh ocean environments, quantum imaging performs superior resolution capabilities over traditional light source imaging techniques.

Large language models (LLMs) have shown great promise in many domains, yet their potential to transform wireless communications, where the escalating complexity of the network outpaces traditional model-based methods, remains largely untapped. Addressing this gap is critical for the next generation of intelligent and adaptive 6G systems. In this work, we develop a specialized dataset aimed at enhancing the evaluation and fine-tuning of LLMs specifically for wireless communication applications. The dataset includes a diverse set of multi-hop questions, including true/false and multiple-choice types, spanning varying difficulty levels from easy to hard. By utilizing advanced language models for entity extraction and question generation, rigorous data curation processes are employed to maintain high quality and relevance. Additionally, we introduce a Pointwise V-Information (PVI) based fine-tuning method, providing a detailed theoretical analysis and justification for its use in quantifying the information content of training data with 2.24% and 1.31% performance boost for different models compared to baselines, respectively. To demonstrate the effectiveness of the fine-tuned models with the proposed methodologies on practical tasks, we also consider different tasks, including summarizing optimization problems from technical papers and solving the mathematical problems related to non-orthogonal multiple access (NOMA), which are generated by using the proposed multi-agent framework. Simulation results show significant performance gain in summarization tasks with 20.9% in the ROUGE-L metrics. We also study the scaling laws of fine-tuning LLMs and the challenges LLMs face in the field of wireless communications, offering insights into their adaptation to wireless communication tasks. This dataset and fine-tuning methodology aim to enhance the training and evaluation of LLMs, contributing to advancements in LLMs for wireless communication research and applications.

Cell-free (CF) massive multiple-input multiple-output (MIMO) provides a ubiquitous coverage to user equipments (UEs) but it is also susceptible to interference. Rate-splitting (RS) effectively extracts data by decoding interference, yet its effectiveness is limited by the weakest UE. In this paper, we investigate an RS-based CF massive MIMO system, which combines strengths and mitigates weaknesses of both approaches. Considering imperfect channel state information (CSI) resulting from both pilot contamination and noise, we derive a closed-form expression for the sum spectral efficiency (SE) of the RS-based CF massive MIMO system under a spatially correlated Rician channel. Moreover, we propose low-complexity heuristic algorithms based on statistical CSI for power-splitting of common messages and power-control of private messages, and genetic algorithm is adopted as a solution for upper bound performance. Furthermore, we formulate a joint optimization problem, aiming to maximize the sum SE of the RS-based CF massive MIMO system by optimizing the power-splitting factor and power-control coefficient. Importantly, we improve a generative AI (GAI) algorithm to address this complex and non-convexity problem by using a diffusion model to obtain solutions. Simulation results demonstrate its effectiveness and practicality in mitigating interference, especially in dynamic environments.

This paper investigates the deep learning based approaches for simultaneous wireless information and power transfer (SWIPT). The quality-of-service (QoS) constrained sumrate maximization problems are, respectively, formulated for power-splitting (PS) receivers and time-switching (TS) receivers and solved by a unified graph neural network (GNN) based model termed SWIPT net (SWIPTNet). To improve the performance of SWIPTNet, we first propose a single-type output method to reduce the learning complexity and facilitate the satisfaction of QoS constraints, and then, utilize the Laplace transform to enhance input features with the structural information. Besides, we adopt the multi-head attention and layer connection to enhance feature extracting. Furthermore, we present the implementation of transfer learning to the SWIPTNet between PS and TS receivers. Ablation studies show the effectiveness of key components in the SWIPTNet. Numerical results also demonstrate the capability of SWIPTNet in achieving nearoptimal performance with millisecond-level inference speed which is much faster than the traditional optimization algorithms. We also show the effectiveness of transfer learning via fast convergence and expressive capability improvement.

This paper investigates adaptive transmission strategies in embodied AI-enhanced vehicular networks by integrating vision language models (VLMs) for semantic information extraction and deep reinforcement learning (DRL) for decision-making. The proposed framework aims to optimize both data transmission efficiency and decision accuracy by formulating an optimization problem that incorporates the Weber-Fechner law, serving as a metric for balancing bandwidth utilization and quality of experience (QoE). Specifically, we employ the large language and vision assistant (LLAVA) model to extract critical semantic information from raw image data captured by embodied AI agents (i.e., vehicles), reducing transmission data size by approximately more than 90% while retaining essential content for vehicular communication and decision-making. In the dynamic vehicular environment, we employ a generalized advantage estimation-based proximal policy optimization (GAE-PPO) method to stabilize decision-making under uncertainty. Simulation results show that attention maps from LLAVA highlight the model's focus on relevant image regions, enhancing semantic representation accuracy. Additionally, our proposed transmission strategy improves QoE by up to 36% compared to DDPG and accelerates convergence by reducing required steps by up to 47% compared to pure PPO. Further analysis indicates that adapting semantic symbol length provides an effective trade-off between transmission quality and bandwidth, achieving up to a 61.4% improvement in QoE when scaling from 4 to 8 vehicles.

This paper investigates intelligent reflecting surface (IRS)-assisted multiple-input single-output (MISO) visible light communication (VLC) networks utilizing the rate-splitting multiple access (RSMA) scheme. In these networks, an eavesdropper (Eve) attempts to eavesdrop on communications intended for legitimate users (LUs). To enhance information security and energy efficiency simultaneously, we formulate a secrecy energy efficiency (SEE) maximization problem by jointly optimizing the beamforming vectors, RSMA common rates, direct current (DC) bias, and IRS alignment matrices. The problem is constrained by total power budget, quality of service (QoS) requirements, linear operating region of light emitting diodes (LEDs), and common information rate allocation. Due to the non-convex and NP-hard nature of the formulated problem, we propose a deep reinforcement learning (DRL)-based dual-sampling proximal policy optimization (DS-PPO) approach. The approach leverages dual sample strategies and generalized advantage estimation (GAE). In addition, the maximum ratio transmission (MRT) and zero-forcing (ZF) are adopted to design the beamforming vectors. Simulation results show that the proposed DS-PPO approach outperforms traditional baseline approaches. Moreover, the implementation of the RSMA scheme and IRS contributes to overall system performance, achieving approximately 19.67% improvement over traditional multiple access schemes and 25.74% improvement over networks without IRS deployment.

In response to the needs of 6G global communications, satellite communication networks have emerged as a key solution. However, the large-scale development of satellite communication networks is constrained by complex system models, whose modeling is challenging for massive users. Moreover, transmission interference between satellites and users seriously affects communication performance. To solve these problems, this paper develops generative artificial intelligence (AI) agents for model formulation and then applies a mixture of experts (MoE) approach to design transmission strategies. Specifically, we leverage large language models (LLMs) to build an interactive modeling paradigm and utilize retrieval-augmented generation (RAG) to extract satellite expert knowledge that supports mathematical modeling. Afterward, by integrating the expertise of multiple specialized components, we propose an MoE-proximal policy optimization (PPO) approach to solve the formulated problem. Each expert can optimize the optimization variables at which it excels through specialized training through its own network and then aggregate them through the gating network to perform joint optimization. The simulation results validate the accuracy and effectiveness of employing a generative agent for problem formulation. Furthermore, the superiority of the proposed MoE-ppo approach over other benchmarks is confirmed in solving the formulated problem. The adaptability of MoE-PPO to various customized modeling problems has also been demonstrated.

This paper investigates simultaneous wireless information and power transfer (SWIPT)-enabled cell-free massive multiple-input multiple-output (CF-mMIMO) networks with power splitting (PS) receivers and non-orthogonal multiple access (NOMA). By exploiting the conjugated beamforming method, the closed-form expressions of the information rate and the total harvested power at each user equipment (UE) are derived. To improve the system spectral efficiency, a sum rate maximization problem is formulated subjecting to the quality of service requirement at each UE and the power budget constraint at each access point by optimizing the UE clustering, the power control coefficients, and the PS ratios. To solve the formulated non-convex and mixed combinatorial problem, a machine learning-based approach is designed. Particularly, the UE clustering is first optimized by using a K-means based method and then the power control coefficients and the PS ratios are jointly optimized by a proposed multi-agent deep Q-network (MA-DQN) based method. The impact of the discount factor of the MA-DQN based method on the derived result is discussed. It is proved that by setting the discount factor as zero, the performance loss is negligible. Based on this observation, a zero-discount MA-DQN (0-γ MA-DQN) based method is further proposed to improve the computational efficiency. Also, the computational complexity of the proposed machine learning-based approach is analyzed. Simulation results show that the proposed machine learning-based approach outperforms various existing approaches. Moreover, it indicates that CF-mMIMO and NOMA could enhance the propagation performance of SWIPT while the proposed machine learning-based approach could facilitate resource allocation.

This paper investigates reconfigurable intelligent surface (RIS)-assisted simultaneous wireless information and power transfer (SWIPT) networks with rate splitting multiple access (RSMA). An energy efficiency (EE) maximization problem is formulated subject to the power budget at the transmitter and the quality of service (QoS) requirements of both information communication and energy harvesting, where the beamforming vectors, the power splitting (PS) ratios, the common message rates, and the discrete phase shifts are jointly optimized. To tackle the non-convex problem with both discrete and continuous variables, a deep reinforcement learning-based approach is proposed with the proximal policy optimization (PPO) framework. Different from traditional optimization approaches which optimizes the beamforming vectors and phase shifts separately and alternatively, our proposed PPO-based approach optimizes all the variables in unison. Besides, to perform beamforming design in action space, the beamforming vectors for the common stream and the private stream are respectively designed based on the maximum-ratio transmission and the zero forcing to enhance both energy and information transmission. To evaluate the performance of the PPO-based approach, a successive convex approximation (SCA) and Dinkelbach’s method based solution scheme (named SCA-D scheme) is also presented. Simulation results show that the system EE obtained by the proposed PPO-based approach is close to that obtained by the SCA-D scheme while outperforming various benchmarks. The RSMA contributes to the EE of the system greatly compared with traditional scheme. As for the case of time-varying channels, the proposed PPO-based approach is with much smaller running time by only sacrificing a slight EE performance compared with the SCA-D scheme.

This paper proposes a multi-agent double deep Q network (DDQN)-based approach to jointly optimize the beamforming vectors and power splitting (PS) ratio in multi-user multiple-input single-output (MU-MISO) simultaneous wireless information and power transfer (SWIPT)-enabled heterogeneous networks (HetNets), where a macro base station (MBS) and several femto base stations (FBSs) serve multiple macro user equipments (MUEs) and femto user equipments (FUEs). The PS receiver architecture is deployed at FUEs. An optimization problem is formulated to maximize the achievable sum information rate of FUEs under the constraints of the achievable information rate requirements of MUEs and FUEs and the energy harvesting (EH) requirements of FUEs. Since the optimization problem is challenging to handle due to the high dimension and time-varying environment, an efficient multi-agent DDQN-based algorithm is presented, which is trained in a centralized manner and runs in a distributed manner, where two sets of deep neural network parameters are jointly updated and trained to tackle the problem and avoid overestimation. To facilitate the presented multi-agent DDQN-based algorithm, the action space, the state space and the reward function are designed, where the codebook matrix is employed to deal with the complex transmit beamforming vectors. Simulation results validate the proposed algorithm. Notable performance gains are achieved by the proposed algorithm due to considering the beam directions in the action space and the adaptability to the Doppler frequency shifts. Besides, the proposed algorithm is shown to be superior to other benchmark ones numerically.