In B5G and 6G wireless networks, large-scale antenna arrays introduce spherical wavefronts in the near-field region, presenting both opportunities and challenges for communication and positioning. To address this, we propose a novel near-field positioning method using a bio-inspired dendritic liquid neural network (DLNN). The DLNN features a multi-branch dendritic structure, with each neuron processing signals through multiple parallel dendrites and a dynamic gating mechanism to adapt the weighting and fusion of outputs. This enhances its ability to capture complex spatiotemporal features. Additionally, the liquid layer with leaky integration processes time-series data, enabling strong adaptability and robust temporal memory. Simulation results show that DLNN outperforms other deep learning models in localization accuracy and robustness, with RMSE values of 0.1706 meters at 40 dBm and 0.2642 meters at 10 dBm, compared to 0.2836 meters and 0.2986 meters from other models. Its inference latency of 0.0580 seconds demonstrates excellent accuracy, speed, and energy efficiency, making DLNN highly suitable for real-time positioning in resource-constrained environments.

Digital Twins (DTs) have been proposed for monitoring and decision-making of complex systems, without being in the presence of the target system. However, this capability typically requires non-trivial amounts of data from target devices. In the case of lunar orbit and surface operations, obtaining data takes place over bandwidth-constrained links with long propagation delays and intermittent connectivity. These limitations are already evident in emerging architectures such as NASA’s Artemis program, where continuous data streaming from lunar assets is not always feasible. This paper proposes TwinSync, a mission-aware networking framework for maintaining synchronized digital twin states over dynamic cislunar networks. The approach combines a Cislunar Network Twin (CNT), which predicts time-varying topology and communication opportunities, with an Age-of-Synchronization (AoS)-aware scheduling mechanism that prioritizes deadline-constrained updates. Consequently, TwinSync ensures that missioncritical Time-Triggered (TT) traffic is supported by design. Simulation results show that TwinSync achieves zero TT deadline violations for deadline-constrained traffic across all evaluated scenarios. It also improves overall network utilization compared to representative deep space networking schemes, demonstrating a shift from reactive best-effort networking to predictive, mission-aware operation, where critical objectives are guaranteed a priori. Extended stress testing reveals that TwinSync maintains delivery of Time-Triggered information even under highly constrained traffic rates, dynamic mission phase transitions, and degraded network conditions.

Next-generation Wi-Fi 8 (IEEE 802.11bn) targets ultra-high reliability (UHR) by introducing coordinated beamforming (CoBF). In dense networks with multiple access points (APs), simultaneous downlink (DL) multi-user MIMO (MU-MIMO) transmissions from multiple APs can cause severe intra-basic service set (intra-BSS) and inter-BSS interference. CoBF aided by only partial channel state information (CSI) feedback through medium access control (MAC) layer frame exchange is envisioned to support concurrent DL transmission with mitigated physical-(PHY-)layer interference. To improve the network throughput, not only the interference mitigation algorithm design requires careful design but also the selection of optimal AP CoBF clusters is crucial for dense AP deployments. This paper presents a cross-layer solution combining PHY and MAC layer design to optimize AP cluster formation for Wi-Fi 8 CoBF. At the PHY layer, we introduce two beamforming nulling strategies: full nulling, which completely cancels all intra-BSS and inter-BSS interference when sufficient spatial degrees of freedom are available, and partial nulling, which is used under limited degrees of freedom to reduce interference as much as possible. Based on this, we formulate the cross-layer problem that aims to optimize the network throughput, to which we propose an exact linear programming (LP) optimization to determine the optimal AP cluster formation. A greedy clustering algorithm is proposed as a low-complexity alternate. Simulation results demonstrate that the proposed CoBF approach significantly mitigates interference and achieves substantial throughput gains in dense AP scenarios. Furthermore, the LP-optimized AP clustering yields the higher network throughput than the greedy heuristic and mixed integer linear programming (MILP) by up to 12% and 26%, highlighting the benefits of global optimization in terms of performance and time complexity.

The advent of Large Multimodal Models (LMMs) offers a promising technology to tackle the limitations of modular design in autonomous driving, which often falters in open-world scenarios requiring sustained environmental understanding and logical reasoning. Besides, embodied artificial intelligence facilitates policy optimization through closed-loop interactions to achieve the continuous learning capability, thereby advancing autonomous driving toward embodied intelligent (El) driving. However, such capability will be constrained by relying solely on LMMs to enhance EI driving without joint decision-making. This article introduces a novel semantics and policy dual-driven hybrid decision framework to tackle this challenge, ensuring continuous learning and joint decision. The framework merges LMMs for semantic understanding and cognitive representation, and deep reinforcement learning (DRL) for real-time policy optimization. We starts by introducing the foundational principles of EI driving and LMMs. Moreover, we examine the emerging opportunities this framework enables, encompassing potential benefits and representative use cases. A case study is conducted experimentally to validate the performance superiority of our framework in completing lane-change planning task. Finally, several future research directions to empower EI driving are identified to guide subsequent work.

In this paper, we propose the geometric algebra-informed neural radiance fields (GAI-NeRF), a novel framework for wireless channel prediction that leverages geometric algebra attention mechanisms to capture ray-object interactions in complex propagation environments. Our approach incorporates global token representations, drawing inspiration from transformer architectures in language and vision domains, to aggregate learned spatial-electromagnetic features and enhance scene understanding. We identify limitations in conventional static ray tracing modules that hinder model generalization and address this challenge through a new ray tracing architecture. This design enables effective generalization across diverse wireless scenarios while maintaining computational efficiency. Experimental results demonstrate that GAI-NeRF achieves superior performance in channel prediction tasks by combining geometric algebra principles with neural scene representations, offering a promising direction for next-generation wireless communication systems. Moreover, GAI-NeRF greatly outperforms existing methods across multiple wireless scenarios. To ensure comprehensive assessment, we further evaluate our approach against multiple benchmarks using newly collected real-world indoor datasets tailored for single-scene downstream tasks and generalization testing, confirming its robust performance in unseen environments and establishing its high efficacy for wireless channel prediction.

World models are emerging as a transformative paradigm in artificial intelligence, enabling agents to construct internal representations of their environments for predictive reasoning, planning, and decision-making. By learning latent dynamics, world models provide a sample-efficient framework that is especially valuable in data-constrained or safety-critical scenarios. In this article, we present a comprehensive overview of world models, highlighting their architecture, training paradigms, and applications across prediction, generation, planning, and causal reasoning. We compare and distinguish world models from related concepts such as digital twins, the metaverse, and foundation models, clarifying their unique role as embedded cognitive engines for autonomous agents. We further propose Wireless Dreamer, a novel world model-based reinforcement learning framework tailored for wireless edge intelligence optimization, particularly in low-altitude wireless networks (LAWNs). Through a weather-aware UAV trajectory planning case study, we demonstrate the effectiveness of our framework in improving learning efficiency and decision quality.

Mixture of Experts (MoE) has emerged as a promising paradigm for scaling model capacity while preserving computational efficiency, particularly in large-scale machine learning architectures such as large language models (LLMs). Recent advances in MoE have facilitated its adoption in wireless networks to address the increasing complexity and heterogeneity of modern communication systems. This paper presents a comprehensive survey of the MoE framework in wireless networks, highlighting its potential in optimizing resource efficiency, improving scalability, and enhancing adaptability across diverse network tasks. We first introduce the fundamental concepts of MoE, including various gating mechanisms and the integration with generative AI (GenAI) and reinforcement learning (RL). Subsequently, we discuss the extensive applications of MoE across critical wireless communication scenarios, such as vehicular networks, unmanned aerial vehicles (UAVs), satellite communications, heterogeneous networks, integrated sensing and communication (ISAC), and mobile edge networks. Furthermore, key applications in channel prediction, physical layer signal processing, radio resource management, network optimization, and security are thoroughly examined. Additionally, we present a detailed overview of open-source datasets that are widely used in MoE-based models to support diverse machine learning tasks. Finally, this survey identifies crucial future research directions for MoE, emphasizing the importance of advanced training techniques, resource-aware gating strategies, and deeper integration with emerging 6G technologies.

AI-enabled wireless communications have attracted tremendous research interest in recent years, particularly with the rise of novel paradigms such as low-altitude integrated sensing and communication (ISAC) networks. Within these systems, feature engineering plays a pivotal role by transforming raw wireless data into structured representations suitable for AI models. Hence, this article offers a comprehensive investigation of feature engineering techniques in AI-driven wireless communications. Specifically, we begin with a detailed analysis of fundamental principles and methodologies of feature engineering. Next, we present its applications in wireless communication systems, with special emphasis on ISAC networks. Finally, we introduce a generative AI-based framework, which can reconstruct signal feature spectrum under malicious attacks in low-altitude ISAC networks. The case study shows that it can effectively reconstruct the signal spectrum, achieving an average structural similarity index improvement of 4%, thereby supporting downstream sensing and communication applications.

The proliferation of radar systems in vehicles, infrastructure, and IoT devices has created an unprecedented opportunity for hidden communication channels that leverage existing hardware. This paper introduces SecRadCom, a novel framework that embeds covert communication in frequency-modulated continuous wave (FMCW) radar systems while preserving sensing functionality. Our key innovation is a cross-layer approach combining physical-layer signal embedding with network-layer protocols specifically designed for radar-based covert communications. We present three embedding techniques—phase-attached modulation, constrained frequency hopping, and phase-coded orthogonal sequencing—and develop a comprehensive architecture for establishing reliable multi-node communication networks. We formally prove the theoretical security bounds, achieving a probability of detection below 10−3 while maintaining mean communication rates up to 100 kbps. Our hardware-in-the-loop implementation on commercial USRP SDR platforms demonstrates minimal sensing performance degradation (<1.2% in range accuracy) while providing 36 dB security gain against adversaries. Extensive evaluations in vehicular and smart infrastructure scenarios show that SecRadCom outperforms state-of-the-art covert communication techniques by 2.1× in throughput and 1.7× in energy efficiency while maintaining robust operation under various channel conditions and network densities.

Floor plans can provide valuable prior information that helps enhance the accuracy of indoor positioning systems. However, existing research typically faces challenges in efficiently leveraging floor plan information and applying it to complex indoor layouts. To fully exploit information from floor plans for positioning, we propose a floor plan-assisted fusion positioning algorithm (FP-BP) using Bluetooth low energy (BLE) and pedestrian dead reckoning (PDR). In the considered system, a user holding a smartphone walks through a positioning area with BLE beacons installed on the ceiling, and can locate himself in real time. In particular, FP-BP consists of two phases. In the offline phase, FP-BP programmatically extracts map features from a stylized floor plan based on their binary masks, and constructs a mapping function to identify the corresponding map feature of any given position on the map. In the online phase, FP-BP continuously computes BLE positions and PDR results from BLE signals and smartphone sensors, where a novel grid-based maximum likelihood estimation (GML) algorithm is introduced to enhance BLE positioning. Then, a particle filter is used to fuse them and obtain an initial estimate. Finally, FP-BP performs post-position correction to obtain the final position based on its specific map feature. Experimental results show that FP-BP can achieve a real-time mean positioning accuracy of 1.14 m, representing an improvement of over 29

The proliferation of Internet-of-things (IoT) infrastructures and the widespread adoption of traffic encryption present significant challenges, particularly in environments characterized by dynamic traffic patterns, constrained computational capabilities, and strict latency constraints. In this paper, we propose DMLITE, a diffusion model and large language model (LLM) integrated traffic embedding framework for network traffic detection within resource-limited IoT environments. The DMLITE overcomes these challenges through a tri-phase architecture including traffic visual preprocessing, diffusion-based multi-level feature extraction, and LLM-guided feature optimization. Specifically, the framework utilizes self-supervised diffusion models to capture both fine-grained and abstract patterns in encrypted traffic through multi-level feature fusion and contrastive learning with representative sample selection, thus enabling rapid adaptation to new traffic patterns with minimal labeled data. Furthermore, DMLITE incorporates LLMs to dynamically adjust particle swarm optimization parameters for intelligent feature selection by implementing a dual objective function that minimizes both classification error and variance across data distributions. Comprehensive experimental validation on benchmark datasets confirms the effectiveness of DMLITE, achieving classification accuracies of 98.87%, 92.61%, and 99.83% on USTC-TFC, ISCX-VPN, and Edge-IIoTset datasets, respectively. This improves classification accuracy by an average of 3.7% and reduces training time by an average of 41.9% compared to the representative deep learning model.

In this paper, we propose a robust, scalable, and interpretable WiFi Channel State Information (CSI) based sensing system that integrates functional data analysis (FDA) and large language models (LLMs) to achieve high performance on human activity and identity recognition tasks. Specifically, FDA is applied to reconstruct and extract structured features from CSI sequences, which are then processed by a lightweight LLM backbone incorporated with an uncertainty aware mechanism. To further enhance confidence calibration and feature representation, we design a composite loss function that integrates generative classification, contrastive learning, and uncertainty aware objectives. We conduct experiments on multiple public CSI-based sensing datasets. The results demonstrate competitive recognition accuracy and meaningful confidence calibration compared with representative deep learning baselines.

In this paper, we investigate a near-field integrated sensing and communication (ISAC) system, where the base station is equipped with a uniform linear array composed of several widely-spaced subarrays. Due to the near-field target sensing, we consider extended targets (ETs) instead of single-point targets. In particular, we model the ET by its center point and contour points, and then derive the expression of the Cramér-Rao bound (CRB) on the ET parameter estimation. Aiming to maximize the CRB, we propose a dual-loop triple alternating optimization (DTAO) scheme based on penalty dual decomposition to jointly optimize subarray selection, digital beamformer and analog beamformer, subject to the constraints of communication-to-interference ratio, total transmit power and binary constraints of subarray selection and constant modulus of analog beamformer. In the inner loop, the digital beamformer, analog beamformer and subarray selection vector are alternately optimized, while in the outer loop, the dual variables and penalty coefficients are iteratively updated until the convergence condition is met. Numerical results show the validity of the near-field CRB on the ET parameter estimation, and reveal the proposed DTAO scheme can enhance the sensing performance of the ISAC system.

In the upcoming sixth-generation (6G) era, supporting field robots for unmanned operations has emerged as an important application direction. To provide connectivity in remote areas, the space–air–ground integrated network (SAGIN) will play a crucial role in extending coverage. Through SAGIN connections, the sensors, edge platforms, and actuators form sensing–communication–computing–control (SC3) loops that can automatically execute complex tasks without human intervention. Similar to the reflex arc, the SC3 loop is an integrated structure that cannot be deconstructed. This necessitates a systematic approach that takes the SC3 loop rather than the communication link as the basic unit of SAGINs. Given the resource limitations in remote areas, we propose a radio-map-based task-oriented framework that uses environmental and task-related information to enable task-matched service provision. We detail how the network collects and uses this information and present task-oriented scheduling schemes. In the case study, we use a control task as an example and validate the superiority of the task-oriented closed-loop optimization scheme over traditional communication schemes. Finally, we discuss open challenges and possible solutions for developing nerve system-like SAGINs.

The rapid advancement toward sixth-generation (6G) wireless networks has significantly intensified the complexity and scale of optimization problems, including resource allocation and trajectory design, often formulated as combinatorial problems in large discrete decision spaces. However, traditional optimization methods, such as heuristics and deep reinforcement learning (DRL), face practical challenges in meeting stringent latency and scalability requirements, especially in large-scale, highly dynamic, and reconfiguration-sensitive deployments in increasingly heterogeneous and resource-constrained network environments. Large language models (LLMs) present a transformative paradigm by enabling natural language-driven problem formulation, context-aware reasoning, and adaptive solution refinement through advanced semantic understanding and structured reasoning capabilities. This paper provides a systematic and comprehensive survey of LLM-enabled optimization frameworks tailored for wireless networks. We first introduce foundational design concepts and distinguish LLM-enabled methods from conventional optimization paradigms. Subsequently, we critically analyze key enabling methodologies, including natural language modeling, solver collaboration, and solution verification processes. Moreover, we explore representative case studies to demonstrate LLMs’ transformative potential in practical scenarios such as optimization formulation, low-altitude economy networking, and intent networking. Finally, we discuss current research challenges, examine prominent open-source frameworks and datasets, and identify promising future directions to facilitate robust, scalable, and trustworthy LLM-enabled optimization solutions for next-generation wireless networks.

Large-language-model (LLM) agents for wireless systems have emerged rapidly, but nearly all of them live at the text or network-management layer. Knowledge-layer models answer questions about signals, orchestration-layer agents route intents over text and KPI metadata, and image-layer prompting operates on pre-computed CSI heatmaps; none of these consume or produce raw physical-layer RF signals. We introduce SenseAgent, a signal-grounded wireless sensing agent organized as a training-free mixture of experts (MoE-TF-DM), in which each expert is a training-free diffusion module specialized for one RF modality (RFID activity, WiFi CSI, or IQ modulation) and an LLM serves as both the modality gate and the natural-language interpreter over 11 signal-processing tools. Unlike classical mixture-of-experts baselines for wireless, whose priors live in pretrained expert weights and require retraining for each new class or environment, SenseAgent’s experts bind the prior to reference samples and adapt at inference time from a handful of few-shot exemplars. We support this architecture with a tool-selection evaluation harness of 30 natural-language queries that exercises top-1 accuracy, description robustness, and an LLM-only ablation. The agent reaches 96.7% top-1 tool selection at verbose default, drops to 23.3% when tool descriptions are truncated to a single sentence, and falls to zero when the LLM is asked to produce the same outputs without any tools, confirming that the tool layer is constitutive rather than decorative. On RadioML 2016.10a, the training-free IQ expert improves over a source-only baseline on 28 of 30 benchmark cells spanning a range of few-shot budgets and signal-to-noise ratios, with a mean accuracy gain of 6.8 percentage points, while running at roughly 1/180-th the synthesis time of a matched trained DDPM, directly quantifying the advantage of a data-bound prior under few-shot target-domain conditions. Lightweight case studies across IQ modulation, RFID, and WiFi CSI exercise zero-retraining adaptation through the full ReAct loop. Full quantitative per-modality studies are the subject of ongoing journal-length work; this invited paper establishes the agent framework, its evaluation, and a research agenda for signal-grounded sensing agents in 6G AI-native networks.

Wi-Fi-based human activity recognition (HAR) has emerged as a focal point within the Internet of Things landscape, due to its nonintrusive sensing capabilities and inherent privacy-preserving advantages. In existing Wi-Fi-based HAR research, channel state information (CSI) is primarily utilized to capture activity-related features and enable recognition. However, CSI-based cross-domain HAR remains challenged by issues, such as redundant subcarriers, limited samples in the target domain, and high sensitivity of CSI to environmental variations. To address these challenges, this article proposes Wi-DMAR, a Wi-Fi-based cross-domain HAR framework that integrates three key modules. First, an adaptive subcarrier selection module computes the correlation between each subcarrier and the principal components, identifies subcarriers with high contribution, preserves essential activity-related features, reduces data dimensionality, and lowers computational overhead. Second, a conditional diffusion-based data augmentation module employs a Transformer-based feature extractor to capture domain-specific representations of the target-domain data, and optimizes the domain consistency loss and domain-guided diffusion loss to generate pseudo-samples that resemble the target-domain distribution, thereby mitigating the sample scarcity. Third, an activity recognition module based on sample similarity learning reformulates the traditional label classification problem into a sample comparison task, by quantifying similarity between samples, and it performs the activity recognition and enhances the cross-domain generalization. Experimental results demonstrate that Wi-DMAR achieves the superior recognition accuracy compared with state-of-the-art cross-domain HAR methods, such as DiffAR and MetaAct. Ablation studies further confirm that each core component contributes positively to performance improvements.

In this paper, we investigate the low-complexity distributed combining scheme design for near-field cell-free extremely large-scale multiple-input-multiple-output (CF XL-MIMO) systems. Firstly, we construct the uplink spectral efficiency (SE) performance analysis framework for CF XL-MIMO systems over centralized and distributed processing schemes. Notably, we derive the centralized minimum mean-square error (CMMSE) and local minimum mean-square error (LMMSE) combining schemes over arbitrary channel estimators. Then, focusing on the CMMSE and LMMSE combining schemes, we propose five low-complexity distributed combining schemes based on the matrix approximation methodology or the symmetric successive over relaxation (SSOR) algorithm. More specifically, we propose two matrix approximation methodology-aided combining schemes: Global Statistics \& Local Instantaneous information-based MMSE (GSLI-MMSE) and Statistics matrix Inversion-based LMMSE (SI-LMMSE). These two schemes are derived by approximating the global instantaneous information in the CMMSE combining and the local instantaneous information in the LMMSE combining with the global and local statistics information by asymptotic analysis and matrix expectation approximation, respectively. Moreover, by applying the low-complexity SSOR algorithm to iteratively solve the matrix inversion in the LMMSE combining, we derive three distributed SSOR-based LMMSE combining schemes, distinguished from the applied information and initial values.

As Gaussian Splatting (GS) is increasingly integrated into products and production pipelines across industry, organizations and researchers have begun developing standardization practices to formally accommodate this emerging technology. Originally regarded primarily as a 3D rendering technique for novel view synthesis, 3DGS can be applied more broadly to represent a wide range of 3D computer graphics content. However, as GS is a relatively recent technology, much of its development remains at an experimental stage, resulting in a sparse standardization landscape. As 3D modeling becomes increasingly comprehensive and the applications and implementations of GS rapidly diversify, many organizations have recognized the need for standardization, introducing new formats such as SPZ or adapting GS to existing 3D asset standards like glTF. In this article, we review the initial standardization efforts led by organizations such as Metaverse Standards Forum, Niantic, Khronos, and MPEG, which aim to unify baseline testing conditions, performance metrics, and development practices for GS. We begin by outlining the implementation and evolution of GS, highlighting how it advances beyond previous 3D scene representation approaches. We then provide an overview of emerging splatting standards. Finally, we conclude by discussing future research directions for GS and examining its current and potential applications.

We consider decentralized federated learning (DFL) in mobile computing networks (MCNs), where dynamically changing neighborhood sets among devices arise from mobility and environmental perturbations. The time-varying topology coupled with inherent system heterogeneity poses significant challenges to achieve stable and efficient convergence in DFL. However, existing studies rarely consider both dynamic connectivity and statistical heterogeneity. To close this gap, this paper proposes a novel DFL framework enhanced with topology learning (DFL-TL) to mitigate the spatio-temporal volatility induced by MCNs, where each mobile device faces coupled constraints on its temporal windows for local updates and spatial scopes for model interaction. We introduce a new bounded neighborhood heterogeneity to jointly measure and constrain both the inter-device heterogeneity and the spectral properties of the topologies. Additionally, we formulate a mixed-integer nonlinear programming (MINLP) problem to jointly optimize learning costs and neighborhood heterogeneity. Through problem decomposition, DFL-TL efficiently identifies optimal resource allocations and adaptive mixing matrices, thereby enabling the selection of optimal training time windows while reducing the adverse effects of dynamic topologies in heterogeneous networks. Furthermore, we establish the iteration complexity of DFL-TL under non-convex settings and show that solving the proposed MINLP formulation leads to a tighter convergence bound. Extensive experiments demonstrate that DFL-TL achieves a faster convergence performance and reduces the wall-clock training time compared to the state-of-the-art baselines.