This article conceives an electromagnetic-compliant architecture for practical deployments of holographic multiple-input multiple-output (HMIMO) systems, addressing an intractable dilemma: “How to avoid mutual coupling under sub-half-wavelength antenna spacing?” To answer this question, we pivot from conductor-based antennas to Rydberg atomic receivers and present design principles of mutual coupling-free HMIMO. In re-examining the fundamental limitations of conductor-based antennas, we demonstrate the inherent superiorities of Rydberg atomic receivers (RARs) from an electromagnetic-compliant perspective. Subsequently, a thorough comparison of the signal detection mechanism between two types of RAR is provided, facilitating a unified review on the operating principles of RAR. Furthermore, we provide a comprehensive exposition on the proposed RAR-based holographic MIMO architecture, along with crucial engineering considerations in practical deployments.
Open Radio Access Networks (O-RAN) introduce unprecedented flexibility, interoperability, and intelligence into next-generation wireless systems, but their disaggregated and software-defined architecture also expands the attack surface and creates new security vulnerabilities. Conventional cryptographic mechanisms, while effective against classical threats, may become insufficient in the presence of quantum-enabled adversaries. This article presents a comprehensive perspective on quantum security for O-RAN, examining how quantum-resilient mechanisms can enhance confidentiality, authentication, and trust across the RAN ecosystem. It discusses post-quantum cryptography (PQC), quantum cryptography, quantum authentication, and quantum-enhanced threat detection within a zero-trust architecture based on continuous verification, least privilege, and micro-segmentation. Their integration with the Near-Real-Time (Near-RT) RAN Intelligent Controller, O-Cloud, and open interfaces is analyzed, together with practical deployment considerations, technology maturity, and adoption timelines. Finally, open research directions are outlined toward secure, resilient, and future-proof O-RAN architectures for 6G networks.
Sixth generation (6G) wireless networks aim to move beyond connected devices toward a “connected cognition” model in which the network understands intent, reasons about constraints, and executes actions autonomously. This article proposes an Agentic-Native 6G architecture embedding intelligence directly into network functions via three pillars: a distributed Knowledge Plane that lifts multimodal telemetry into semantic knowledge graphs within a Semantic Knowledge Base (SKB); autonomous agents equipped with perception modules, Recurrent State Space Model world models, Linear Temporal Logic guardrails, and Large Action Models (LAMs); and an Agent-to-Agent fabric for collaborative semantic inference. We clarify the hierarchy of Knowledge Plane, SKB, and knowledge graphs; distinguish LAMs from LLMs across training corpus, output modality, and action space; and quantify computational overhead. A proof-of-concept O-RAN implementation for industrial robot control outperforms KDN-style heuristics, model-free reinforcement learning, and recent world-model and intent-based baselines across KPI forecasting, latency control under jamming, and out-of-distribution adaptation.
With the advances of underwater networking and multi-agent reinforcement learning (MARL), autonomous underwater vehicle (AUV) cluster network-based intelligent formation control is becoming a promising platform for smart-ocean missions. However, severe and time-varying ocean disturbances remain a major barrier, often causing MARL policies trained under nominal conditions to degrade or fail during deployment. To mitigate the above issues, this article, motivated by embodied intelligence, presents an Underwater Multi-Agent Embodied Intelligence (UMAEI) architecture that organizes underwater cluster autonomy as a closed-loop sense–decide–act system. We argue that an underwater multi-agent embodied intelligence architecture should emphasize both data efficiency and robustness. Accordingly, UMAEI incorporates two complementary mechanisms to improve practicality. First, a continuous-time refinement-based sample augmentation method is proposed to densify sparsely sampled underwater trajectories and improve training data efficiency. Second, a disturbance-compensated robust MARL algorithm embeds a lightweight recursive least-squares compensation module with binary gating to handle strong and unseen disturbances. To enable robust and intelligent formation control in AUV cluster networks, we build on UMAEI and develop a dedicated formation scheme. The scheme consists of a decentralized path-planning reward, a UMAEI-based formation control policy, and an empowerment-based training strategy for AUV cluster networks. Evaluation results showcase that the proposed formation scheme can achieve exact and robust AUV cluster network-based formation in high-fidelity marine environment.
The rapid development of artificial intelligence (AI) has led to the emergence of agent technology driven by large language models (LLMs), which are regarded as valid solutions to numerous cross-field challenges owing to their tremendous intelligence and autonomy. When the agents are applied to the radio access network (RAN), a new paradigm, agentic AI-RAN, rises in response, which integrates communication, sensing, computation, and control through the advanced decision-making capabilities and self-sufficiency of agents. Due to different design purposes, agentic AI-RAN has two unique features compared to conventional AI-empowered RAN. First, an agent in agentic AI-RAN does not only process network data passively as a learning model but also perceives, reasons, plans, and acts upon the radio environment autonomously, which fundamentally reshapes the way RAN functions are orchestrated. Second, an agent in agentic AI-RAN must jointly handle multi-dimensional resources including communication, sensing, computation, and control rather than optimizing along a single dimension as done in standard AI-RAN systems, since the agentic capability emerges only from the co-deployment of these heterogeneous functionalities. Due to these unique features, in this paper, we present an overview on the fundamentals and key challenges in the design of agentic AI-RAN. We first introduce the concept of agentic AI-RAN and provide its basic framework. Then, we focus on agentic AI-RAN techniques in two areas: the co-deployment mechanism of agents over the RAN infrastructure and the agentic interaction with the radio environment. For each area, we identify the main problems and challenges. Then, we will provide a comprehensive treatment of these problems. Two use cases, one based on Py5chesim and another one based on Bubble RAN by NVIDIA, are demonstrated as well. Finally, we point out future research in the field of agentic AI-RAN. In a nutshell, this article provides a holistic set of guidelines on how to design agentic AI-RAN systems over real-world wireless communication networks, aiming to stimulate more innovative research on this emerging paradigm.
Agentic artificial intelligence (Agentic AI) represents a new framework for future sixth-generation (6G) integrated sensing and communications (ISAC) systems. Agentic AI can shift network management from passive, model-driven optimization to active, goal-driven autonomy. By reasoning, planning, and executing complex actions, Agentic AI aims to handle the high complexity of 6G networks. This article surveys this emerging topic, focusing on the basic principles of Agentic AI, reviewing its key applications in ISAC, and identifying its critical challenges. We also briefly discuss how large artificial intelligence models (LAMs) may support agentic reasoning and intent understanding in practical ISAC deployments. Furthermore, six promising research directions are outlined.
In this paper, we first review evolution of radio resource management in mobile networks from traditional statistical and numerical methods towards machine learning (ML) and artificial intelligence (AI) driven approaches. We discuss these methods especially from perspective of joint optimization of multiple radio resource management parameters and settings. However, an application of ML in a common way, i.e., each ML model dealing with one radio resource management parameter, leads to an accumulation of natural inaccuracies due to sub-optimality of individual stand-alone models. Multiple radio resource management parameters can also be handled via big AI models or multi-task learning (MTL) consisting in a large model with multiple outputs, each corresponding to one resource management parameter. Nevertheless, the big AI models and MTL require large datasets for training and, like common ML models, struggle to capture mutual relations and dependencies among individual optimized parameters. Thus, we outline an approach for joint optimization of multiple radio resource management parameters consisting in a coordination of dedicated but tightly integrated small ML models. The coordination of the small ML models is facilitated via mutually exchanged feedback tailored to reflect network performance and allowing to capture mutual dependencies among the optimized radio resource management parameters. We demonstrate benefits of the coordinated small ML models against state-of-the-art concepts in terms of increased network performance and stable training.
Reconfigurable apertures (RAs) modify the electromagnetic interface through radiator relocation, radiation-support selection, programmable scattering, guided-wave extraction, or continuous current shaping. Yet most RA designs keep the common carrier fixed once the aperture state has been chosen. This paper studies when a narrow carrier adjustment remains useful after aperture reconfiguration and shows that the answer depends on the residual carrier sensitivity of the couplings that determine performance. It introduces a unified aperture-dependent propagation model, identifies four operating conditions for non-negligible movable signals (MS) gains, and compares representative RA platforms according to how aperture and carrier control should be coordinated. It then gives a block-level implementation boundary and uses reduced-order MA and PASS tests to illustrate the resulting operating regimes. The main message is that MS becomes effective only when a carrier-visible residual remains after aperture reconfiguration, its main benefit often comes from suppressing limiting couplings rather than uniformly strengthening desired ones, and its gain must survive probing, synchronization, and front-end costs.
Low-altitude wireless networks (LAWNs) are expected to support mission-critical services in future sixth-generation systems, with tightly integrated communication, sensing, computation, and control. Beyond task-specific intelligence, emerging applications increasingly require autonomous behavior, explicit handling of mission intent, and coordinated decision-making across distributed agents. In this article, we introduce the agentic-integrated LAWNs, where autonomous agents and artificial intelligence (AI) models cooperate to translate mission intent into closed-loop network operation. We first present the architectural foundations of agentic-integrated LAWNs. This architecture consists of coupled aerial and edge segments organized into a three-layer framework, which integrates the basic functional, decision and cognition, and collaboration layers. We then discuss key enabling technologies for functional learning and adaptation, multi-agent coordination, and knowledge-driven enhancement. We further provide a case study on coordinated drone navigation within agentic-integrated LAWNs, highlighting improvements in safety and coordination efficiency. Finally, the article outlines future research directions to advance LAWNs.
Artificial Intelligence (AI) has become a key enabler of the envisioned 6G intelligent endogenous architecture, significantly enhancing the autonomous capabilities of communication networks. However, the deep integration of AI into 6G presents considerable challenges in ensuring trustworthiness throughout its lifecycle, including special concerns related to data quality, data and model privacy, model security, model output explainability, and so on. In response, major communication standardization bodies such as 3GPP, ETSI, and ITU-T have proposed several guidelines towards AI trust in communication networks. Nevertheless, a comprehensive review on these standardization efforts across the full lifecycle of AI models is still missing, hindering the development and harmonization of relevant standards. To fill this gap, this paper reviews the existing standards about AI trust in communication networks. We begin by discussing trust-related issues at each phase of an AI model’s lifecycle and proposing corresponding evaluation criteria on establishing trust. We then systematically review existing standards related to AI trust in communication networks, using the proposed criteria as a measure to judge their quality. Based on the review, we identify open issues and suggest future directions, aiming to provide a systematic guidance for advancing the standardization of AI trust.
Semantic communication provides a new paradigm for wireless communications by focusing on task-relevant information rather than bit-level reconstruction. However, most existing semantic communication methods rely on end-to-end learning, making semantic reliability difficult to explicitly model and control. To address this issue, this paper proposes a semantic–physical layer co-design framework based on semantic importance (SemIm) and non-binary polar coding (NB-PC). A dual-scale SemIm evaluation mechanism is first introduced to characterize SemIm from both global semantic statistics and sample-dependent feature saliency. Based on the resulting SemIm ranking, semantic features are mapped onto NB-PC subchannels according to their reliability ordering, enabling semantic-aware unequal error protection (UEP) at the channel-coding level. Furthermore, a block-wise semantic-aware transmission framework is developed to establish an interpretable interface between semantic perception and physical-layer reliability allocation. Simulation results demonstrate that the proposed framework achieves consistent improvements in semantic reconstruction quality and task performance compared with conventional joint source–channel coding (JSCC) and separation-based baselines. In addition, a modular JSCC hardware baseline platform is constructed to verify link stability and provide an engineering testbed for future integration of semantic-aware NB-PC modules.
Emerging wireless networks such as the sixth-generation (6G) mobile systems are expected to operate over highly heterogeneous infrastructures with rapidly evolving service demands, thereby increasing the dependence on large-scale, constraint-coupled, cross-layer optimization for network design and operation. However, the trend leads to a fundamental bottleneck: Converting high-level intents into mathematically consistent formulations, implementable algorithms, and reproducible simulations remains predominantly human-driven, time-intensive, and error-prone. While large language models (LLMs) provide a natural-language interface for intent interpretation and rapid prototyping, monolithic LLM-based pipelines are often limited by insufficient domain grounding, weak constraint awareness, and a lack of execution-based verification and self-correction. These limitations motivate a shift towards agentic AI, where problem-solving is realized through iterative decomposition, explicit planning, tool-integrated execution, and reflection driven by feedback. In this paper, we present ComAgent, a multi-LLM based agentic AI framework that coordinates specialized agents for literature searching, planning, coding, and solution scoring within a closed-loop Perception–Planning–Action–Reflection cycle, turning user intents into solver-ready formulations and reproducible simulation pipelines while continuously self-correcting logical and feasibility errors. We demonstrate the efficacy of ComAgent through two distinct evaluations. In a non-trivial beamforming optimization case study, ComAgent autonomously perceives the problem, designs an algorithm, and generates solutions that achieve a performance comparable to that of expert-designed baselines. Furthermore, on a diverse set of generic wireless tasks, ComAgent outperforms monolithic LLMs. The numerical results demonstrate the potential of the proposed agentic AI framework for various emerging wireless networks.
The transition toward 6G calls for a shift from static connectivity to mobile edge general intelligence, where the network evolves into an autonomous system capable of perceiving and reasoning about its environment. This article explores an agentified Open Radio Access Network (O-RAN) architecture that integrates embodied Artificial Intelligence (AI) agents directly within disaggregated network components. Rather than treating intelligence as a standalone optimization module, the investigated framework adopts an embodied AI perspective, enabling agents to fuse RF telemetry with visual sensing to develop improved situational awareness of the physical world. We also explored a joint communication and compute slicing mechanism that models radio spectrum and computational resources as a unified resource fabric, coordinated through a hierarchical Brain-Reflex control loop. By leveraging Large Language Models (LLM) to express high-level intent and distributed heuristic agents for real-time tactical decisions, the network dynamically balances transmission effort with cognitive processing. Evaluation is performed using a live, AI-in-the-loop digital twin of a mission-critical UAV search-and-rescue scenario. By dynamically interfacing the NS-3 network simulator with a live LLM serving framework for real-time strategic orchestration, we illustrate the architectural viability of this vision through a proof-of-concept evaluation, showing indicative improvements in time-to-discovery and service reliability under a heuristic control baseline. Furthermore, we identify critical open research challenges and outline future research directions to guide the evolution of perceptive 6G systems.
The proliferation of connected and autonomous vehicles within the Internet of Vehicles (IoV) is generating a data deluge, straining current wireless networks which primarily focus on raw data transmission. This creates a semantic gap between data volume and actionable intelligence, hindering the realization of truly intelligent, efficient, and reliable vehicular services for the sixth generation of mobile networks (6G). To bridge this gap, this article proposes a novel architectural vision: Intelligent Semantic Agents empowered IoV (ISA-IoV). ISA-IoV deploys a hierarchy of Semantic Agents, software entities powered by foundation models, across cloud, edge, and end devices to collaboratively perform semantic perception, reasoning, and communication. We define the scalability and adaptability requirements for ISA-IoV and introduce three unique Key Performance Indicators to guide its development. The article further surveys key enabling technologies, discusses major implementation challenges, and outlines promising future directions. By integrating agentic artificial intelligence with semantic communication over a Space-Air-Ground Integrated Network, ISA-IoV presents a concrete roadmap towards semantic-native and intelligence-everywhere 6G vehicular networks.
The groundbreaking development of generative artificial intelligence (AI) is rapidly boosting the ability to generate content such as images and videos, reshaping communication paradigms. This article introduces generative communications (GenCom), a novel paradigm for 6G networks in which large AI models (LAMs) drive semantic understanding, reasoning, and content generation, embedding these into the communication process. Unlike traditional systems that strictly pursue accurate bit transmission, GenCom enables transmitters to convey only minimal yet sufficient information, while receivers leverage shared generative priors and knowledge bases to synthesize the intended output. Communication is thus redefined as controlled generation rather than data reproduction. We formalize the concept of GenCom, clarify its AI-native and generation-driven properties, and present its core mechanisms. A two-layer GenCom architecture supported by key enabling technologies is proposed, and analysis of four representative application scenarios demonstrates that GenCom offers ultra-efficient transmission, semantic-level robustness, and new network functions. Finally, we outline future research directions, including foundational theory and real-time processing, highlighting a promising pathway toward 6G networks.
Reinforcement learning (RL) is a promising AI algorithm supporting latency-sensitive applications in 6G to enable ultra-reliable low-latency communication (URLLC). This work focuses on RL-assisted real-time resource allocation in mobile communications. Location and mobility of cellular user equipments (CUEs) are critical information for rapid convergence of resource allocation, in which deep Q-network (DQN) algorithm is useful to facilitate resource allocation to maximize sum rate of a cellular network, whose up-link channels are shared by device-to-device (D2D) UEs (DUEs). To ensure signal-to-interference-plus-noise ratios (SINRs) of both CUEs and DUEs, CUE’s guaranteed and prohibited areas are defined to align CUE locations with DUEs’ transmit powers for precise power allocation. The simulation results show that the proposed RL approach offers a fast tracking convergence with a balanced quality of services (QoSs) for both CUEs and DUEs. The application of the DQN-based algorithm can be extended beyond D2D communications to support low-latency location/mobility-aware communications.
The convergence of autonomous aerial vehicles and ground-based vehicles is driving the emergence of drone-vehicle networks, which promise transformative capabilities in autonomous mobility, emergency response, environmental monitoring, and smart city applications. To meet the stringent demands for high data rates and ultra-low latency in these dynamic environments, millimeter-wave (mmWave) communication has been identified as a key enabler. This article provides a comprehensive exploration of the mmWave-integrated drone-vehicle networks paradigm. We first conduct a bibliometric analysis to map the evolution and current trends in this emerging field, identifying key technologies and thematic research clusters. We then examine critical challenges that hinder dependable mmWave communication in such networks, focusing on three interdependent areas: accurate channel measurement and modeling, reliable communication link maintenance, and robust security mechanisms. To address these challenges, we propose a set of forward-looking research directions, including computational intelligence-based channel modeling, AI-empowered autonomous resource scheduling, adaptive and lightweight security frameworks, and the integration of large language models for enhanced network management. This article serves as a roadmap to guide future research and technological development in this promising area.
The semantic and goal-oriented communication paradigm is a fundamental shift in the design of next-generation 6G networks, aiming to support an increasingly connected, intelligent, and sustainable digital ecosystem. This paper provides a comprehensive overview of the architecture and operational framework developed within the 6G-GOALS project, with particular emphasis on its core design pillars: ultra-low latency, semantic communications, AI-native integration, energy efficiency, and enhanced network resilience. The paper details the novel architectural components of the 6G-GOALS framework, which builds on and extends the O-RAN architecture by integrating semantic-aware entities and protocols. Key enablers, including distributed AI, real-time, goal-driven decision-making, and adaptive orchestration of network functions, are presented, illustrating how these capabilities work in concert to realize fully semantic-aware, intelligent, and self-adaptive network operations capable of meeting the demands of next-generation connectivity. Finally, we provide an evaluation of the architecture’s potential to meet key performance indicators (KPIs), its alignment with sustainability goals, and its readiness for the evolving digital ecosystem. This analysis is intended to serve as a foundational reference for researchers and industry stakeholders working to advance the 6G vision.
