PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey of Beam Management for mmWave and THz Communications Towards 6G

Qing Xue, Chengwang Ji, Shaodan Ma, Jiajia Guo, Yongjun Xu, Qianbin Chen, Wei Zhang

arXiv:2308.02135v2cs.ITeess.SY

TL;DR

MmWave and THz beam management must handle narrow-beam measurement overhead, fluctuating channels, mobility, and dynamic environments. This paper surveys beam-management procedures and research using AI, RIS, and ISAC, including collaborative learning and sensing-assisted methods. It synthesizes reported advances while identifying unresolved issues such as limited NLoS handling, multi-user sensing, privacy, security, and heterogeneous-sensor coordination.

  • Problem

    Narrow beams, fluctuating channels, user movement, and changing environments make robust mmWave/THz beam acquisition and tracking difficult, especially for mobile scenarios.

  • Method

    The paper conducts a comprehensive survey of mmWave/THz beam management across conventional procedures and AI-, RIS-, and ISAC-enabled approaches.

  • Results

    The survey identifies beam-management advances spanning supervised and reinforcement learning, collaborative training, sensing-assisted tracking, and RIS-assisted beam alignment.

  • Takeaways & Limitations

    AI, sensing, and RIS provide surveyed directions for adapting beam management to dynamic mmWave/THz environments and reducing measurement or training demands.

  • Takeaways & Limitations

    Existing sensing-based methods have limited capability for NLoS targets and require further adaptation to multi-user environments.

Abstract

from arXiv · show

Communication in millimeter wave (mmWave) and even terahertz (THz) frequency bands is ushering in a new era of wireless communications. Beam management, namely initial access and beam tracking, has been recognized as an essential technique to ensure robust mmWave/THz communications, especially for mobile scenarios. However, narrow beams at higher carrier frequency lead to huge beam measurement overhead, which has a negative impact on beam acquisition and tracking. In addition, the beam management process is further complicated by the fluctuation of mmWave/THz channels, the random movement patterns of users, and the dynamic changes in the environment. For mmWave and THz communications toward 6G, we have witnessed a substantial increase in research and industrial attention on artificial intelligence (AI), reconfigurable intelligent surface (RIS), and integrated sensing and communications (ISAC). The introduction of these enabling technologies presents both open opportunities and unique challenges for beam management. In this paper, we present a comprehensive survey on mmWave and THz beam management. Further, we give some insights on technical challenges and future research directions in this promising area.

I. INTRODUCTION

mmWave and THz communications support demanding future wireless services but depend on efficient beam management because narrow beams, fluctuating channels, mobility, and blockage complicate alignment and tracking. AI, RIS, and ISAC are emerging as key technologies for addressing these challenges in 6G.

  • mmWave operates around 30–100 GHz and is promising for indoor and outdoor communications, but may not satisfy future wireless data-traffic growth.
  • 6G targets autonomous, ultra-large-scale, highly dynamic, and intelligent services with stringent data-rate, latency, mobility, reliability, and connection-density requirements.Representative targets include peak data rates of at least 1 Tbit/s, over-the-air latency of 0.01–0.1 ms, and mobility of at least 1000 km/h.
  • Narrow beams generated by large antenna arrays compensate path loss but make mmWave/THz systems heavily reliant on rapid beam training, alignment, and tracking to avoid misalignment or link failure.
  • Rapid channel fluctuation and frequent beam misalignment create key feasibility challenges, requiring efficient initial-access and tracking strategies that periodically identify suitable beam pairs.
  • AI can adapt beam parameters to real-time network dynamics, while RIS can mitigate blockage and expand coverage but complicates beam coordination.
  • ISAC is expected to support beam management in practical dynamic environments where wireless channels vary rapidly.

C. Our Survey Scope and Contributions

The survey reviews emerging beam-management research for mmWave and THz systems, emphasizing AI, sensing, RIS, collaborative learning, predictive ISAC methods, and RIS-assisted architectures. It also examines more than 150 papers, identifies open challenges, and organizes prior standards, state-of-the-art approaches, and future directions.

  • Survey scope: The survey comprehensively examines beam management for both mmWave and THz communications, integrating AI, sensing, and RIS perspectives.
  • Collaborative beam management: It extends prior single-agent reviews by analyzing collaborative beam management across multiple agents and tasks using FL, TL, and split learning.
  • ISAC-enabled beam management: The paper presents the first review of predictive ISAC beam-management solutions, covering sensing-enabled approaches for precise alignment and tracking in high-mobility scenarios.
  • RIS-assisted systems: It provides a pioneering review of beam management in RIS-assisted mmWave and THz systems, including AI-driven and sensing-aided methods and their strengths and limitations.
  • Literature synthesis: The review examines over 150 scholarly papers and synthesizes lessons learned, persistent challenges, and potential future research avenues.
  • Paper organization: The paper is organized around standardized beam-management procedures, emerging AI-, RIS-, and ISAC-based approaches, future work, and conclusions.

A. Beam Management in 3GPP NR

In 3GPP NR, beam management covers beam-related PHY/MAC operations for acquiring, maintaining, measuring, reporting, and recovering directional links. The section also identifies open standardization needs and extensions for AI/ML, non-terrestrial networks, and IAB.

  • 3GPP NR procedure: 3GPP NR beam management includes beam sweeping, measurement, reporting, and determination operations.Sweeping covers spatial areas; measurement evaluates beamformed signals; reporting communicates beam information; determination selects transmit or receive beams.
  • 3GPP NR procedure: Beam measurement commonly uses exhaustive searches of reference-signal beams to derive beam quality from RSRP, with SINR measurement also supported in 5G Release 16.The search follows prespecified intervals and directions.
  • Beam recovery: After beam failure, NR recovery proceeds through failure detection, candidate-beam identification, recovery-request transmission, and response monitoring.The UE continuously monitors whether a beam-failure condition is triggered before beginning recovery.
  • Initial access and tracking: Beam management supports directional initial access for idle UEs and beam tracking that updates steering direction and beam shape as channels change during movement.The section distinguishes initial beam establishment from beam tracking or beam maintenance.
  • Open NR issues: Further work is needed on beam-determination, measurement, reporting, and sweeping requirements because aspects such as periods, accuracy, delay, and physical-layer design remain unsettled.The cited standardization discussion also notes unresolved requirements for selecting beams and radio-resource management.
  • Future NR enhancements: NR beam management is being extended toward AI/ML beam prediction, non-terrestrial mobility, and faster IAB beam switching, coordination, and recovery.AI/ML targets time- or spatial-domain prediction to reduce overhead and latency, while satellite and IAB settings require procedure-specific adaptations.

B. Beam Management in IEEE

IEEE mmWave WLAN and WPAN standards use beamforming training to select directional beams through coarse-to-fine procedures. Existing surveys cover foundational beamforming, 5G standardization, and AI-related work, while newer 6G, RIS, and ISAC developments remain incompletely covered.

  • IEEE beamforming training: IEEE 802.11ad, 802.11ay, and 802.15.3c use beamforming training to select beamforming and combining vectors from predefined codebooks without explicit channel estimation.The two-stage process consists of coarse-grained sector training followed by fine-grained beam refinement.
  • IEEE 802.11ad: 802.11ad splits beamforming training into sector-level sweeping and an optional beam refinement protocol, followed by MIMO setup, training, feedback, and selection subphases.The initiator begins sector sweeping while the responder provides feedback on the highest-quality received sector.
  • IEEE 802.15.3c: IEEE 802.15.3c narrows the spatial area through sector-level training before slicing the best sectors into high-resolution beams for beam-level training.This two-level procedure searches first with low-resolution patterns and then with higher-resolution patterns.
  • Related surveys: Earlier surveys addressed mmWave beamforming architectures, signal processing, RF design, and 5G NR measurement and management procedures.Covered topics include analog, digital, and hybrid beamforming, plus initial-access and tracking measurement frameworks.
  • Survey gap: Existing surveys generally predate emerging 6G technologies or focus on narrower topics, leaving recent AI, RIS-assisted, and ISAC beam-management work incompletely surveyed.The review positions its contribution as covering both mmWave and THz systems and integrating these emerging technology areas.
  • AI-based beam management: AI methods are being studied for beamforming training and tracking, while federated and transfer learning offer distributed-learning and training-speed opportunities.The cited work describes AI-based beam prediction and discusses FL and TL as future trends for Wi-Fi beamforming.
  • RIS and ISAC: RIS can help overcome blockage, but RIS-assisted beam management must coordinate transmission across two hops and therefore complicates system management.Sensing-assisted beam management is also emerging for integrated communication and sensing systems.

III. ENABLING-TECHNOLOGY-BASED 6G BEAM MANAGEMENT: STATE-OF-THE-ART

The survey organizes 6G mmWave and THz beam-management research around AI, sensing-based ISAC, and RIS-assisted methods, while discussing their benefits and challenges.

  • III. Enabling-Technology-Based 6G Beam Management: State-of-the-Art: The survey classifies beam-management research into AI-based, sensing-based ISAC, and RIS-assisted methods, and discusses their possible combination.It presents benefits of AI, ISAC, and RIS before reviewing mechanisms organized by these enabling technologies.
  • AI as an Enabler: AI can model complex nonlinear relationships, adapt to dynamic environments, and use surrounding information to detect blockages or identify bypassing beam pairs.These capabilities address limitations of idealized mathematical models and high-overhead beam resweeping.
  • ISAC as an Enabler: Sensing provides environment awareness that can support beam tracking, blockage prediction, proactive handoff, and reduced blind beam-training overhead.Relevant information includes scatterer geometry and transmitter or receiver locations and directions.
  • Research Landscape: The classification of existing beam-management algorithms and the roadmap to edge AI frame the survey’s technology-based research landscape.The roadmap notes 3GPP Release 17’s introduction of federated learning in global 5G communications and the first FL standard approved by IEEE Std 3652.1TM-2020.
  • RIS as an Enabler: RIS can passively reflect signals toward desired directions, reducing energy consumption relative to active antenna arrays and helping restore blocked direct links.RIS-assisted systems can use virtual links when the AP/BS-to-user path is infeasible because of pathloss or blockage.

B. AI-Empowered Beam Management

AI-empowered beam management uses learning-based methods to reduce search overhead and improve prediction in dynamic mmWave and THz settings. The survey reviews supervised, reinforcement, collaborative, and parallel learning approaches, while noting data, experience, and real-world deployment challenges.

  • Overview: AI beam-management methods can extract information from previous results to limit subsequent search areas, potentially reducing overhead and improving prediction accuracy.The survey groups reported approaches into isolated and collaborative training modes.
  • Isolated Training: Isolated training includes supervised and reinforcement learning, with supervised learning mapping feature vectors to labels from labeled historical data.Supervised learning is reported as the most frequently used AI technique for mmWave and THz beam management.
  • Supervised Learning: FCNN, LSTM, CNN, and conventional machine-learning methods are applied to beam alignment, tracking, beam selection, and beam-quality prediction.Examples use CNNs with partial or wide-beam measurements and LSTM models for temporal channel or beam evolution.
  • Collaborative Training: Collaborative training addresses communication overhead, privacy concerns, limited edge datasets, and generalization through federated learning, split learning, and transfer learning.Transfer learning is described as knowledge transfer that improves training efficiency when data are lacking, subject to source-target commonalities.
  • Parallel Learning: Parallel learning distributes model-update computation or experience collection across learners, reducing training time while maintaining performance in reported reinforcement-learning frameworks.Examples include parallel DRL on a GPU and vehicles acting as active learners for beam association or handover in high-mobility mmWave networks.
  • Summary: The survey reports strong progress from CNN, LSTM, MAB, and DRL methods, but identifies training-sample costs, experience requirements, and real-world adoption challenges.Supervised methods require extensive and frequently updated samples, whereas reinforcement learning obtains data through environment interaction and feedback.

C. Beam Management for ISAC Systems

ISAC-assisted beam management uses sensing information from radar, communication signals, hybrid sensing, or dedicated sensors to improve beam selection, tracking, and prediction. The survey identifies gains in communication rate, tracking precision, and training overhead, while highlighting limitations involving NLoS targets, multi-user settings, resource allocation, and multi-beam implementation.

  • ISAC sensing and beam management: Sensing can provide transmitter/receiver locations and environmental geometry to support beam tracking, blockage prediction, proactive handoff, and beam-direction selection.These capabilities can reduce blind beam training and help avoid interference with adjacent users.
  • ISAC sensing and beam management: ISAC beam management spans radar sensing, communication signal sensing, hybrid signal sensing, and dedicated sensors.The survey presents these four sensing categories as the main technology types and illustrates their typical use cases.
  • Challenges and open issues: Multi-beam ISAC separates communication and sensing directions, yet full-duplex MIMO implementations must handle substantial transmitter-receiver leakage and inter-beam interference.Co-located or integrated radar and communication systems also require cooperative operation when their signals overlap in time or frequency.
  • Communication signal sensing: 5G NR sensing reuses DMRS, SRS, SSB, CSI-RS, and PRS signals for channel sensing, beam prediction, blockage perception, user tracking, and beam or cell switching.The survey notes that practical results remain limited despite many feasibility studies.
  • Semantic sensing: Semantic information from sparse-channel parameters and environmental features may enable direct beam prediction using radar, GPS, Wi-Fi, cameras, or semantic segmentation.The paper distinguishes parameter semantics, such as AOA, AOD, loss, and delay, from environment semantics describing key scatterers.
  • Challenges and open issues: Sensing integration can increase communication rate, improve beam-tracking precision, and reduce extensive training overhead, but existing methods mainly address LoS links and need adaptation to multi-user environments.Multi-target radar positioning can lose precision, while communication-signal sensing requires careful frequency-, time-, and spatial-domain resource allocation.

D. Beam Management for RIS-Enhanced Systems

RIS-assisted mmWave/THz links require beam management across a dual-hop AP/BS–RIS–user architecture, making beam training costly and technically complex. The survey covers sweeping, hierarchical, multi-beam, advanced, AI-based, and sensing-assisted methods, emphasizing opportunities from AI and computer vision alongside persistent overhead, interference, and cost trade-offs.

  • Challenges and conventional methods: RIS-assisted beam management must coordinate the AP/BS–RIS and RIS–user hops because passive RIS elements cannot autonomously generate or decode beams.This dual-hop structure makes sophisticated beam management essential for realizing passive beamforming gains.
  • Challenges and conventional methods: Beam-sweeping methods use finite active or passive beamforming codebooks without requiring complex channel estimation, making them beneficial for large RISs.The method must account for the cascaded link rather than directly reusing an AP/BS–user beam-training procedure.
  • Challenges and conventional methods: Exhaustive search evaluates all possible beam tuples, whereas hierarchical multi-resolution codebooks reduce complexity by testing wider beams before narrower beams.For cascaded links, exhaustive search spans the AP/BS, RIS, and user training codebooks; hierarchical search trades search cost against early-stage beamforming gain.
  • Challenges and conventional methods: Multi-beam training reduces single-beam training time for large RISs with pencil-like beams, but practical systems face inter-beam interference, reduced passive beamforming effectiveness, and larger codebooks.These drawbacks limit straightforward deployment of multi-beam search.
  • AI-assisted methods: AI-based RIS methods include DRL, DNN, RNN, genetic algorithms, and other learning approaches for phase configuration, beamforming, beam alignment, beam tracking, and beam prediction.Reported applications include multi-user downlink optimization, indoor positioning, mobile beam steering, and flying-RIS THz drone communications.
  • AI- and sensing-assisted methods: AI/ML-based RIS approaches can achieve comparable performance to conventional methods while reducing computational complexity, and sensing-assisted RIS beam management can exploit user locations to expedite channel estimation and beamforming.The survey also identifies computer vision as a possible way to design RIS reflection coefficients while reducing training overhead and additional feedback links.
  • Open directions: AI algorithms suit mobile RIS communications requiring rapid beam steering, but RIS-assisted ISAC beam management remains underdeveloped and semi-passive RIS architectures increase deployment cost.The survey presents AI, RIS, and ISAC integration as a promising direction requiring further investigation.

E. Lessons Learned: Summary and Insights

The survey’s lessons section frames AI-based beam management as a distinct research area and points toward summarizing its implications for future wireless systems. The supplied passages introduce the lessons but do not state their substantive content.

  • Lessons learned: Table VIII is identified as summarizing key characteristics of existing mmWave and THz beam management techniques before the lessons are presented.The supplied passage does not provide the table’s characteristics or comparisons.
  • Lessons learned: The paper explicitly introduces lessons learned for AI-based beam management in wireless communication systems.The supplied passage does not enumerate the lessons themselves.

1) Adaptability and Flexibility:

Adaptive beam management uses changing environmental, sensing, and RIS information to adjust beamforming, while accounting for overhead and network coordination.

  • 1) Adaptability and Flexibility:: AI-based beam management adapts beamforming parameters dynamically to mobility, interference, multipath propagation, and changing environmental conditions.Its stated goal is real-time performance optimization.
  • 1) Adaptability and Flexibility:: Surveyed beam management techniques differ in their general overhead, as summarized in Table VIII.The supplied table material provides the comparison heading but no individual overhead values.
  • 1) Adaptability and Flexibility:: Sensing-assisted beam management combines sensor information to improve environmental awareness, dynamic adaptation, and beamforming robustness.The surveyed sensing sources include radar, LiDAR, and environmental monitors.
  • 1) Adaptability and Flexibility:: RIS-assisted beam management dynamically steers transmitted or reflected beams using real-time feedback to adapt to channel conditions and user requirements.The approach also considers interference mitigation and broader network-control integration.

8) Real-World Deployment Challenges:

Real-world deployment requires integrating AI, sensing, and RIS capabilities while addressing hardware, coordination, resource-allocation, and scalability concerns.

  • 8) Real-World Deployment Challenges:: Practical RIS deployment must address hardware complexity, calibration, and scalability across diverse deployment scenarios.These constraints are presented as obstacles to viable RIS-enabled beam management.
  • 8) Real-World Deployment Challenges:: Integrated AI, sensing, and RIS beam management combines AI decision-making, environmental awareness, and RIS reconfigurability for dynamic wireless environments.The paper associates this integration with enhanced spectral efficiency, improved coverage, and better quality of service.
  • 8) Real-World Deployment Challenges:: AI can allocate beam, spectrum, and power resources using sensing-assisted inputs, while RIS controllers adjust reflective properties for resource optimization.The proposed integration also establishes feedback and learning among components.

IV. CHALLENGES AND OPEN ISSUES

The paper discusses technical concerns that arise when implementing 6G mmWave and THz communications with AI, sensing, and RIS.

  • IV. CHALLENGES AND OPEN ISSUES: Efficient 6G mmWave and THz communication using AI, sensing, and RIS raises several potential technical concerns for beam management.The passage introduces the subsequent discussion of these concerns without specifying them here.

A. AI-Empowered 6G Framework

The AI-empowered 6G framework considers denser and more heterogeneous networks with stricter requirements, emphasizing edge collaboration, sensing-based localization, generalization, and lifecycle management.

  • A. AI-Empowered 6G Framework: 6G is expected to involve denser small-BS/AP and user deployments, greater technology and application heterogeneity, and stricter performance requirements than 5G.These characteristics motivate deeper research on AI for 6G wireless networks.
  • A. AI-Empowered 6G Framework: Collaborative edge AI executes training and inference at network edges to reduce computing, communication, storage, and engineering resources while providing low latency and high reliability.The passage also notes that real-world mobility environments are often partially observable.
  • A. AI-Empowered 6G Framework: Sensing AI analyzes sensor-collected physical-environment data for tasks including detection, localization, and motion or activity recognition.Localization of network nodes and objects is identified as important for efficient beam management.
  • A. AI-Empowered 6G Framework: Model generalization is challenging because optimal mmWave and THz beam angles depend strongly on propagation environments, while collecting large local training datasets can be costly or impossible.Each node may benefit from environment-specific data, but obtaining enough such data is constrained.
  • A. AI-Empowered 6G Framework: AI/ML lifecycle management for beam management remains insufficiently addressed in concrete scenarios despite preliminary progress in 3GPP standardization.The paper identifies this as a topic requiring future attention.

B. ISAC-Enabled 6G Framework

ISAC can support beam management by sensing blockages, positioning users, and improving environmental awareness. The section distinguishes radar-type integration approaches and highlights collaboration, multimodal fusion, federated sensing, and RIS-assisted sensing as research directions.

  • ISAC sensing can detect blockages and position users, providing instructions for faster beam alignment and tracking.
  • Radar-type ISAC implementations are categorized as loose or tight integration according to how communication and sensing hardware and waveforms are shared.
  • Collaborative Sensing: Collaborative sensing extends coverage beyond single-node sensing and supports environment sensing, 3D positioning, imaging, and real-time beam management.
  • Collaborative Sensing: Collaborative sensing introduces high cost, structural complexity, synchronization difficulty, and unresolved coordination challenges for mobile or heterogeneous sensors.
  • Collaborative Sensing: AI is identified for multimodal data fusion, target recognition, and decision-making, while federated sensing addresses personalization, privacy, and communication efficiency.
  • Supported by RIS: RIS-assisted ISAC may improve sensing coverage, accuracy, and resolution by adding an alternative line-of-sight observation path, although research remains limited.

C. RIS-Enhanced 6G Framework

RIS-enhanced beam management must address the overhead and coordination demands of two-hop, passive-surface-assisted links. Open issues include mobility, multiple RISs and users, MIMO operation, and architectures that add active sensing or relaying capability.

  • Beam management becomes more costly in RIS-assisted two-hop systems, motivating more efficient mechanisms for mobile mmWave/THz communications.
  • Support for Mobility: RIS passivity prevents direct signal sensing, complicating beam training and requiring coordination between passive RIS training and active AP/BS beam training under mobility.
  • Multi-Cell Multi-RIS: Multi-cell multi-RIS deployments require users to jointly determine and coordinate beams across dense cells and multiple reflecting surfaces.
  • Multi-User and MIMO: RIS beam management must extend beyond predominantly single-user studies to multi-user MIMO settings with shared or partitioned reflecting beams.
  • Multi-Cell Multi-RIS: Multiple RISs can provide path diversity and enable cross-verification of estimated path parameters, but their integration creates additional management challenges.
  • Bring in Active Ability: Semi-passive, relay-type, and active RIS designs are being explored to overcome beam-management limitations caused by fully passive surfaces.

D. THz Beam Management Towards 6G

THz beam management inherits mmWave challenges but becomes more demanding because higher propagation loss requires narrower beams and because extremely large arrays and RISs create near-field effects. Existing far-field-oriented methods therefore leave important gaps.

  • THz communications require narrower beams than mmWave systems to compensate for higher propagation loss, further complicating beam management.
  • Extremely large-scale THz MIMO and RIS deployments introduce near-field effects that make beam management more challenging.
  • Near-field codebooks must consider both transmitter–receiver direction and distance, producing extremely high training overhead compared with far-field design.
  • Most current contributions assume far-field propagation, while far-field codebooks do not match near-field channels and can cause serious performance loss.

V. CONCLUSIONS

The survey addresses beam misalignment in rapidly varying mmWave/THz channels by comprehensively reviewing beam-management technologies across AI, ISAC, and RIS paradigms. It classifies existing approaches, discusses their extensibility and limitations, and identifies future concerns.

  • Rapid channel fluctuation and frequent beam misalignment make efficient initial access and tracking essential for mmWave/THz communication systems.
  • The survey comprehensively reviews state-of-the-art beam-management technologies for AI-empowered, ISAC-enabled, and RIS-enhanced 6G networks.
  • AI-based beam-management papers are classified by training mode, AI paradigm, and model, with their key contributions summarized.
  • ISAC beam-management approaches are organized into radar sensing, communication-signal sensing, hybrid sensing, and dedicated-sensor categories.
  • For highly mobile V2X systems, sensing-aided techniques are envisioned as promising candidates for real-time vehicle detection and tracking that facilitate beam management.
  • The survey concludes that AI-driven and sensing-aided frameworks will play an essential role in implementing RIS-enhanced networks, while identifying further technical concerns.
Loading 2308.02135v2…