PaperlayerView exact revision on arXiv

Source-linked AI summary

The Integrated Sensing and Communication Revolution for 6G: Vision, Techniques, and Applications

Nuria González-Prelcic, Musa Furkan Keskin, Ossi Kaltiokallio, Mikko Valkama, Davide Dardari, Xiao Shen, Yuan Shen, Murat Bayraktar, Henk Wymeersch

arXiv:2405.01816v1eess.SP

TL;DR

Future 6G networks need to integrate sensing and communication to support sensing services and resilience in dynamic environments. This paper presents an ISAC vision and surveys enabling methods, with simulations examining their performance in practical scenarios.

  • Problem

    6G networks require integrated sensing and communication to connect physical and radio-frequency environments while supporting sensing services and resilience.

    Source: §VIII · p. 87

  • Method

    The paper presents an ISAC cellular-network vision and surveys signal-processing, optimization, and machine-learning methods for network sensing and sensing-aided communication.

    Source: §VIII · p. 87

  • Results

    The paper evaluates bistatic, multistatic, monostatic, positioning, and sensing-aided communication technologies and reports their performance in practical scenarios.

    Source: §VIII · p. 87

  • Takeaways & Limitations

    Communication and sensing should be considered together when designing future wireless networks.

    Source: §VIII · p. 87

Abstract

from arXiv · show

Future wireless networks will integrate sensing, learning and communication to provide new services beyond communication and to become more resilient. Sensors at the network infrastructure, sensors on the user equipment, and the sensing capability of the communication signal itself provide a new source of data that connects the physical and radio frequency environments. A wireless network that harnesses all these sensing data can not only enable additional sensing services, but also become more resilient to channel-dependent effects like blockage and better support adaptation in dynamic environments as networks reconfigure. In this paper, we provide a vision for integrated sensing and communication (ISAC) networks and an overview of how signal processing, optimization and machine learning techniques can be leveraged to make them a reality in the context of 6G. We also include some examples of the performance of several of these strategies when evaluated using a simulation framework based on a combination of ray tracing measurements and mathematical models that mix the digital and physical worlds.

I. INTRODUCTION … A. ISAC infrastructure

The paper presents ISAC as a communication-centric 6G technology that integrates sensing and communication across shared hardware, waveforms, resources, algorithms, and network data. It surveys the supporting infrastructure and techniques, including joint sensing, wide-aperture technologies, radio SLAM, sensing-assisted communication, distributed data and computation, and learning for sensing and network self-optimization.

I. INTRODUCTION

  • ISAC integrates sensing and communication by exploiting shared hardware, waveforms, signal processing, machine learning, and channels to enable applications beyond conventional communication.The paper frames ISAC as a renewed research area motivated by devices such as connected vehicles and by opportunities that extend beyond conventional integration goals.Source: §I · p. 3
  • The paper adopts a communication-centric ISAC perspective with tight waveform and time-frequency integration, while reviewing how sensing can assist network operation.It surveys frameworks, frequency-dependent features, recent techniques, and advances intended to make ISAC practical in 6G.Source: §I · p. 3

VII. TECHNOLOGIES FOR SENSING ASSISTED COMMUNICATION

  • The surveyed techniques derive sensing information from communication signals, support joint monostatic operation, exploit wide apertures, and include radio SLAM and sensor-aided communication.Signals may come from fronthaul, downlink, uplink, or sidelink; full-duplex operation enables simultaneous communication and radar-like reflection processing for position and velocity, while RIS and distributed MIMO can improve position estimation using additional paths.Source: §VII · p. 4
  • Sensing information such as user position can reduce link configuration and reconfiguration overhead and enable early blockage detection, supporting communication adaptation.These approaches are especially relevant at mmWave and sub-THz frequencies, where large arrays and bandwidths improve angular and delay resolvability and channel sparsity helps map parameters to environmental objects.Source: §VII · pp. 4–5

A. ISAC infrastructure

  • An ISAC network offers sensing and communication as services through heterogeneous terrestrial and non-terrestrial infrastructure, including cellular base stations, RIS, relays, satellites, and unmanned aerial vehicles.RIS can generate favorable reflections between base stations and user equipment to expand coverage and improve sensing accuracy and target-detection probability, particularly at higher frequencies.Source: A. ISAC infrastructure · p. 5
  • Sensing uses UE cameras, radars, and LIDARs, infrastructure co-location for a bird’s-eye view, and reused communication signals with additional hardware and data fusion.More capable hardware, such as a full-duplex base station for monostatic sensing, can support this integration.Source: A. ISAC infrastructure · p. 6
  • Learning combines stored sensor and communication-performance data with distributed training, model updates, and inference across UEs, base stations, and the core network.Data storage, sharing, and computation vary across these components.Source: A. ISAC infrastructure · p. 6
  • ISAC infrastructure is itself smart, using protocols and algorithms that exploit models and data across network elements.The network includes heterogeneous devices and components with differing communication, sensing, computation, and learning capabilities.Source: A. ISAC infrastructure · p. 7
  • Learning supports self-optimization: AI-driven base stations can balance sensing and communication needs, while central processors can reconfigure infrastructure coordination.This learning role extends beyond sensing services to broader network operation.Source: A. ISAC infrastructure · p. 7

B. Spectrum and MIMO technologies … 1) Sensing, Positioning, Localization:

This section examines how spectrum, propagation, array aperture, and MIMO architecture shape 6G sensing performance, including the effects of analog, hybrid, and low-resolution designs. It also clarifies sensing, positioning, and localization terminology and describes how radio-derived information supports object-state estimation and sensor fusion.

1) Spectrum Considerations:

  • Spectrum selection trades coverage, mobility, resolution, data rate, and propagation behavior across FR1, FR2, upper mid-band, and sub-THz bands.FR1 offers low path loss and high mobility but poor delay resolution; FR2 improves resolution with limited coverage; upper mid-band targets a coverage-throughput trade-off; sub-THz offers high resolution and data rates over short ranges.Source: 1) Spectrum Considerations · pp. 7–8
  • Multi-band networks are promising for diverse sensing services, but require further work on transceiver and antenna design, propagation, resource allocation, and geometric evaluation models.Ray tracing or common databases are emphasized over conventional stochastic channel models for evaluating sensing capabilities across bands.Source: 1) Spectrum Considerations · p. 8

2) MIMO architectures:

  • MIMO architecture strongly affects channel-parameter extraction and joint sensing-communication precoder and combiner design because hardware constraints vary across frequency bands.FR1 can support one RF chain per antenna with digital processing, whereas mmWave constraints motivate analog, hybrid, and low-resolution architectures.Source: 2) MIMO architectures · pp. 8–9
  • Hybrid designs provide frequency-selective spatial processing and multibeam operation, making them suitable for multistream, multiuser, and multitarget scenarios while approaching all-digital sensing and communication performance.Analog beamforming is limited by phase quantization, frequency-flat processing, and lack of amplitude tunability; low-resolution fully digital designs reduce cost and power but incur quantization noise.Source: 2) MIMO architectures · pp. 9–10
  • Future upper-mid-band systems are expected to use extremely large arrays, many-RF-chain hybrid MIMO, and multiband arrays, leaving joint array configuration for communication and sensing open.MIMO configurations will vary with channel features and bandwidth as carrier frequency increases.Source: 2) MIMO architectures · p. 10

C. Network sensing modes

  • The section establishes terminology for distinguishing the different types of network sensing and their relationship to concepts used throughout the paper.Source: C. Network sensing modes · p. 11

1) Sensing, Positioning, Localization:

  • Sensing processes received radio signals to extract service-relevant information, generally reflecting transmitter, receiver, and environmental geometry but not necessarily requiring it.Source: 1) Sensing, Positioning, Localization · p. 11
  • Positioning localizes UEs from uplink, downlink, or sidelink transmissions, whereas localization also estimates passive-target positions and detects their presence; here, sensing denotes radar-like object-state detection.Source: 1) Sensing, Positioning, Localization · p. 11
  • 6G sensing measurements can be fused with external cameras, lidar, or radar to provide complementary information, including seeing around automotive corners.Source: 1) Sensing, Positioning, Localization · p. 11

2) Mono-, Bi-, and Multistatic Sensing: … E. Network operation assisted by sensing

This section examines monostatic, bistatic, and multistatic sensing, along with sensing-assisted network operation in ISAC systems. It covers sensing evolution and open modeling and standardization gaps, and discusses ISAC maps, resilient adaptation, beam codebooks, blockage mitigation, and reduced beam-training overhead.

2) Mono-, Bi-, and Multistatic Sensing:

  • Monostatic sensing co-locates transmitter and receiver with a shared clock and known transmitted signal, whereas bistatic sensing separates them and must address synchronization and coordinate uncertainty.Monostatic measurements can use pilot or data signals; bistatic sensing with UE transmitters or receivers generally relies on pilots and may involve SLAM, while BS pairs may support time or phase synchronization.Source: 2) Mono-, Bi-, and Multistatic Sensing · pp. 11–12
  • Multistatic sensing uses multiple physically separated transmitters and/or receivers, requiring pilots or multiplexing and measurement fusion adapted to available synchronization.Different levels of time or phase synchronization lead to different ways of fusing measurements from the receivers.Source: 2) Mono-, Bi-, and Multistatic Sensing · p. 12

1) Evolution from positioning to sensing, a standardization perspective:

  • 3GPP positioning evolved from a 50 m LTE regulatory target to 3 m in early 5G NR indoor use cases and 1 m in release 17 industrial indoor IoT use cases.The section argues that sensing’s protocols and architecture will differ significantly from positioning, with new sensing signals potentially standardized in 6G if existing signals are insufficient.Source: 1) Evolution from positioning to… · pp. 12–13

2) Wireless network based use cases and requirements:

  • Network sensing can enable services across smart-home, transport, industrial, and UAV domains, while 3GPP has begun standardizing sensing use cases and requirements,.Examples include intrusion detection, automotive maneuvering and navigation, autonomous-vehicle tracking, robot collision avoidance, and UAV tracking or detection.Source: 2) Wireless network based use cases… · p. 13

3) 3GPP Propagation modeling for sensing:

  • Current 3GPP channel models do not support detailed sensing evaluation: sensing-specific parameters and aspects, sensing-target modeling, physically consistent geometric reflections, and spatially consistent channels for object tracking must be added.Source: 3) 3GPP Propagation modeling for sensing · pp. 13–14

E. Network operation assisted by sensing

  • Sensing data can improve network adaptability and resilience to propagation changes such as blockage, with machine learning assisting environmental representation and network adaptation.Machine learning can fuse sensing data with digital maps and communication models and provide recommendations for dynamic environments –.Source: E. Network operation assisted by sensing · p. 14
  • An ISAC map combines physical-world and radio-world measurements, including object properties and propagation-relevant information, to support sensing-assisted network operation.The map extends conventional radio maps by incorporating objects’ type, size, and trajectory alongside network-relevant sensing information.Source: E. Network operation assisted by sensing · p. 14
  • Tracking user 2’s automotive radar signal at the base station provides radar-channel information as a prior, reducing the overhead of establishing the user 2–base-station link.The radar-aided prior supports beam training for user 2.Source: E. Network operation assisted by sensing · pp. 15–16
  • User location can guide a beam codebook adapted to the statistical propagation environment around that location.This uses the sensed representation of the propagation environment to improve communication operation.Source: E. Network operation assisted by sensing · p. 16
  • The ISAC map fuses sensing data with previously effective network configurations, such as location-specific reduced beam codebooks, to support network operation.The urban-scenario map illustrates how sensing data and the technology used to obtain it can be integrated for communication enhancement.Source: E. Network operation assisted by sensing · p. 16
  • Tracking a moving target can predict user blockage, allowing the network to proactively find an alternative path and mitigate its impact.Blockage prediction and management are identified as major network-operation benefits of sensing data.Source: E. Network operation assisted by sensing · p. 16

III. WAVEFORMS, RESOURCE ALLOCATION AND CHANNEL PARAMETER ESTIMATION IN ISAC NETWORKS … 1) Why channel estimation for localization or sensing is different:

This section examines MIMO-OFDM waveforms and unified channel models for ISAC, focusing on estimating time-varying multipath parameters for communication, localization, and sensing. It addresses the resolution, accuracy, overhead, complexity, interference, ambiguity, filtering, and high-mobility considerations shaping these methods.

1) Fundamentals:

  • MIMO-OFDM enables efficient channel-parameter estimation by flexibly injecting known reference signals across time, frequency, and space, with 2D FFT/IFFT processing supporting delay/range and Doppler/velocity estimation.The basic resolutions are ∆τ = N/∆f and ∆f_D = ∆f/M; wider bandwidth improves delay estimation, while longer observation intervals improve Doppler estimation.Source: 1) Fundamentals · pp. 17–18
  • The sensing range of basic 2D-transform implementations is limited by the cyclic-prefix duration, reaching about 350 m with 30 kHz subcarrier spacing in current C-band 5G NR networks.Shorter symbol and cyclic-prefix durations at mmWave can make this limitation more pronounced.Source: 1) Fundamentals · p. 18
  • OFDM-based ISAC must address inter-carrier interference from long symbol durations and ambiguity sidelobes from FFT processing, cyclic prefixes, and frequency-sparse reference signals.Waveform optimization using the MIMO-OFDM subcarrier structure can affect ambiguity and target-parameter estimation performance.Source: 1) Fundamentals · p. 18

2) Unified Communication/Localization/Sensing Signal Model with MIMO-OFDM:

  • The unified MIMO-OFDM receive model covers communications, localization, and sensing in monostatic and bistatic configurations using analog precoding and combining, digital baseband precoding, and additive Gaussian noise.It is extended to time-varying channels and is used throughout the paper.Source: 2) Unified… · pp. 18–19
  • The model parameterizes each path or target by complex gain, delay, Doppler shift, angle of arrival, and angle of departure, with gain interpretations differing across localization and sensing.Communications can often neglect Doppler as time-invariant, whereas high-mobility localization and sensing require more comprehensive Doppler modeling.Source: 2) Unified… · p. 19
  • The unified MIMO-OFDM signal model incorporates pulse-shaping and filtering effects, while fast-time intra-symbol phase progressions can be neglected below 30 m/s but cause ICI in high-mobility scenarios.Practical implementations may leave roll-off-region subcarriers unused so the active-subcarrier frequency response remains flat; high-mobility effects must instead be modeled explicitly.Source: 2) Unified… · p. 20, 1) Why channel estimation for… · pp. 20–21

1) Why channel estimation for localization or sensing is different:

  • Localization and sensing require explicit delay estimation, including the transmitter–receiver clock offset, whereas communication estimation can often remain in the frequency domain.Absolute-delay estimation makes clock offset an additional parameter in the estimation process.Source: 1) Why channel estimation for… · p. 20
  • Sensing observes the channel over longer periods to estimate Doppler shifts and velocity, making joint space-time-frequency estimation more computationally demanding than communication-only estimation.Communication models may neglect Doppler within the coherence time, while sensing exploits Doppler shifts for velocity estimation.Source: 1) Why channel estimation for… · p. 20
  • High-resolution ISAC estimation increases algorithmic complexity and training overhead because localization and sensing require more accurate channel parameters than communications alone.A 1 m localization target at 50 m requires approximately 1° angle accuracy, pushing estimation algorithms and training length.Source: 1) Why channel estimation for… · pp. 20–21

2) Techniques for channel estimation: · 3) Spatial designs for channel estimation:

This section covers channel-estimation techniques and spatial training designs for improving estimation efficiency, accuracy, and localization. It examines maximum-likelihood, compressed-sensing, and subspace-based methods alongside directional and derivative-beam designs for angle-of-departure estimation.

2) Techniques for channel estimation:

  • Channel-estimation techniques are classified into maximum-likelihood –, compressed-sensing, –, and subspace-based estimators, with frequency-band suitability varying across methods.Most methods sound the channel using pilot sequences and training precoders/combiners, then extract multipath parameters from the received measurements.Source: 2) Techniques for channel estimation · p. 21
  • For sub-6 GHz or moderate arrays, conventional ML is optimal but complex, whereas SAGE variants [49]– provide super resolution at moderate complexity.Expectation-maximization methods provide high-accuracy estimates but remain computationally expensive.Source: 2) Techniques for channel estimation · p. 21
  • At mmWave, low-SNR operation, hybrid analog combining, and large arrays motivate sparse channel estimation, which can also provide localization information.Compressed-sensing methods recover sparse vectors using angular and delay dictionaries, but high-resolution overcomplete dictionaries can impose prohibitive computational or memory costs.Source: 2) Techniques for channel estimation · pp. 21–22
  • Subspace methods such as ESPRIT – estimate angles, delays, and Doppler shifts at moderate complexity, but standard ESPRIT requires models without filtering effects.A hybrid approach combines beamspace ESPRIT for angular estimation with dictionary-based sparse recovery for delays when filtering is present [59].Source: 2) Techniques for channel estimation · p. 23

3) Spatial designs for channel estimation:

  • Training precoders and combiners can be optimized beyond communication codebooks to reduce channel-estimation overhead or improve accuracy, including through low mutual coherence.Design objectives may target sparse recovery or accurate angle estimation; in the stated downlink localization example, the new codebook provides significantly better performance.Source: 3) Spatial designs for channel… · p. 24
  • The new directional/derivative codebook – augments standard directional beams with derivative beams, whereas the conventional codebook follows 5G NR mmWave practice,.The codebooks sample an AoD uncertainty interval U = [θ − Δθ, θ + Δθ], with normalized columns.Source: 3) Spatial designs for channel… · pp. 24–25
  • Derivative beams complement directional beams by producing sharp curvature around the intended angle, making small angular perturbations detectable through amplitude changes.This enables precise angle mapping from complex amplitude measurements and is supported by CRLB analysis,.Source: 3) Spatial designs for channel… · p. 25
  • For evaluation, directional-beam columns use uniform time sharing, while directional/derivative columns use CRLB-criterion-optimized time sharing.AoD RMSE is evaluated against the UE’s AoD using the maximum-likelihood estimator [64, Eq. (11)].Source: 3) Spatial designs for channel… · p. 25
  • Using Fdir/der instead of the traditional 5G codebook Fdir yields substantial improvements in AoD estimation accuracy, suggesting significant potential for achieving extreme location accuracy in 6G.Source: 3) Spatial designs for channel… · p. 25

C. Resource allocation · 1) Fundamentals: · 2) Problem formulations:

ISAC resource allocation uses unified waveforms and allocates radio resources to trade off sensing and communication performance while reducing hardware costs. Its formulations rely on distinct communication and sensing metrics, but a universal joint evaluation framework remains open, motivating sensing-oriented, communication-oriented, and joint optimization designs.

1) Fundamentals:

  • Unified waveforms simultaneously support sensing and communication, improving spectrum efficiency and reducing hardware costs through resource allocation for their performance tradeoff.Directional and derivative beams can jointly enable high-accuracy UE tracking in 5G/6G mmWave scenarios.Source: 1) Fundamentals · pp. 25–26
  • Communication performance is measured by channel capacity or SINR, whereas sensing performance is measured by target-state estimation error using Fisher information analysis.For position estimation, the squared position error bound uses the equivalent Fisher information matrix derived from the FIM [69].Source: 1) Fundamentals · pp. 26–27

2) Problem formulations:

  • A universal framework jointly evaluating ISAC sensing and communication performance remains open, so resource allocation uses multiple objective formulations rather than one unified problem.Time, frequency, space, power, and code resources are allocated across sensing-oriented, communication-oriented, and joint formulations.Source: 2) Problem formulations · p. 27
  • Sensing-oriented formulations allocate power and spatial resources to optimize sensing metrics such as SPEB or detection probability under communication and hardware constraints.Joint waveform, precoding, and combining design can minimize average SPEB, but non-convexity and tight variable coupling make the problem difficult; SDR and SCA provide high-quality solutions at acceptable computation costs –.Source: 2) Problem formulations · pp. 27–28
  • Communication-oriented formulations maximize mutual information or throughput through power allocation and beamforming while enforcing sensing-accuracy or beam-pattern constraints.Auxiliary-variable decomposition and convex relaxation yield efficient suboptimal solutions,, while waveform design exposes subspace and deterministic-random tradeoffs.Source: 2) Problem formulations · pp. 28–29
  • Joint formulations give sensing and communication equal status by optimizing weighted performance measures such as communication rate and estimation rate.The weights can represent S&C priorities, while value-of-service objectives use metrics such as sensing CRLB and communication SINR,.Source: 2) Problem formulations · p. 29
  • Joint resource allocation can instead minimize total resource consumption while guaranteeing sensing and communication performance through constraints.Robust power-allocation and spatial-design strategies account for uncertainties in network parameters,.Source: 2) Problem formulations · pp. 28–29

IV. TECHNOLOGIES FOR JOINT BISTATIC AND MULTISTATIC SENSING AND COMMUNICATION … 2) Minimal problems:

This section examines joint bistatic and multistatic sensing and communication, covering positioning, object localization, and SLAM alongside the fundamentals and solvability of minimal-infrastructure positioning. It also considers how complementary multi-antenna AoD and AoA measurements can estimate a UE’s position and orientation.

A. Introduction

  • Bistatic sensing uses one transmitter and one receiver, whereas multistatic sensing uses several, with bistatic communication sensing extensively studied and multistatic sensing comparatively limited in communications,,,.The section frames positioning, object detection/localization, and SLAM as distinct uses of separated transmitters and receivers.Source: A. Introduction · pp. 29–30

1) Fundamentals of position and orientation estimation:

  • Positioning estimates 3D UE location from signals involving known BS positions and orientations, typically by estimating line-of-sight AoA, AoD, ToA, and Doppler before solving for the UE state.Dedicated pilots facilitate channel-parameter estimation and improve resolution and accuracy, while least-squares or maximum-likelihood recovery is generally non-convex and may require approximations or relaxations.Source: 1) Fundamentals of position and… · p. 30
  • Nuisance parameters such as clock bias, CFO, and 3D UE orientation must be estimated alongside location, enabling joint network synchronization and complete 6D UE-pose estimation,,.Non-line-of-sight paths may be discarded after detection or exploited to solve positioning [91], [92].Source: 1) Fundamentals of position and… · p. 31

2) Minimal problems:

  • Minimal positioning problems seek the smallest measurement or infrastructure set that preserves solvability, while additional measurements, multipath, beacons, RISs, sidelinks, data-driven methods, and SLAM can reduce BS reliance.These alternatives trade infrastructure reduction against complexity, ambiguity, measurement conditions, or the need for resolved multipath.Source: 2) Minimal problems · p. 31
  • Four synchronized single-antenna BSs with line-of-sight provide sufficient time-of-arrival measurements to determine the 3D UE location and 1D clock bias.The clock bias can alternatively be removed using three correlated time-difference-of-arrival measurements.Source: 2) Minimal problems · p. 32
  • Two BSs with planar arrays provide two-dimensional angle-of-departure measurements whose intersection uniquely determines the UE point under noise-free conditions.The section uses this example to show how additional angular measurements can reduce infrastructure reliance.Source: 2) Minimal problems · p. 33
  • Different measurement types provide complementary information to the TDoA measurements described in Example 2.Additional examples are summarized in Table I, including the role of uncontrolled and controlled multipath.Source: 2) Minimal problems · pp. 33–34
  • 2D AoD measurements from two multi-antenna BSs estimate the UE’s 3D position, while 2D AoA then determines its 3D orientation.The orientation is uniquely determined because the two lines have four parameters, whereas UE orientation has three degrees of freedom.Source: 2) Minimal problems · pp. 33–34, Figure 10 · p. 34

3) Positioning in sub-6 GHz: · 4) Positioning in mmWave:

This section examines positioning in sub-6 GHz and mmWave bands, covering their respective propagation and hardware constraints and the benefits of mmWave’s larger bandwidths and arrays. It also considers ray-tracing comparisons and processing approaches, including super-resolution methods and matched filtering, across bands and bandwidths.

3) Positioning in sub-6 GHz:

  • In sub-6 GHz, 5–20 MHz bandwidth yields 15–60 m distance resolution, causing paths within 15 m to merge and positioning errors of tens of meters in clutter.Small UE arrays also limit angle resolution and prevent accurate orientation estimation, while BS arrays of roughly 64–128 elements help only when paths are angle-separated.Source: 3) Positioning in sub-6 GHz · p. 34
  • Sub-6 GHz channel richness and site-specific effects undermine statistical channel models and make conventional FFT- or correlation-based positioning perform relatively poorly.Machine-learning fingerprinting can reduce errors below 10 m but requires labeled fingerprint–location pairs; super-resolution methods instead resolve closely spaced paths when SNR permits.Source: 3) Positioning in sub-6 GHz · pp. 34–35

4) Positioning in mmWave:

  • At FR2, bandwidths up to 400 MHz provide sub-meter distance resolution, while sub-THz bandwidths of 1 GHz or more can enable resolution below 30 cm.These gains make delay-based positioning feasible in clutter, but require stricter and more challenging synchronization among BSs.Source: 4) Positioning in mmWave · p. 36
  • At mmWave, arrays with 64 or more elements support superior angle resolution, while good angle resolution can provide a path toward orientation estimation.These benefits involve analog or hybrid architectures, lower-dimensional beamspace processing, hardware impairments, lack of per-device calibration, and greater sensitivity to blockages.Source: 4) Positioning in mmWave · pp. 36–37
  • MmWave’s narrow beams couple communication and positioning, complicate simultaneous connections to several BSs, and motivate location-aided beam training and single-BS positioning.The single-BS example exploits uncontrolled multipath by estimating LoS and reflected-path angles and delays; the LoS estimate can provide a coarse position for maximum-likelihood refinement.Source: 4) Positioning in mmWave · pp. 37–38
  • Multipath-based methods are limited by multipath strength, angle-based methods incur large errors far from the multi-antenna BS or RIS, and the examples assume far-field propagation.The considered algorithms are matched filtering across frequency and space and ESPRIT-based super-resolution processing.Source: 4) Positioning in mmWave · pp. 38, 40
  • The evaluation uses 101 ray-traced urban-intersection instances, with a moving UE, a central BS, four surrounding buildings, and matched physical aperture sizes across frequency bands.Ray-tracer outputs provide channel gains, delays, angles of arrival, and angles of departure; the compared single-BS RTT method estimates position from line-of-sight delay and azimuth/elevation angles.Source: 4) Positioning in mmWave · pp. 38, 40
  • Under equal bandwidth utilization, both algorithms perform better at mmWave than sub-6 GHz despite higher mmWave path loss, because electrically large arrays improve angular resolution and path resolvability.Increasing bandwidth further enhances performance, supporting mmWave positioning through large bandwidths and electrically large arrays.Source: 4) Positioning in mmWave · pp. 40–41
  • PDPs show higher path gains at sub-6 GHz, while distant UEs can produce more diverse multipath through reflections from surrounding buildings.The ray-tracing study examines rich and sparse scattering environments caused by the presence or absence of large reflectors.Source: 4) Positioning in mmWave · p. 40, Figure 13 · p. 41
  • More than an order-of-magnitude accuracy degradation affects MF or correlation-based processing compared with super-resolution approaches, while ESPRIT significantly outperforms MF across both bands and tested mmWave bandwidths.ESPRIT benefits from resolving closely spaced paths, particularly in dense multipath environments.Source: 4) Positioning in mmWave · pp. 40–41, Figure 14 · p. 41

6) Machine learning for positioning:

Machine learning addresses the limitations of geometric and conventional fingerprinting-based positioning by learning mappings from channel fingerprints, while hybrid geometric/deep models and attention-based tracking improve accuracy in challenging wireless environments. The section reports sub-meter localization for 95% of LOS users, 20 cm tracking accuracy for 95% of users, and improved robustness in NLOS channels.

6) Machine learning for positioning:

  • Geometric positioning degrades across scenarios and frequency bands because NLOS multipath, channel-estimation errors, and practical impairments reduce localization accuracy.At mmWave, trilateration and triangulation often require links to several base stations, while accurate single-link methods incur high pilot overhead and complexity.Source: 6) Machine learning for positioning · pp. 41–42
  • ML-based fingerprinting learns the fingerprint-to-position mapping offline and performs online regression without storing or searching the full database.Richer CSI fingerprints and deep networks can address RSS limitations caused by coarse information, device dependence, and multipath variability.Source: 6) Machine learning for positioning · p. 42
  • CNNs, LSTMs, and Transformers extract spatial, subcarrier, or attention-based channel features, while transfer learning adapts outdated fingerprints using limited new measurements.Hybrid geometric/deep approaches can provide very high accuracy with small training datasets, [142].Source: 6) Machine learning for positioning · p. 43
  • Attention-based tracking networks replace potentially inaccurate mathematical evolution models, using prior channel and position estimates to improve position tracking.V-ChATNet uses attention over estimate sequences and keeps location error below 20 cm for 95% of the time in realistic ray-traced vehicular channels.Source: 6) Machine learning for positioning · p. 43

Example 5 (ML-based positioning)

  • 20 cm accuracy is achieved for 95% of users in combined LOS/NLOS channels, significantly outperforming Kalman-filter tracking, while localization reaches sub-meter accuracy for 95% of LOS users and 50% of NLOS users.The hybrid system uses PathNet, geometric localization, and ChanFormer; V-ChATNet further refines tracking, with results shown in Fig. 15(a) and Fig. 15(b).Source: 6) Machine learning for positioning · p. 44

7) Sidelink positioning: · 8) RIS-aided positioning:

The sections describe sidelink and RIS-aided positioning as complementary approaches that use UE cooperation or programmable reflections to enhance localization. They outline their operating stages, measurements, practical challenges, and demonstrated positioning accuracy.

7) Sidelink positioning:

  • Sidelink positioning uses UE-to-UE signals and measurements, with a target UE assisted by anchor UEs through configuration, transmission and measurement, and position-calculation stages.Positioning reference signals provide measurements, and ToA/AoA information is fused to estimate absolute or relative position at the network location server or UE.Source: 7) Sidelink positioning · p. 44
  • Sidelink positioning complements GNSS and conventional localization for V2X, with research covering high-accuracy architectures, sub-6 GHz multipath bounds, and near-field localization,,.Attention-based ChanFormer and V-ChATNet strategies are evaluated against MOMP + GeoLoc for initial access and tracking using ray-tracing data from [142], [143].Source: 7) Sidelink positioning · pp. 44–45
  • Cooperative localization generalizes sidelink positioning by jointly inferring multiple UE positions from network measurements and exploiting relative information from UE–UE links.Its theoretical and algorithmic benefits are emphasized for harsh indoor and urban-canyon propagation environments, with limits, cooperation laws, and resource-allocation methods studied in,, [160]–[162], [163], [164].Source: 7) Sidelink positioning · p. 45

8) RIS-aided positioning:

  • RIS positioning provides an additional position reference, measurements such as AoA, AoD, delay, and Doppler, and unique AoA or AoD measurements.For a passive RIS, entries have at most unit amplitude and controllable phase because it cannot amplify.Source: 8) RIS-aided positioning · pp. 46–47
  • RIS-aided positioning must determine RIS location and orientation and separate RIS signals from uncontrolled multipath, including the line-of-sight path.Calibration methods and dedicated RIS control sequences address these issues, particularly in multi-RIS scenarios.Source: 8) RIS-aided positioning · p. 47
  • In RIS-aided mmWave positioning, a modified low-complexity MOMP algorithm jointly estimates sparse parameters from environmental and RIS paths using separate dictionaries.The positioning formulation handles cases where line-of-sight paths exist from both BS and RIS or from only one source, using linear equations and indoor reflection properties,.Source: 8) RIS-aided positioning · p. 47
  • 80% of users achieve positioning error below 20 cm with a 32 × 32 RIS when line-of-sight paths exist from both the BS and RIS.This result comes from ray-tracing-based indoor smart-factory simulations using an 8 × 8 BS, 20 dBm transmit power, 60 GHz carrier frequency, and 100 MHz bandwidth.Source: 8) RIS-aided positioning · pp. 47–48, Figure 16 · p. 48

9) Carrier phase positioning: … D. Radio SLAM in bistatic scenarios

This section covers carrier-phase positioning, bistatic and multistatic sensing, and bistatic Radio SLAM within ISAC, focusing on precise ranging, non-connected-object localization, and joint UE–map estimation. It addresses ambiguity, detection and association, multisensor fusion, and Radio SLAM approaches for estimating UE pose, clock, and environmental maps in infrastructure-limited or NLoS conditions.

9) Carrier phase positioning:

  • Carrier-phase measurements enable millimeter-wave ranging to a few millimeters and C-band ranging at centimeter-level accuracy, but integer ambiguity prevents direct measurement of full wavelengths,,.Differential or double-differential measurements using a known fixed reference UE can solve or relax this ambiguity.Source: 9) Carrier phase positioning · pp. 48–49

C. Bistatic and multistatic sensing of non-connected objects

1) Fundamentals of mapping and target tracking, multi-site processing:

  • Mapping and multi-target tracking differ from positioning because they require detection amid clutter, estimation of complex object states, unknown data association, and multisensor fusion.Phase-coherent base stations can process aggregated waveforms as a distributed sensor, whereas non-synchronized stations fuse locally estimated channel parameters.Source: 1) Fundamentals of mapping and target… · p. 50

D. Radio SLAM in bistatic scenarios

  • Bistatic Radio SLAM jointly estimates the UE position and landmark locations, using NLoS paths to improve accuracy, reduce infrastructure dependence, and support localization without LoS.Its objective is the joint posterior p(x1:t, m | y1:t), which factorizes into the trajectory posterior and the map posterior conditioned on that trajectory.Source: D. Radio SLAM in bistatic scenarios · p. 52
  • RBPF methods represent each particle by a UE trajectory and associated map, with the UE state comprising pose and clock.The pose includes position and orientation, while the clock captures parameters needed to synchronize the UE’s local clock.Source: D. Radio SLAM in bistatic scenarios · p. 53
  • Bistatic Radio SLAM estimates the UE state in a BS-defined global reference system, enabling global-state estimation from a single snapshot and filtering-divergence detection.Snapshot and filtering approaches are both relevant; the preferred choice depends on the system and application scenario, while smoothing is unexplored in bistatic Radio SLAM.Source: D. Radio SLAM in bistatic scenarios · p. 53
  • An indoor bistatic SLAM experiment at 60 GHz with 400 MHz PRS bandwidth used a single BS transmitting PRSs to a UE to jointly estimate pose and the surrounding-environment map.The SLAM problem was solved using an RBPF-PHD filtering solution.Source: D. Radio SLAM in bistatic scenarios · p. 54
  • Filtering supports mixed LoS/NLoS operation, uses prior posteriors as regularization for state identifiability, and improves accuracy through sequential measurement processing.It also addresses misdetections, clutter and multi-bounce false detections, and measurement ambiguities that can associate observations with incorrect landmarks.Source: D. Radio SLAM in bistatic scenarios · pp. 54–55

V. TECHNOLOGIES FOR JOINT MONOSTATIC SENSING AND COMMUNICATION · A. Introduction · B. Basic SI modeling and isolation solutions

Monostatic cellular ISAC collocates transmission and reception, enabling UEs and gNBs to act as standalone cellular radars but requiring simultaneous transmit–receive operation and strong self-interference suppression. The section models SI waveforms and outlines antenna, beamforming, RF, and digital cancellation approaches, including hardware-impairment modeling.

A. Introduction

  • Monostatic sensing collocates transmitter and receiver entities –, allowing UEs and gNBs to provide standalone environmental awareness without other network entities.The transmit waveform illuminates landmarks while supporting situational awareness and coordinate tracking.Source: A. Introduction · p. 55
  • Because cellular transmit periods are much longer than pulsed-radar intervals, monostatic sensors must receive while transmitting and therefore operate as IBFD or STAR systems –.This simultaneous operation is needed to measure reflected and scattered waves at meaningful distances.Source: A. Introduction · p. 56
  • Direct TX–RX coupling is a powerful short-range target that can mask sensing echoes, so isolation must precede the LNA through antenna, RF, and baseband cancellation –.A macro-gNB EIRP can reach about +70 dBm while a 100 MHz receiver noise floor is around –90 dBm.Source: A. Introduction · p. 56

B. Basic SI modeling and isolation solutions

  • The main antenna-interface choices are shared antennas with circulators or separate TX and RX antenna sets, with separate systems generally preferred for higher spatial isolation.Separate antennas also enable more effective beamforming-based TX–RX isolation, while RF and digital cancellation principles apply to both arrangements.Source: B. Basic SI modeling and isolation… · pp. 56–57
  • The SI waveform model incorporates precoded subarray transmissions, physical TX–RX channels, TX/RX beamforming, and transmitter and receiver hardware impairments.The formulation includes PA nonlinearity and oscillator phase noise through transmitter and receiver hardware models.Source: B. Basic SI modeling and isolation… · pp. 57–58
  • Both TX and RX beamforming shape observable SI, with TX beamforming especially important because it controls SI at the RX LNA input.Optimization can trade SI suppression against target illumination and achievable communication rate –.Source: B. Basic SI modeling and isolation… · p. 58
  • Hardware-aware modeling enables digital cancellation of PA, LNA, and analog distortion, improves receiver dynamic-range optimization, and can turn oscillator phase noise into a sensing resource.This benefit is tied to modeling hardware imperfections rather than treating them only as impairments.Source: B. Basic SI modeling and isolation… · p. 58

C. Triple-function precoding/combining in full duplex transceivers … RHSI ¯AT,

The section develops triple-function precoding and combining for full-duplex monostatic sensing during tracking and initial access, jointly supporting communication, target gain, and self-interference suppression. It formulates hardware-constrained designs and presents SI-aware codebooks that enable target detection where the initial codebook fails.

1) Precoding/combining for tracking:

  • For tracking, TX and RX spatial filters simultaneously enable downlink communication, provide high target-direction gain, and contribute to SI suppression through multibeam precoding.The multibeam precoder illuminates the target angle and multiple communication-channel paths.Source: 1) Precoding/combining for tracking · pp. 58–59

+ WHHSI

  • The SI channel is treated as frequency-flat or represented at the center frequency for wideband operation, while the FD BS is stationary so SI remains constant across symbols.The codebook objective balances SI coupling reduction against beam gain on the angular grids.Source: + WHHSI · p. 59, RHSI ¯AT · p. 61
  • The tracking design maximizes sum downlink rate under transmit and receive target-gain, SI-suppression, analog-hardware, and power constraints.Communication performance is quantified by subcarrier-dependent rate R_n, while sensing performance uses transmit and receive target gain thresholds τ_T and τ_R.Source: + WHHSI · p. 59

W, FRF,

  • Because precoders and combiners are coupled by analog hardware constraints, alternating optimization with convex or semidefinite relaxations can solve the joint design subproblems.Prior approaches include gain relaxation and null-space projection in, rate maximization with analog SI cancellation and DFT codebooks in, and generalized-eigenvalue precoding in [204].Source: W, FRF · p. 60

2) Precoding/combining for initial access:

  • Initial access requires wide angular coverage, so communication beam-training precoders and combiners can also support radar sensing while sharing only communication and sensing functions.Unlike tracking, initial access does not require the full triple-function requirement because spatial resources need not be shared with SI suppression in the same way.Source: 2) Precoding/combining for initial… · p. 61

RHSI ¯AT,

  • SI-aware initial-access codebooks are designed to minimize beam-pair coupling through the SI channel while maintaining high AoA/AoD gain and unit-modulus analog realizability.The TX and RX codebooks use alternating minimization, convex relaxation, and block-coordinate descent, with beam training pairing matching AoD and AoA grid angles.Source: 2) Precoding/combining for initial… · p. 61, RHSI ¯AT · p. 61
  • SI-aware codebooks efficiently suppress SI and yield high radar SINR at the target angle, whereas the initial codebook is corrupted by SI and makes detection impossible.This result is shown for a point target with two 8 × 8 arrays separated by 10λ; Fig. 22 compares the initial and SI-aware codebooks.Source: RHSI ¯AT · p. 61, Figure 22 · p. 62
  • The described monostatic designs serve downlink users only; integrating uplink users remains a key challenge but could significantly improve overall spectrum efficiency.The BS precoder supports sensing and communication with SI suppression, while the communication setting otherwise considers half-duplex operation.Source: RHSI ¯AT · p. 62

D. Joint monostatic sensing and communication under non-idealities: repurposing challenges into benefits … E. Radio SLAM in monostatic scenarios

This section examines monostatic ISAC under shared-oscillator impairments and develops radio SLAM for infrastructure-less localization and mapping. It covers how impairments can provide target-dependent geometric information and how smartphone and THz ISAC devices use radio reflections for simultaneous position estimation and map generation.

D. Joint monostatic sensing and communication under non-idealities: repurposing challenges into benefits

1) OFDM monostatic radar sensing without non-idealities:

  • Without non-idealities, SISO OFDM monostatic radar removes data-symbol effects and applies IDFT/DFT processing to form a delay-Doppler spectrum for detection and estimation.CFAR or super-resolution methods such as MUSIC and ESPRIT can then estimate target parameters.Source: 1) OFDM monostatic radar sensing… · pp. 63–64

2) OFDM monostatic radar sensing under ICI:

  • ICI arises when high mobility or small subcarrier spacing destroys subcarrier orthogonality, increasing sidelobes and masking weaker targets in standard range processing.The conventional model is approximately valid when velocities are low or subcarrier spacing is large enough that intra-symbol phase progression is negligible.Source: 2) OFDM monostatic radar sensing under… · p. 64

3) OFDM monostatic radar sensing under PN:

  • PN distorts subcarrier orthogonality and reduces radar dynamic range, thereby degrading detection performance, especially for weak targets.With an ideal oscillator, the PN matrix becomes all ones and the model reverts to the non-idealities-free case.Source: 3) OFDM monostatic radar sensing under… · p. 65

EXPLOITATION OF IMPAIRMENTS IN JOINT MONOSTATIC SENSING AND COMMUNICATIONS

  • ICI provides an additional Doppler dimension whose unambiguous velocity is N times greater, enabling high-speed ambiguity resolution and improved target resolvability.PN’s delay-dependent statistics similarly resolve range ambiguity for distant targets, and ICI can distinguish targets sharing a delay-Doppler-angle bin.Source: EXPLOITATION OF IMPAIRMENTS IN JOINT… · p. 66
  • PN exploitation can elevate radar performance beyond an ideal radar without PN, repurposing a sensing impairment into a benefit.PN statistics resolve far-target range ambiguity through covariance matching, unlike the ambiguous range information in b(τ).Source: EXPLOITATION OF IMPAIRMENTS IN JOINT… · p. 66

E. Radio SLAM in monostatic scenarios

  • Radio SLAM extends classical robotic SLAM’s goal of detecting landmarks and simultaneously estimating landmark and user positions to monostatic radio scenarios.The mapping problem is introduced as a coupled “chicken and egg” challenge.Source: E. Radio SLAM in monostatic scenarios · p. 67
  • Smartphone ISAC devices can perform Radio SLAM, enabling infrastructure-less localization and automatic digital-map generation while preserving privacy and reducing energy use versus lidar-based solutions.Radio SLAM identifies landmarks from specular reflections of signals emitted by an antenna-array-equipped mobile user.Source: E. Radio SLAM in monostatic scenarios · p. 68
  • Multipath, diffuse reflections, antenna side lobes, and limited reflection angles can create ghost landmarks or invisible obstacles, making Radio SLAM harder than lidar- or vision-based SLAM.At millimeter-wave frequencies, obstacles are typically visible in monostatic sensing only when signals impinge nearly perpendicularly.Source: E. Radio SLAM in monostatic scenarios · p. 69
  • THz bandwidth and large arrays provide high spatial and angular resolution, while diffuse reflections at THz frequencies can improve object visibility for nonperpendicular incidence.Experimental THz Radio SLAM remains at an early stage, with initial work assessing scan matching and modified Fourier-Mellin methods.Source: E. Radio SLAM in monostatic scenarios · pp. 69–70
  • 12 cm localization RMSE was obtained at 300 GHz in a large-office experiment using circular scanning, a ±90° span, and an 18° beamwidth.The experiment evaluated mapping and trajectory reconstruction with the modified Fourier-Mellin algorithm using real-world THz radar measurements.Source: Figure 28 · p. 69, E. Radio SLAM in monostatic scenarios · p. 70

VI. SENSING AND COMMUNICATING WITH WIDE APERTURES … C. Communication and sensing with ELAAs

This section examines integrated sensing and communication with extremely large-scale arrays, focusing on near-field operation and the spatial and distance information enabled by spherical wavefronts. It also discusses key modeling and estimation challenges and approaches based on focusing, distance-angle optimization, polar-domain representations, and geometry-based channel estimation.

A. Introduction

  • XL-MIMO increases antenna counts by at least an order of magnitude, producing electrically large apertures that improve spectral efficiency and spatial resolution for communication and sensing –.An ELAA is defined here as an antenna whose dimension D is much larger than the wavelength λ.Source: A. Introduction · p. 71

B. Near-field vs. far-field communication and sensing

  • Near-field spherical wavefronts provide distance information in addition to angle, and at millimeter-wave and THz frequencies the near-field may span practical distances of meters or hundreds of meters.This regime challenges far-field channel models but can improve communication and sensing capabilities.Source: B. Near-field vs. far-field… · p. 71

C. Communication and sensing with ELAAs

  • Near-field ELAA focusing controls interference by both angle and distance, enabling same-angle user discrimination and high-rank communication even under strong line-of-sight conditions.Near-field phase profiles also support source localization, while ELAA-based sensing can use narrowband signals with lower latency and fewer synchronization or ranging procedures.Source: C. Communication and sensing with ELAAs · pp. 72, 73, 75
  • A single near-field ELAA can estimate a source’s position from its aperture phase profile, while an array-equipped source enables six-dimensional position-and-orientation estimation.In the far field, the phase profile loses the distance-dependent term, so only the angle of arrival can be estimated; near-field ELAA sensing avoids ad-hoc synchronization and time-base ranging protocols.Source: C. Communication and sensing with ELAAs · pp. 73–74
  • Prior work minimizes the near-field joint distance-and-angle sensing Cramér–Rao bound under per-user communication-rate constraints and studies fully digital and hybrid architectures.Another approach uses low-complexity two-stage estimation, refining a far-field target estimate under the correct near-field condition [242].Source: C. Communication and sensing with ELAAs · p. 75
  • ELAA ISAC faces near-field modeling challenges because large apertures create spatially non-wide-sense-stationary channels, varying propagation perspectives, powers, and polarizations across array regions.These effects arise because different array regions observe the environment from different perspectives and may receive distinct propagation paths.Source: C. Communication and sensing with ELAAs · p. 75
  • Highly accurate ELAA channel estimation requires many parameters and substantial signaling overhead, while near-field location-domain sparsity makes angular-domain algorithms inapplicable.A polar-domain channel representation captures spherical-wave characteristics.Source: C. Communication and sensing with ELAAs · p. 75
  • In strong LoS conditions, channel estimation can align with localization because the channel is determined by the source position using only three parameters.This geometry-based relationship further intertwines sensing and communication.Source: C. Communication and sensing with ELAAs · p. 75
  • Dense-element ELAAs require accurate modeling of mutual coupling effects between elements.The section also illustrates distributed-array positioning and multistatic sensing using phase-coherent processing across radio units.Source: C. Communication and sensing with ELAAs · pp. 75–76, Figure 32 · p. 76

D. Distributed MIMO/cell-free massive MIMO for multi-perspective localization and sensing · E. Sensing and localization aided by large RIS

Distributed MIMO enables multi-perspective localization and sensing through a large, coordinated aperture, while large RISs support near-field localization and sensing even under blockage or element failures. These benefits require stringent synchronization and geometric knowledge for D-MIMO, whereas RIS-assisted localization can achieve 20–30 cm trajectory error in a blocked indoor scenario.

D. Distributed MIMO/cell-free massive MIMO for multi-perspective localization and sensing

  • D-MIMO comprises widely separated radio units connected to distributed units for baseband processing and a central unit for higher-layer coordination.Architectures differ in node topology, local versus central processing, and synchronization, while retaining massive-MIMO operations such as pilots and multi-user beamforming.Source: D. Distributed MIMO/cell-free massive… · p. 76
  • D-MIMO operation resembles carrier-phase positioning, multistatic sensing, and ELAA processing, but differs through its large, dense, distributed architecture.Typical operation assumes phase synchronization among RUs to enable spatial focusing toward UEs.Source: D. Distributed MIMO/cell-free massive… · pp. 76–77
  • D-MIMO localization and sensing require phase stability plus RU locations known within a fraction of the wavelength, placing users and objects in the array’s near field.These stringent requirements are likely feasible mainly at FR1, while higher-frequency operation is more constrained.Source: D. Distributed MIMO/cell-free massive… · p. 76
  • Large distributed apertures provide more uniform communication, localization, and sensing coverage, with enhanced multipath resolution and improved accuracy but higher computational complexity.The distributed architecture uses many geographically separated radio units coordinated through distributed and central units.Source: D. Distributed MIMO/cell-free massive… · pp. 76–77

E. Sensing and localization aided by large RIS

  • Near-field large-RIS measurements can estimate a UE’s 3D position when the direct UE–BS path is blocked or some RIS elements fail.The model uses the RIS near-field response as a function of UE position across different times, and a narrowband phase-profile variant is also proposed in.Source: E. Sensing and localization aided by… · pp. 77–78
  • 20–30 cm localization error is maintained along an indoor UE trajectory despite obstacles partially shadowing a long RIS.The UE determines its location using signals reflected only by the RIS, with dynamically varied element reflection coefficients enabling BS–RIS–UE delay measurements.Source: E. Sensing and localization aided by… · p. 78
  • Large RIS configurations can scan an area for target detection and localization with limited bandwidth by exploiting the BS–RIS–target propagation paths.A monostatic single-antenna BS can harness the RIS’s high spatial resolution for sensing.Source: E. Sensing and localization aided by… · p. 78

VII. TECHNOLOGIES FOR SENSING ASSISTED COMMUNICATION … 2) Radar-aided beam configuration:

This section examines sensing-assisted communication strategies for reducing beam-configuration costs and mitigating blockage in challenging mmWave networks. It covers position-, learning-, and radar-aided approaches, including covariance mapping and radar-based vehicle-parameter estimation for updating and tracking communication beams.

VII. TECHNOLOGIES FOR SENSING ASSISTED COMMUNICATION

  • MmWave beam adaptation is difficult because large arrays, bandwidths, blockage, penetration losses, and challenging circuits increase configuration costs and channel variability.Initial access can take up to 5s with simple analog beamforming, while blocked-LOS throughput falls 2× via an NLOS path and 4× when switching to 4G because no NLOS path is available.Source: §VII · pp. 78–79

1) Position-aided beam training:

  • Position-aided beam training uses the user’s position-estimated direction to define a reduced beam set, accounting for angular estimation error and yielding overhead reduction of (1 −|L|/NT).Source: 1) Position-aided beam training · pp. 79–80

2) Radar-aided beam configuration:

  • Radar-aided configuration reduced communication overheads by 77% in a realistic ray-traced vehicular environment.Radar covariance can serve as a prior for communication covariance because their main channel directions are highly similar at nearby mmWave frequencies.Source: 2) Radar-aided beam configuration · p. 81
  • Deep learning maps radar covariance to communication covariance to compensate for frequency and location mismatches, producing further overhead reductions and higher effective rates,.The mismatch arises because radar and communication modules may use different frequencies or vehicle locations, and the approach supports multiple users.Source: 2) Radar-aided beam configuration · p. 81

SIMULATION OF A RADAR-AIDED MMWAVE VEHICULAR COMMUNICATION SYSTEM: PARAMETERS FOR THE COMMUNICATION

  • The radar-aided vehicular simulation uses four users, ray-traced urban environments, a 9600-sample training set, a 2400-sample test set, and four radar-directed beams.It compares exhaustive search, radar-covariance beam training without mismatch compensation, and communication-covariance training after ML-based covariance mapping using sum rate versus coherence time.Source: 2) Radar-aided beam configuration · p. 81, SIMULATION OF A RADAR-AIDED MMWAVE… · pp. 82–83

Simulation of a radar-aided mmWave vehicular communication system: Parameters for the communication

  • Radar-aided beam training provides significant performance gains across all considered coherence times, while ML-based mismatch prediction is more beneficial at shorter coherence times.The approach maps radar covariance to communication covariance after ML-based processing.Source: SIMULATION OF A RADAR-AIDED MMWAVE… · p. 83
  • A joint radar-communication base station estimates a vehicle’s angle, distance, and velocity, then updates beamformers using the predicted communication-link angle, improving performance over other benchmarks.This method targets the beam-tracking stage of vehicular communication.Source: SIMULATION OF A RADAR-AIDED MMWAVE… · p. 83
  • A related method estimates range, speed, angle of arrival, and path loss from radar-derived delay and Doppler using factor-graph message passing, then uses angular information for beam tracking.The approach exploits parameters estimated at the base station for the link associated with a particular vehicle.Source: SIMULATION OF A RADAR-AIDED MMWAVE… · p. 83

3) Vision-aided beam training: … VIII. CONCLUSION

This section presents sensing-aided beam training and blockage management, using environmental and channel data with multimodal learning to reduce training overhead and support proactive link management. It concludes by highlighting integrated sensing and communication as a key 6G feature that enables new sensing services and improves network operation.

3) Vision-aided beam training:

  • Vision-aided beam training uses images and position information to identify users, reflectors, and LoS paths, improving beam selection and potentially reducing training overhead.Images can also track users in dynamic environments for beam prediction.Source: 3) Vision-aided beam training · p. 84
  • Reconstructed 3D environments provide spatial information for neural beam selection, while image-based prediction can accurately forecast several future beams in dynamic vehicular scenarios.Realistic-environment dataset creation remains an open challenge for learning methods requiring paired channels and images.Source: 3) Vision-aided beam training · p. 84
  • UE-mounted cameras can detect surrounding objects and vehicles, provide beam decisions, and predict beam coherence duration; simulations found them more beneficial than BS cameras.This approach uses 3D modeling and ray tracing.Source: 3) Vision-aided beam training · p. 84

4) Exploiting LIDAR point clouds and multimodal information:

  • LIDAR is also combined with GPS, camera images, or depth maps to support multimodal beam-training strategies for vehicles and robots.These sensors provide physical information about environmental obstacles and structure.Source: 4) Exploiting LIDAR point clouds and… · p. 85
  • LIDAR-aided beam training identifies obstacles and combines point clouds with positions or other data to classify LOS/NLOS states and recommend reduced beam sets.Curriculum learning improves classification in complex NLOS cases, while federated learning lets connected vehicles fine-tune a shared beam-selection network and reduce overhead for later participants.Source: 4) Exploiting LIDAR point clouds and… · p. 85

B. Blockage prediction and management

  • At mmWave and THz frequencies, blockage can cause severe attenuation or link failure, motivating proactive sensing-based prediction and management beyond reactive beam retraining.Reported attenuation reaches 20 dB for vehicles, 40 dB for people, and 20 dB for hands.Source: B. Blockage prediction and management · pp. 85–86

VIII. CONCLUSION

  • Communication and sensing should be designed together because ISAC enables new sensing services and provides data that enhances network operation, making integration a key 6G feature.Source: §VIII · p. 87
  • The paper presents a comprehensive ISAC cellular-network vision covering network sensing and sensing-aided communication, including bistatic, multistatic, monostatic, wide-aperture, positioning, and radio-SLAM approaches.Source: §VIII · p. 87

B. Blockage prediction and management

  • Sensor-aided beam training reduces overhead, while blockage detection and proactive prediction can be achieved in mmWave systems using sensors and machine-learning solutions.Range-angle maps from a roadside-unit radar can serve as LSTM inputs for blockage prediction, whose effectiveness was shown experimentally.Source: §VIII · p. 87, B. Blockage prediction and management · p. 87

VIII. CONCLUSION

  • The paper includes examples illustrating the performance that these sensing and communication technologies can achieve in practical scenarios.Source: §VIII · p. 87

Related papers

Loading 2405.01816v1…