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

The paper addresses how future cellular networks can integrate sensing, learning, and communication to provide sensing services and support network operation. It presents a vision and overview of signal-processing, optimization, and machine-learning approaches for 6G ISAC, with examples evaluated through ray-tracing and mathematical simulation models. The paper reports positioning and sensing results, including errors of 0.55 m, 2.43°, and 1.53 ns for position, orientation, and synchronization error, respectively.

  • Problem

    Future wireless networks need to integrate sensing, learning, and communication, while stringent performance requirements motivate advanced ISAC solutions.

  • Method

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

  • Results

    0.55 m, 2.43°, and 1.53 ns are reported for position, orientation, and synchronization error, respectively, in a radio-SLAM bistatic scenario.

  • Takeaways & Limitations

    The paper concludes that communication and sensing should be considered together in future cellular networks.

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

ISAC integrates sensing and communication in future cellular systems, using shared capabilities to support new applications and network operation. The paper presents a communication-centric 6G vision, surveys enabling techniques across frequency bands and sensing modes, and reviews sensing-aided communication.

  • Higher frequencies, large arrays, and large bandwidths produce waveforms and signal-processing algorithms that are naturally well-suited for sensing.
  • Connected vehicles illustrate sensing and communication devices that combine cameras, radars, or LIDAR, creating challenges for sensor and communication data fusion.
  • ISAC exploits similarities in hardware, waveforms, signal processing, machine learning, and sensing and communication channels to define new applications for future wireless networks.
  • The paper adopts a communication-centric ISAC perspective with tight integration of waveform and time and frequency resources for sensing and communication.
  • The survey covers network sensing modes including bistatic, multistatic, monostatic, positioning, wide-aperture sensing, and radio-SLAM.
  • Sensing-aided communication can leverage sensing information to reduce link-configuration overhead and enable early blockage detection.

B. Spectrum and MIMO technologies

6G sensing performance depends on spectrum selection, propagation conditions, bandwidth, array aperture, and MIMO architecture. The paper contrasts architectures and bands by their resolution, coverage, mobility, hardware complexity, and suitability for joint sensing and communication.

  • Spectrum Considerations: FR1 offers broad coverage and high mobility but poor delay resolution because its bandwidth is limited.Its lower carrier frequency reduces path loss and Doppler frequencies, while propagation is less geometric and more statistical.
  • Spectrum Considerations: FR2 provides good resolution through directional arrays but has limited coverage and supports only moderate mobility.Higher path loss, pronounced shadowing, and fewer propagation paths constrain its coverage and mobility profile.
  • Spectrum Considerations: The upper mid-band offers a potential trade-off between data rate and wide coverage, while sub-THz offers high resolution for short-range imaging and mapping.Sub-THz operation is envisioned for nearly static conditions, with fine high-gain beams, very short range, and diffusive reflections.
  • Spectrum Considerations: Multi-band networks can combine or switch among bands to support diverse sensing services, but transceiver, antenna, propagation, and resource-allocation issues require further study.The paper also calls for geometric ray-tracing models or common databases to evaluate sensing across 6G bands.
  • MIMO architectures: MIMO architecture determines the received signal model, channel-parameter extraction methods, and precoder or combiner design for joint sensing and communication.The paper discusses analog, hybrid, low-resolution, and fully digital architectures under differing hardware constraints.
  • MIMO architectures: Analog beamforming has limited amplitude control, quantized phases, and frequency-flat spatial processing, making it unsuitable for multistream, multiuser, or multitarget scenarios.Hybrid architectures add digital processing with fewer RF chains, enabling multibeam operation and performance close to all-digital solutions, while complicating estimation and optimization.
  • MIMO architectures: Hybrid precoding and combining factor into analog and digital matrices, and the added digital stage enables frequency-selective spatial processing.Hybrid designs use fewer RF chains than antennas and provide more degrees of freedom than analog architectures.
  • MIMO architectures: Low-resolution fully digital architectures reduce power consumption and cost through low-resolution converters, but high quantization noise compromises performance.At mmWave frequencies, space and power constraints prevent using one high-resolution RF chain per antenna.

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

The paper distinguishes monostatic, bistatic, and multistatic sensing by transmitter–receiver arrangement, synchronization, and measurement reference frame. It situates sensing within 3GPP’s evolution from positioning toward broader sensing services and ISAC-enabled network adaptation.

  • Sensing configurations: Monostatic sensing co-locates transmitter and receiver with a shared clock and known transmitted signal, enabling pilot- or data-based measurements in the transmitter’s coordinate system.Measurements can include ToA, AoA, and detected objects; the transmitter may be a base station or mobile UE.
  • Sensing configurations: Bistatic sensing separates transmitter and receiver, requiring pilot-based sensing when a UE is involved and accounting for unknown UE location through a SLAM problem.When both nodes are base stations, synchronization and transmitted-data knowledge may be assumed; otherwise, the LoS path can reference later multipath.
  • Sensing configurations: Multistatic sensing uses several physically separated transmitters and/or receivers, requiring pilots or multiplexing and measurement fusion under potentially different synchronization levels.Several transmitters need interference avoidance, while receiver fusion depends on available time or phase synchronization.
  • Evolution from positioning to sensing: Positioning specifications evolved from a 50 m LTE regulatory target to 5G NR requirements down to 3 m indoors and 1 m for Release 17 industrial indoor IoT.The evolution includes positioning signals, measurements, procedures, and architecture standardized across multiple 3GPP releases.
  • Evolution from positioning to sensing: Sensing may differ substantially from positioning because its protocol and architecture target passive scatterers, and new sensing signals may be standardized in 6G if existing signals are insufficient.The paper also notes that many sensing forms are outside its coverage.
  • Network operation assisted by sensing: An ISAC-map fuses physical-world and radio-world measurements, representing users, blockers, static scatterers, and inferred object properties such as type, size, and trajectory.Such data can support location-aware beam design, blockage prediction, network adaptation, and reduced training overhead.

III. WAVEFORMS, RESOURCE ALLOCATION AND CHANNEL PARAMETER ESTIMATION IN ISAC NETWORKS

The section examines multicarrier MIMO-OFDM waveforms for communication, localization, and sensing, including their processing benefits, parameter resolutions, signal-model foundations, and practical limitations. It emphasizes the trade-offs among flexible reference-signal design, sensing accuracy, waveform ambiguity, cyclic-prefix range, and mobility effects.

  • Waveform design and resource use: Complementary filtering and windowing can enhance waveform spectral containment while preserving transparent transmitter, receiver, or joint implementation.OFDM/OFDMA also supports MIMO precoding, beamforming, link adaptation, and scheduling.
  • Practical limitations: OFDM-based ISAC must address inter-carrier interference from long symbol durations, along with sensitivity to phase noise, carrier-frequency offsets, and Doppler spread.OTFS is described as an alternative multicarrier scheme designed for increased robustness against Doppler.
  • Waveform fundamentals: MIMO-OFDM supports flexible reference-signal injection and efficient channel-parameter estimation for both communication and sensing receivers.Two-dimensional FFT/IFFT processing over multiple symbols enables delay/range and Doppler/velocity estimation.
  • Waveform fundamentals: The basic resolutions are Δτ = N/Δf for delay and ΔfD = Δf/M for Doppler, with N and M denoting frequency- and time-transform sizes.Wider bandwidth improves ranging, while longer observation intervals improve Doppler and velocity estimation.
  • Practical limitations: OFDM’s cyclic-prefix length directly limits sensing range because target reflections and dominant scattering components must fit within the CP duration.With 30 kHz subcarrier spacing in 3.5 GHz 5G NR, target distances remain on the order of 350 m; reduced CP lengths at mmWave can make this limitation more pronounced.
  • Waveform design and resource use: Waveform ambiguity and target-parameter estimation can be affected by FFT and cyclic-prefix sidelobes, sparse reference signals, and subcarrier structure.Waveform optimization is presented as a way to address these effects, with MIMO-OFDM providing a basis for optimization.
  • Unified signal model: The unified time-varying MIMO-OFDM model represents received signals using precoding, the channel, transmitted data streams, and additive white Gaussian noise.Its channel parameters include path gain, delay, Doppler, angles, and sensing-related radar cross section and attenuation.
  • Channel parameter estimation: Localization and sensing require higher channel-estimation accuracy than communication-only precoder or combiner design, increasing estimation complexity and training-sequence length.The resulting training overhead affects the design of communication and sensing systems.

2) Techniques for channel estimation:

Channel estimation methods address frequency-, array-, and hardware-dependent challenges through likelihood, sparse-recovery, and subspace techniques. At mmWave, sparsity-based methods and specialized beam designs support high-resolution estimation, but resolution can increase computational and memory costs.

  • Estimator categories: Channel estimation methods include maximum-likelihood, compressed-sensing, and subspace-based estimators, with suitability depending on frequency band and system architecture.Maximum-likelihood methods can provide high resolution but often have high complexity, while SAGE offers super-resolution at moderate complexity.
  • MmWave challenges: MmWave estimation is difficult because initial beams may be misaligned and operate at low SNR, hybrid combining compresses observations, and large arrays increase channel dimensionality.These properties make many techniques developed for lower frequencies or smaller arrays impractical.
  • Sparse recovery: Sparse-recovery methods model frequency-selective channels as sparse vectors whose dictionary atoms represent combinations of AoD, AoA, and delay.Time-domain formulations include delay, while frequency-domain dictionaries are built from transmit and receive steering vectors.
  • Sparse recovery: Estimating channel parameters becomes support identification in an overcomplete dictionary, with ℓ1 relaxation used to recover sparse channel gains.Higher angular and delay resolution requires larger dictionaries, increasing computational or memory requirements.
  • Subspace methods: Subspace-based estimators can provide high-resolution angles, delays, and Doppler information, but ESPRIT-based methods require channel models without filtering effects.A hybrid beamspace-ESPRIT and sparse-recovery approach is described for models that include filtering effects.
  • Spatial designs: Directional and derivative beams improve AoD estimation accuracy compared with the traditional 5G codebook and can support high-accuracy tracking.The cited discussion reports substantial improvements with Fdir/der compared with Fdir and identifies potential for extreme location accuracy in 6G.

C. Resource allocation

ISAC resource allocation jointly manages communication and sensing objectives using metrics such as mutual information, SINR, SPEB, and detection probability. Formulations range from function-oriented optimization to joint weighted objectives, but remain difficult because of non-convexity and tight variable coupling.

  • Resource-allocation goals: Unified waveforms perform sensing and communication simultaneously, enabling resource allocation to trade off both functions and reduce hardware costs.Radio resources are allocated according to the performance metrics of the dual functions.
  • Performance metrics: ISAC evaluates communication through capacity or SINR and sensing through target-state estimation error, including Fisher-information-based metrics such as SPEB.The sensing state may include position, orientation, velocity, and other quantities of interest.
  • Open challenges: Resource-allocation design lacks a universal theoretical framework and involves tradeoffs including subspace and deterministic-random tradeoffs in waveform design.The paper also discusses robust strategies that account for uncertainty in network parameters.
  • Sensing-oriented formulation: Sensing-oriented formulations optimize sensing metrics such as SPEB or detection probability subject to communication constraints and hardware, power, and structure constraints.The resulting joint problem is non-convex and tightly coupled, motivating SDR, SCA, and related approximations.
  • Communication-oriented formulation: Communication-oriented formulations maximize mutual information or throughput through power allocation and beamforming while imposing sensing constraints.Convex-relaxation techniques and auxiliary-variable decompositions provide efficient suboptimal solutions.
  • Joint formulation: Joint formulations assign equal status to sensing and communication through weighted performance objectives, value of service, or resource-consumption objectives with performance constraints.One example maximizes wcRc + wsRs, where estimation rate measures target-state entropy reduction.

IV. TECHNOLOGIES FOR JOINT BISTATIC AND MULTISTATIC SENSING AND COMMUNICATION

Bistatic and multistatic sensing use signals from one or multiple spatially separated transmitters and receivers to support positioning, object sensing, and SLAM. Positioning combines estimated channel parameters with geometric models, while complementary measurements and multipath can reduce infrastructure dependence.

  • Sensing architectures: Bistatic sensing uses one transmitter and one receiver, whereas multistatic sensing uses several transmitters or receivers.Bistatic sensing is readily implemented in communication systems, while multistatic sensing has received more limited treatment in communications.
  • Sensing scenarios: The covered scenarios include UE positioning, object detection and localization using known transmitter and receiver locations, and SLAM combining UE localization with environmental mapping.Traditional SLAM based only on backscattered signals is treated separately.
  • Positioning pipeline: Positioning generally uses dedicated pilots to estimate LoS angles, delays, and Doppler before mapping them to UE location and nuisance parameters.Nuisance parameters include clock bias, carrier-frequency offset, and unknown UE orientation.
  • Positioning pipeline: UE state recovery is a nonlinear estimation problem involving measured channel parameters, 3D location, and nuisance parameters, commonly solved with least-squares or maximum-likelihood methods.Because the problem is generally non-convex, heuristics, approximations, or relaxations may be used to seek the global optimum.
  • Minimal problems: Four delay observations from synchronized single-antenna BSs are sufficient to determine 3D UE location and 1D clock bias.The example uses LoS time-of-arrival estimates from each BS; correlated TDoA measurements can remove clock bias.
  • Reducing infrastructure dependence: Additional angles, carrier phase, multipath, and new beacon technologies can reduce reliance on multiple connected BSs or improve positioning accuracy.Carrier phase provides extremely precise but ambiguous location information, while resolvable NLoS paths can improve positioning.
  • Minimal problems: A single-antenna UE with two synchronized BSs can yield a unique 3D location from intersecting lines, while a planar UE array can additionally determine 3D orientation.The examples illustrate how complementary measurements support minimal positioning and pose-estimation problems.

3) Positioning in sub-6 GHz:

Sub-6 GHz positioning is constrained by limited bandwidth, small UE arrays, and rich site-specific multipath, making conventional signal processing unreliable. Machine learning and super-resolution methods address these constraints, while mmWave benefits from larger bandwidths, arrays, and sparser channels.

  • Sub-6 GHz limitations: Small UE arrays provide limited angle resolution, preventing accurate UE orientation estimation, while larger BS arrays can improve resolution when paths are angularly separated.Massive-MIMO BSs commonly use roughly 64-128 antenna elements.
  • Sub-6 GHz limitations: Rich multipath obscures site-specific geometry, so conventional statistical channel models are questionable for positioning and conventional FFT or correlation methods perform relatively poorly.The channel includes many path clusters affected by shadowing, diffraction, reflections, and scattering.
  • Mitigation strategies: Machine-learning fingerprinting can reduce location errors below 10 meters, but requires labeled fingerprint-location pairs and incurs training complexity.Super-resolution methods provide a second direction by resolving closely spaced paths when the SNR supports it.

6) Machine learning for positioning:

Machine learning extends positioning beyond geometry-based methods by learning site-specific channel fingerprints and temporal patterns. The surveyed designs include fingerprint regression, transfer learning, hybrid geometric-neural approaches, and attention-based tracking.

  • Motivation: Geometric localization degrades in indoor and outdoor scenarios because multipath and interference disrupt the channel-to-environment relationship.
  • Fingerprinting: CSI fingerprinting uses an offline database of channel parameters and associated locations, then infers position online by comparing or mapping new fingerprints.
  • Fingerprinting: ML-based fingerprinting learns the fingerprint-to-position mapping offline and performs online regression without storing or searching the full database.Transfer learning can reconstruct an outdated fingerprint database using old fingerprints and a small number of new measurements.
  • ML-based positioning example: V-ChATNet achieves 20 cm accuracy for 95% of users across combined LoS and NLoS channels, significantly outperforming Kalman Filtering.The tracking network uses previous channel and position estimates to refine the new position.

7) Sidelink positioning:

Sidelink positioning uses UE-to-UE measurements to estimate absolute or relative position and can complement conventional methods in challenging environments. Its operation comprises group configuration, signal measurement, and position calculation, with cooperative localization extending the framework to multiple UEs.

  • Sidelink positioning: Sidelink positioning fuses link-level measurements between UEs to calculate a target UE’s absolute or relative position.Processing can occur at a network location server or at the UE itself.
  • Sidelink positioning: The sidelink procedure has configuration, signal transmission and measurement, and position calculation stages.
  • Sidelink positioning: During measurement, UEs exchange positioning reference signals; position calculation then uses time-of-arrival and angle-of-arrival information.
  • Applications: Sidelink positioning is presented as a promising complement to GNSS for V2X applications.
  • Cooperative localization: Cooperative localization jointly infers multiple UE positions from network measurements and gains relative-position information through UE-UE links.The framework includes theoretical limits, cooperation-efficiency analysis, and resource-allocation strategies.

9) Carrier phase positioning:

Carrier-phase measurements support highly accurate ranging and positioning, but their use is complicated by integer ambiguity. Sensing and tracking additionally require detection, complex-state estimation, data association, and multi-sensor fusion.

  • Carrier phase positioning: Few-millimeter ranging accuracy is feasible in millimeter-wave networks, while C-band networks can achieve centimeter-level ranging accuracy.Relative carrier-phase measurements enable delay estimation and ranging with accuracies that are fractions of a wavelength.
  • Carrier phase positioning: Carrier phase is insensitive to integer multiples of the wavelength, creating an integer ambiguity that prevents direct measurement of full wavelengths in TX–RX distance.Differential or double-differential measurements using a reference device can solve or relax this ambiguity.
  • Carrier phase positioning: Bayesian filtering combined with cellular carrier-phase measurements can support super-resolution and low-latency 6DoF tracking of 5G-enabled XR headsets without additional sensors.Carrier-phase methods are also conceptually related to GNSS real-time kinematic positioning.
  • Sensing and tracking: Unlike positioning, sensing must detect an unknown number of objects amid clutter and estimate more complicated states that may include velocity, shape, extent, and material type.Moving objects are handled through multi-target tracking, while static objects are treated as landmarks.
  • Sensing and tracking: Mapping and multi-target tracking require associating measurements at time t with objects detected at earlier times, producing a growing combinatorial problem that needs dedicated complexity-mitigation routines.Measurements themselves carry no object identity information, and objects can generate multiple measurements.
  • Sensing and tracking: Phase-coherent receivers can form a distributed sensor using aggregated raw I/Q waveforms, whereas unsynchronized receivers fuse information after local channel-parameter estimation.Different receiver fields of view make multi-sensor fusion challenging.

V. TECHNOLOGIES FOR JOINT MONOSTATIC SENSING AND COMMUNICATION

Monostatic ISAC uses collocated transmission and reception to let cellular nodes sense their surroundings independently, but simultaneous operation requires full-duplex reception and strong suppression of self-interference. Separate antenna systems, cancellation techniques, beamforming, and hardware-aware modeling address these requirements.

  • Monostatic sensing: Monostatic sensing can turn UEs and gNBs into cellular radars that provide standalone situational awareness without relying on other network entities.A node can use its own transmit waveform to illuminate landmarks while tracking its own coordinates relative to a reference point.
  • Monostatic sensing: Cellular monostatic sensors must receive while transmitting because millisecond-scale TDD transmit periods are too long for meaningful radar waveform interaction at distance.This makes the sensor an in-band full-duplex, or simultaneous transmit-and-receive, system.
  • Self-interference suppression: Direct TX–RX coupling can behave as an extremely powerful short-range target that masks other targets unless suppressed through antenna, RF, and baseband cancellation.Isolation must be established before the LNA to prevent self-interference from desensitizing or blocking the sensing receiver.
  • Self-interference suppression: Separate TX and RX antenna systems can provide larger isolation than circulators, whose isolation is commonly around 20 dB, by spatially separating the antenna systems.The discussion primarily assumes separate antenna systems, while RF and baseband cancellation principles apply to both arrangements.
  • Beamforming and hardware: TX and RX beamforming both contribute to observable self-interference, with TX beamforming able to control the interference waveform already at the RX LNA input.Optimization can trade off self-interference suppression, target illumination, and achievable communication rate.
  • Beamforming and hardware: Hardware-aware models enable digital cancellation of nonlinear distortion from power amplifiers, LNAs, and other analog components, while oscillator phase noise may also become useful for sensing.The chapter treats accurate hardware modeling as important for receiver dynamic-range optimization and sensing design.

C. Triple-function precoding/combining in full duplex transceivers

Full-duplex monostatic ISAC requires precoders and combiners to communicate, illuminate targets, and suppress self-interference under hybrid-architecture constraints. SI-aware codebooks and joint optimization address these coupled objectives, enabling target detection where conventional initial-access beams fail.

  • Triple-function design: The TX and RX spatial filters jointly support downlink communication, high gain toward targets, and self-interference suppression.A multibeam precoder illuminates the target angle and multiple communication-channel paths.
  • Triple-function design: Joint precoder–combiner design maximizes downlink sum rate over subcarriers subject to target-gain, self-interference-suppression, hardware, and power constraints.The optimization is difficult because precoders and combiners are coupled and constrained by analog hardware.
  • Optimization approaches: Alternating optimization, convex approximations, semidefinite methods, null-space projection, analog cancellation, and hybrid decomposition are used to handle the coupled design problem.Different approaches trade rate, target gain, residual SI, and implementability under phase-shifter or subarray constraints.
  • SI-aware codebooks: Initial directional codebooks produce high residual self-interference, motivating SI-aware TX and RX codebooks for sensing at a full-duplex base station.The codebooks are designed over angular grids with unit-modulus entries so they can be realized using phase shifters.
  • SI-aware codebooks: SI-aware codebooks efficiently suppress self-interference and yield high radar SINR at the target angle, whereas the initial codebook makes target detection impossible.The example uses two 8 × 8 planar arrays separated by 10λ in a single-point-target scenario.
  • Access and tracking: The same SI-aware codebook can support tracking, while initial access reduces the triple-function requirement because communication and sensing resources need not yet be shared.Beam training for initial access uses wide angular coverage similar to radar target discovery.

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

The paper models joint monostatic OFDM sensing and communication, showing that shared-oscillator impairments affect sensing differently from communications. ICI and PN can degrade detection, but their Doppler- and delay-dependent structure can also provide additional sensing information.

  • System model: A monostatic ISAC transceiver sends data to a remote receiver while a co-located radar receiver senses environmental objects.The setup uses shared transmitter and radar hardware, with a remote communications receiver using an independent oscillator.
  • System model: Range-Doppler processing removes data-symbol effects and applies IDFT across subcarriers and DFT across symbols to obtain a sensing spectrum.CFAR, MUSIC, and ESPRIT can then support detection and delay-Doppler estimation.
  • ICI: ICI increases sidelobe levels and can mask weaker targets, especially for high-mobility scenarios or small subcarrier spacing.The standard model loses validity when intra-symbol Doppler shifts destroy subcarrier orthogonality.
  • PN: PN distorts subcarrier orthogonality and reduces radar dynamic range, degrading detection performance for weak targets.In monostatic sensing, shared-oscillator PN has delay-dependent statistics that can convey unambiguous range information.
  • ICI: ICI also supplies an additional Doppler dimension that can resolve high-speed Doppler ambiguity and distinguish targets sharing a delay-Doppler-angle bin.The associated unambiguous velocity is N times greater because of the higher-frequency time-domain sampling.
  • PN: Exploiting the delay dependence of PN can make radar performance exceed that of an ideal radar without PN.The paper presents this as repurposing a traditionally detrimental impairment into a sensing benefit.

VI. SENSING AND COMMUNICATING WITH WIDE APERTURES

Extremely large apertures and near-field propagation expand the joint communication-and-sensing capabilities of 6G systems. Spherical wavefronts provide distance information and enable focusing, high-rank LoS communication, and fine localization.

  • Wide apertures: XL-MIMO increases antenna counts to hundreds or thousands, producing electrically large apertures with improved spectral efficiency and spatial resolution.The paper uses ELAA for antennas whose physical dimension D is much larger than the wavelength λ.
  • Near-field operation: At millimeter-wave and higher frequencies, the radiating near-field can span practical distances of several or hundreds of meters.This can occur even with physically small but electrically large antennas.
  • Near-field operation: Spherical near-field wavefronts encode distance in addition to angle, supporting focusing and richer sensing and positioning.The near-field phase profile can be analyzed to determine a transmitting source’s position.
  • Near-field operation: ELAA focusing can discriminate users at the same viewing angle by controlling interference in both angular direction and distance.In contrast, far-field beam steering primarily distinguishes users by angle.
  • Communication degrees of freedom: Diffraction-based focusing supports multiple communication modes, and ELAA can provide high-rank communication even in strong LoS conditions.Capacity-approaching near-field MIMO requires intricate array phase profiles that can be approximated with multiple focused beams.
  • Localization: A single near-field ELAA can localize a source through the impinging-wave phase profile, while an array-equipped source enables 6D position-and-orientation estimation.The paper also notes that narrowband signals can support ELAA localization and sensing without ad-hoc synchronization or time-base ranging protocols.

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

Distributed MIMO extends sensing and localization across a large, irregularly deployed aperture, improving coverage and multipath resolution while imposing stringent synchronization and location requirements. RIS-based sensing and localization further support near-field and NLoS operation.

  • Distributed MIMO: D-MIMO uses radio units distributed over a wide area and coordinated through distributed and central units with varying processing and synchronization architectures.Its operation otherwise resembles massive MIMO, including pilots, multi-user beamforming, and uplink data transmission.
  • Distributed MIMO: Localization and sensing require phase stability plus radio-unit locations known within a fraction of the wavelength.These requirements place users and objects in the near field of the distributed array while they remain in the far field of individual elements.
  • Distributed MIMO: D-MIMO provides more uniform communication, localization, and sensing coverage, with enhanced multipath resolution and accuracy.The larger aperture comes at the expense of high computational complexity.
  • RIS localization: A near-field RIS response observed across time can enable 3D UE localization despite blocked LoS or failing RIS elements.The model represents the near-field response as a function of the UE position.
  • RIS localization: RIS-reflected delay measurements can keep localization error within 20−30 cm along a trajectory despite obstacles partially shadowing the RIS.A narrowband phase-profile version of the algorithm is also proposed.
  • RIS sensing: RIS configurations can scan an area so a monostatic base station detects and localizes targets using reflected BS-RIS-target paths, even with limited bandwidth.The approach exploits the RIS’s high spatial resolution and dynamically varying reflection configurations.

VII. TECHNOLOGIES FOR SENSING ASSISTED COMMUNICATION

Sensing-assisted communication addresses blockage and adaptation challenges in high-frequency, large-array systems by using position, radar, and learned environmental information. The reviewed strategies reduce beam-training overhead and improve effective communication rates, while radar–communication mismatches remain a constraint.

  • Sensing-assisted operation: ISAC maps combine environmental localization with past configuration outcomes to build position-dependent beam codebooks and track potential blockers.These maps support proactive communication adaptation to blockage.
  • Beam training: Location-informed beam selection achieves an overhead reduction of (1 −|L|/NT) by restricting search to beams covering estimated angular uncertainty.The approach computes a reduced codebook using the estimated user direction and its error ΔϕUE.
  • Beam training: Position-based and machine-learning methods reduce beam-search overhead by testing smaller sets of location-dependent beam pairs in LOS and NLOS channels.Inverse fingerprinting learns beam-pair subsets from past measurements, with online learning refining recommendations.
  • Radar-assisted configuration: Radar covariance can guide vehicular beam configuration because radar and communication channels at nearby mmWave frequencies have similar main directions.This provides a statistical-CSI prior rather than requiring full instantaneous CSI.
  • Radar-assisted configuration: 77% communication-overhead reduction is achieved using radar covariance in a ray-traced vehicular environment.Deep learning can translate radar covariance into communication covariance to compensate for channel and hardware mismatches, yielding further overhead reductions and higher effective rates.
  • Radar-assisted configuration: Radar-aided beam-training gains remain significant across coherence times, while the benefit of machine-learning-based covariance translation varies with operating conditions.The comparison includes exhaustive search, uncompensated radar-covariance training, and mismatch-compensated strategies.

3) Vision-aided beam training:

Vision-aided sensing uses cameras and other sensors to support beam training, beam prediction, and blockage management in dynamic wireless environments. The reviewed approaches combine environmental information with learning to reduce overhead, improve beam decisions, and anticipate link disruptions.

  • Vision-aided beam training: 3D scene reconstruction and deep neural networks can use camera views and UE position to predict selected beam indices.The reconstructed environment captures locations of scatterers, while ray tracing simulations show that several future beams can be accurately predicted.
  • Vision-aided beam training: Camera images provide spatial information about users, reflectors, and scatterers that can support beam selection and tracking in dynamic environments.Images may be fused with UE position and other information to enhance beam selection and determine line-of-sight conditions.
  • Vision-aided beam training: UE-mounted cameras can detect surrounding objects and vehicles, provide beam decisions, and predict the beam coherence interval.Simulation results using 3D modeling and ray tracing found UE images more beneficial than BS camera images.
  • Multimodal sensing: LIDAR, cameras, GPS, depth maps, and position data can be combined with learning models to identify obstacles, classify LOS or NLOS conditions, and reduce beam-training overhead.LIDAR-based designs can recommend a reduced beam set, while curriculum and federated learning improve classification or reduce later-user overhead.
  • Blockage prediction and management: Sensor-aided blockage management detects or predicts blockages and supports proactive actions such as BS selection, frequency handover, and link management.The reviewed work reports effective sensor-based and machine-learning-based blockage detection and prediction, including experimental and ray tracing-based evaluations.
  • Conclusion: The article presents ISAC as a 6G network vision encompassing network sensing, sensing-aided communication, bistatic and multistatic sensing, monostatic sensing, radio SLAM, and sensor-aided beam training.The overview emphasizes sensing-assisted beam training for overhead reduction and argues that communication and sensing should be considered together.
Loading 2405.01816v1…