PaperlayerView exact revision on arXiv

Source-linked AI summary

A Tutorial on Environment-Aware Communications via Channel Knowledge Map for 6G

Yong Zeng, Junting Chen, Jie Xu, Di Wu, Xiaoli Xu, Shi Jin, Xiqi Gao, David Gesbert, Shuguang Cui, Rui Zhang

arXiv:2309.07460v2cs.ITeess.SP

TL;DR

6G makes real-time CSI acquisition increasingly difficult, while generating location-tagged channel data that can reveal local wireless environments. This tutorial develops CKM-based environment-aware communications, surveys construction and usage methods, and highlights evidence that CKM can reduce training overhead, including experimentally comparable performance to exhaustive beam sweeping over 4096 beam pairs. Practical deployment remains constrained by hardware and signal-processing limitations in some target scenarios.

  • Problem

    6G network density, antenna size, bandwidth, mobility, and intelligence make acquiring real-time CSI for all links increasingly challenging.

  • Method

    The paper presents CKM-based environment-aware communications, using historical location-channel data to construct and utilize maps without requiring explicit environment modeling.

  • Results

    CKM reduces channel-training overhead and supports environment-aware inference, with one prototype achieving performance comparable to exhaustive sweeping over 4096 beam pairs.

  • Takeaways & Limitations

    CKM provides a framework for using location-tagged channel data in proactive, environment-aware 6G communication design.

  • Takeaways & Limitations

    CKM-based approaches remain practically constrained by devices and architectures with limited hardware or signal-processing capabilities.

Abstract

from arXiv · show

Sixth-generation (6G) mobile communication networks are expected to have dense infrastructures, large antenna size, wide bandwidth, cost-effective hardware, diversified positioning methods, and enhanced intelligence. Such trends bring both new challenges and opportunities for the practical design of 6G. On one hand, acquiring channel state information (CSI) in real time for all wireless links becomes quite challenging in 6G. On the other hand, there would be numerous data sources in 6G containing high-quality location-tagged channel data, e.g., the estimated channels or beams between base station (BS) and user equipment (UE), making it possible to better learn the local wireless environment. By exploiting this new opportunity and for tackling the CSI acquisition challenge, there is a promising paradigm shift from the conventional environment-unaware communications to the new environment-aware communications based on the novel approach of channel knowledge map (CKM). This article aims to provide a comprehensive overview on environment-aware communications enabled by CKM to fully harness its benefits for 6G. First, the basic concept of CKM is presented, followed by the comparison of CKM with various existing channel inference techniques. Next, the main techniques for CKM construction are discussed, including both environment model-free and environment model-assisted approaches. Furthermore, a general framework is presented for the utilization of CKM to achieve environment-aware communications, followed by some typical CKM-aided communication scenarios. Finally, important open problems in CKM research are highlighted and potential solutions are discussed to inspire future work.

I. INTRODUCTION

6G combines expanded spectrum, antenna arrays, network density, sensing, passive communication, and distributed intelligence to support demanding applications. These developments create both opportunities and challenges, especially the need for scalable channel acquisition under larger channel dimensions and constrained hardware.

  • 6G vision: 6G research targets autonomous vehicles, network-connected robots, and mixed reality through broader wireless capabilities than current 5G networks.The envisioned network includes ubiquitous wireless intelligence, global coverage, and connectivity for people, things, and intelligence.
  • Potential technologies: 6G technologies span cell-free massive MIMO, non-terrestrial networks, above-6 GHz spectrum, evolving MIMO, and delay-Doppler modulation.These technologies address user-centric coverage, aerial connectivity, high data rates, positioning accuracy, and high-mobility interference.
  • Integrated functions: Passive communication and integrated localization, sensing, and communication reuse infrastructure and radio resources while enabling sensing or localization outputs to support communication.Backscatter and symbiotic radio add passive devices, while ILSAC integrates localization, sensing, and communication functionalities.
  • Emerging trends: 6G trends include denser nodes, larger antennas, wider spectrum, lower-cost hardware, higher-quality positioning, and greater intelligence.Edge devices such as base stations and user equipment can process sensing data in distributed ways, while dense deployments support broader coverage and connectivity.
  • CSI acquisition challenge: Larger antenna arrays, wider bandwidth, dense-node joint processing, and mobility increase channel dimensionality, making complete real-time channel estimation costly or impossible.MIMO channels contain NSC matrices of dimension MUE × MBS, while antenna counts and bandwidth are expected to grow substantially in 6G.
  • Hardware constraints: Reducing RF-chain and component costs through analog or hybrid beamforming creates hardware and signal-processing limitations for training-based channel acquisition.Analog beamforming shares one RF chain across array elements, while hybrid architectures use fewer RF chains than antenna elements.

5) More intelligence:

6G's increasing intelligence, data availability, and channel complexity create both a CSI-acquisition challenge and an opportunity for environment-aware communications. CKM organizes location-specific channel knowledge to support communication design and reduce reliance on costly real-time inference.

  • 5) More intelligence:: 6G intelligence enables networks to analyze massive data from sensors, UEs, and IoT devices using advanced AI and computation.
  • 5) More intelligence:: Denser networks and larger-dimensional channels make real-time CSI acquisition harder, especially with simpler hardware and increasing training or beam-sweeping overhead.Pilot training and beam sweeping become more burdensome as channel dimensions and network density increase; analog beam sweeping can take seconds in a 64-antenna BS and 16-antenna UE example.
  • 5) More intelligence:: 6G also provides abundant location-specific channel data, including received signal strength, channel gains, AoA, and AoD, from BSs and mobile devices acting as sensors.
  • 5) More intelligence:: Environment-aware communication uses a priori knowledge of the actual local environment to improve communication-network design and address difficult real-time CSI acquisition.
  • 5) More intelligence:: CKM is a site-specific, location-tagged database of channel knowledge that can facilitate or even obviate sophisticated real-time CSI acquisition.The paper surveys CKM concepts and classifications, construction methods, utilization frameworks, and applications including training-free, light-training, predictive, localization, and sensing communications.

II. CHANNEL KNOWLEDGE MAP: BASIC CONCEPT

CKM maps transmitter and/or receiver locations to location-specific channel knowledge, providing the conceptual basis for environment-aware communications. The section introduces CKM organization, construction, utilization, and representative communication scenarios.

  • The paper covers CKM construction, performance evaluation, utilization, and typical communication scenarios.
  • Typical CKM-enabled applications include training-free and light-training communications, predictive communication, and localization and sensing.
  • CKM maps a location vector q ∈ R^D to channel knowledge for wireless links.
  • In BS-centric communication, a BS-maintained CKM can provide location-specific channel knowledge for multiple UEs across their trajectories.

B. Basic Principles of CKM

CKM infers location-specific channel knowledge from historical location-channel pairs, avoiding explicit modeling of the radio environment. It supports multiple communication modes and channel-knowledge types for environment-aware networking.

  • Environment awareness: Because previously visited locations tend to have similar wireless environments, CKM can reduce channel uncertainty and improve channel inference accuracy.UE mobility is identified as a major source of channel variation and can be characterized through trajectories.
  • CKM construction: CKM uses historical location-channel pairs to infer new channel knowledge without explicitly characterizing the environment or the underlying channel function.In quasi-static environments, the historical data can represent the local mapping and support CKM construction.
  • Data and storage: CKM construction may use data collected during communications or numerical computations, rather than requiring dedicated site surveys.The data can be gathered from multiple devices and stored hierarchically to mitigate network-wide storage demands.
  • Communication modes: B2X CKM supports BS-to-UE communications within coverage, whereas X2X CKM supports D2D links between potentially mobile transmitters and receivers.X2X CKM requires both transmitter and receiver locations, increasing its dimensionality and data-management requirements.
  • Channel knowledge types: CKMs may provide region-specific modeling parameters, location-specific large-scale knowledge, or fine-grained small-scale channel knowledge.Small-scale CKMs can include instantaneous channel gains, path parameters, or complete channel impulse responses, but require richer information.

D. CKM Typical Usage Scenarios

CKM is intended for channel-inference settings where real-time training is infeasible, costly, or insufficient, including unseen locations, non-cooperative nodes, high-dimensional channels, and constrained hardware. Compared with time-domain prediction, CKM can reuse data across devices and domains while reducing training overhead.

  • Unseen locations: CKM supports prediction for locations not yet or ever visited, enabling a network-wide view needed for proactive wireless-network decisions.Supported applications include predictive resource allocation, foresighted handover, and node sleeping or wake-up without periodic channel training.
  • Non-cooperative nodes: CKM can infer channels for non-cooperative nodes, unlike training-based acquisition that typically requires transmitter and receiver cooperation.The paper gives eavesdropping and cognitive-radio systems as examples where cooperation may be unavailable.
  • Larger-dimensional channels: CKM is useful for high-dimensional channels whose training, feedback, and processing costs grow with massive MIMO, wide bandwidth, mobility, and low-latency requirements.Location-specific prior knowledge can reduce the burden of acquiring such channels in real time.
  • Hardware limitations: CKM is especially appealing for semi-passive devices and analog, hybrid, or low-resolution architectures with limited hardware or signal-processing capabilities.These constraints make conventional training-based channel acquisition more difficult.
  • Comparison with existing inference: Unlike time-domain prediction and out-of-band inference, CKM can perform inter-device inference across time, frequency, and space using historical data from multiple devices.Time-domain prediction still needs frequent training, while out-of-band inference depends critically on cross-band correlation.

3) Channel-to-channel mapping:

Channel-to-channel mapping infers channel knowledge across frequency and spatial domains from a known channel, while CKM instead explicitly uses location information. The paper situates CKM among sensing-aided communications, localization methods, and ISAC, emphasizing its potential to reduce real-time acquisition overhead.

  • Channel-to-channel mapping: Channel-to-channel mapping infers a user’s channel for another antenna set and possibly another frequency from known CSI.Unlike CKM, it does not explicitly use user location information.
  • Channel-to-channel mapping: The bijective location-to-channel assumption in channel-to-channel mapping constrains the acquired antenna set and its geometric relationships.CKM does not rely on this assumption, broadening its applicable scenarios.
  • Sensing-aided communications: Radar-, lidar-, and vision-aided communications can reduce real-time training overhead but generally require additional hardware, waveforms, and signal processing.These requirements increase communication-system cost, size, and complexity.
  • Localization-related methods: CKM constructs site-specific databases and infers location-specific channel knowledge, conceptually reversing fingerprinting-based localization.The paper expects CKM to work indoors and outdoors because it relaxes the bijective mapping criterion.
  • CKM and ISAC: CKM and ISAC address different primary challenges: ISAC integrates communication and sensing, whereas CKM targets difficult CSI acquisition.ISAC can nevertheless support CKM construction and utilization by providing location-tagged data and device or scatterer localization.
  • Summary: CKM uses historical measurements or offline ray tracing to provide a priori local channel knowledge, potentially reducing real-time acquisition overhead and improving channel estimation quality.The paper presents CKM as a promising technique for environment-aware communications.

2) Measurement data:

Measurement-based CKM construction acquires location-specific data through dedicated offline campaigns or online network measurements, then applies spatial interpolation methods to estimate channel knowledge at unmeasured locations. The section compares canonical methods including IDW, KNN, local polynomial regression, Kriging, and kernel-based approaches.

  • Measurement data: Dedicated offline measurements use planned routes, controlled mobility, configured scenarios, and data cleaning, but require intensive labor.Online measurement can use UE radio measurements collected through minimization of drive tests.
  • Environment model-free methods: Environment model-free CKM construction treats the task as spatial interpolation or extrapolation without explicitly modeling geometry-induced correlations.These methods apply when parametric propagation models are unavailable.
  • Interpolation methods: Canonical interpolation methods include Kriging, kernel regression, matrix completion, KNN, IDW, and polynomial regression.Matrix completion exploits a low-rank assumption when the radio map is represented on a grid.
  • Canonical interpolation: IDW estimates a location by weighted averaging, with weights commonly proportional to inverse distance raised to a parameter α.Choosing α is difficult when the propagation field is complicated and prior information is limited.
  • Canonical interpolation: KNN averages measurements from the K nearest locations using IDW or kernel-based weights.Gaussian and Laplacian kernels are examples of possible weighting functions.
  • Canonical interpolation: Local polynomial regression adapts a polynomial model to each target location through weighted least squares, emphasizing nearby measurements.Cross-validation can select the weighting parameter σ because it strongly affects interpolation performance.
  • Kriging: Kriging constructs an unbiased estimator by selecting weights that minimize mean-squared error under a spatial statistical model.Simple, ordinary, and universal Kriging differ in their assumptions about the mean function.
  • Kriging: The Kriging solution is obtained from a linear system with an unbiasedness constraint, and its estimation variance has a closed form.The resulting predictor is described as the best unbiased prediction under the stated stationary, constant-mean assumptions.

3) Kriging for static CKM construction:

Kriging supports CKM construction for interference, SINR, and shadowing maps, and can be extended with state-space modeling to track time-varying propagation fields. The approach combines spatial interpolation with temporal filtering and requires suitable statistical or finite-dimensional representations.

  • Static CKM construction: CKM construction must jointly consider data acquisition, message passing, information processing, and resource allocation and control.Related map applications include interference, SINR, and shadowing maps.
  • Static CKM construction: Interference and SINR maps can be constructed by directly applying the Kriging principle to terminal measurements stored in a network database.The network architecture must reflect the type of CKM being constructed.
  • Static CKM construction: Shadowing-map construction first estimates large-scale shadowing from sensor RSS measurements, then computes interference using the shadowing map and secondary-user power allocation.Kriging interpolates the full shadowing map from limited estimated entries.
  • Time-varying CKM construction: Time-varying CKMs can exploit spatio-temporal channel correlation through a Kriged Kalman filtering approach.Measurements are modeled as the time-varying field plus measurement noise across sampled locations and time slots.
  • Time-varying CKM construction: The temporal model represents the field with spatially correlated, temporally white Gaussian processes and evolves it through a spatial correlation function.A basis expansion provides a finite-dimensional representation for the continuous state-space model.
  • Time-varying CKM construction: The resulting state-space formulation includes separate state-evolution and measurement equations for Kalman filtering.The measurement vector contains sampled field values plus measurement noise.

5) Kernel Regression:

Kernel regression offers a nonparametric framework for CKM estimation and is closely related to Kriging through kernel Gram matrices. Kernel methods can also reduce representation dimension and accommodate multi-scale channel structure.

  • Kernel Regression: Kernel-based function estimation provides a nonparametric regression framework, and RKHS and Kriging are equivalent when covariance and kernel Gram matrices coincide.This links kernel regression to statistical field-estimation methods.
  • Kernel Regression: RKHS represents interpolating functions as weighted sums of kernels centered at measurement locations.The kernel choice imposes structural properties; for example, spline kernels generate low-curvature functions.
  • Kernel Regression: The RKHS representation theorem yields a finite-dimensional optimal interpolator from the measurement samples.Its regularized least-squares coefficients are obtained by solving a Gram-matrix system.
  • Kernel Regression: Under the zero-mean stationary setting, the RKHS estimator corresponds to the Kriging estimator, with noisy measurements contributing a σ^2I term to the covariance matrix.The resulting estimator is expressed as a kernel-weighted combination of observations.
  • Kernel Regression: High-dimensional CKMs motivate low-dimensional representations before estimation and require attention to measurement quantization.These are identified as practical complexity considerations for complicated CKM data structures.
  • Kernel Regression: A narrowband power-spectrum model separates location-dependent path loss from transmitter PSD terms, reducing construction to a 2D power map when common PSD terms are known or estimated.The reduced map can then be reformulated using kernel regression.
  • Kernel Regression: A dual-kernel approach can track slowly varying path loss and fast-varying shadowing components in multi-scale wireless channel models.CKMs commonly focus on large-scale channel knowledge, while the dual kernel captures additional shadowing variation.
  • Kernel Regression: Matrix and tensor completion construct CKMs after discretizing the channel map over grid points, with non-uniform grids potentially improving identifiability for non-uniform measurements.The discretized map becomes a matrix or tensor with sparse observations assigned to grid cells.

7) Matrix and Tensor Completion:

Matrix and tensor completion exploit low-rank structure to reconstruct CKMs from sparse measurements, while interpolation assistance balances local information with global completion. Resolution and identifiability remain central constraints.

  • Matrix completion: Low-rank channel matrices can be completed from sparse observations under sufficient random measurements or suitable deterministic observation patterns.Uniformly random sampling requires O(M log^2 M) observations under the stated conditions.
  • Tensor completion: Tensor models extend CKM construction to higher-dimensional, frequency, and multimodal domains.A multi-emitter spectrum cartography model can use an LL1 tensor structure, reconstructed with alternating least squares.
  • Noise handling: Observation noise combines measurement noise with discretization noise caused by samples falling away from grid centers.Weighted least-squares objectives can address these errors while balancing data fidelity and nuclear-norm regularization.
  • Identifiability: Matrix completion faces a resolution–identifiability trade-off because a fixed measurement budget makes finer grids increasingly sparse.The paper notes an O(M log^2 M) sampling guideline but recommends choosing a small M with an empirical safety margin.
  • Interpolation assistance: Interpolation-assisted matrix completion combines local correlation from interpolation with global structure from matrix completion.A moderate, adaptive window size performs best; small and large windows approach pure matrix completion and full interpolation, respectively.

9) Deep Learning with Neural Network:

Deep learning and environment-assisted models construct CKMs by learning propagation relationships from measurements, geometry, or both. Environment reconstruction can represent electromagnetic effects through spatial loss fields, virtual obstacles, or segmented propagation regions.

  • Deep learning: Neural networks can replace difficult propagation models and use multipath geometry with CNNs to extrapolate angle and delay information.Satellite images and transceiver positions can also be encoded as feature images for U-net-based radio map prediction.
  • Environment-assisted modeling: Environment model-assisted construction reconstructs propagation geometry as an intermediate step, exposing correlations shared by measurements from the same environment.The resulting environment model can provide a full spatial CKM as a byproduct.
  • Spatial loss field: Spatial loss fields model incremental shadowing loss over locations and integrate path-dependent weights between transmitter–receiver pairs.Direct-path, normalized-ellipse, and inverse-area elliptical models provide alternative weighting functions.
  • Spatial loss field: The spatial loss field can be discretized into a linear system and estimated from measurements using matrix operations.The discretized model is written as s = Bg, with the estimate given as ĝ = B^-1s.
  • Virtual obstacle model: Virtual obstacles represent radio attenuation with electromagnetic classes rather than reconstructing the visual environment directly.Solid, light, or absent obstacles correspond respectively to deep shadow, slight attenuation, or line-of-sight propagation.
  • Segmented propagation model: Propagation regions can be learned without geometry or derived from simplified ray tracing, then used in segmented channel models parameterized by propagation and environment variables.Least-squares estimation can recover the model parameters and reconstruct the CKM.

D. Performance Evaluation

Performance evaluation uses reconstruction error on radio maps, comparing interpolation, regression, nearest-neighbor, and matrix-completion methods. The reported indoor experiment finds the interpolation-assisted NNM-t approach strongest among the tested methods.

  • Evaluation criteria: Radio-map construction is commonly evaluated using pixel-by-pixel MSE or MAE against the constructed CKM.The section evaluates both environment model-free and environment model-assisted approaches.
  • Indoor RSS experiment: The indoor RSS dataset contains 166 measurements over a 14×34 m^2 area, with 20–140 randomly selected samples used for reconstruction.The sampling rate is defined as M′/M.
  • Compared methods: Seven representative methods are compared, including local polynomial interpolation, Kriging, thin-plate splines, Gaussian-process regression, nearest neighbors, ALS, and NNM-t.NNM-t combines nuclear-norm matrix completion with trust-region constraints obtained from local polynomial interpolation.
  • Results: NNM-t performs best among the seven tested methods across the reported CKM reconstruction results.The comparison places NNM-t above pure matrix-based ALS and pure interpolation-based methods.

2) Environment model-assisted approaches for environment and radio map construction:

Environment model-assisted CKM construction combines propagation and geographic modeling to support radio-map reconstruction and CKM utilization. The approach addresses CSI acquisition challenges but involves data, computation, and identifiability trade-offs.

  • Construction evaluation: Joint environment and radio-map construction can be evaluated on simulated urban RSS measurements using neural and segmented propagation models.The compared models include U-net, RadioUNet, a segmented model, and a DL-based segmented model.
  • Construction evaluation: Virtual-obstacle maps generated from RSS measurements can closely match the geometry and distribution of structures in the true city map.The segmented model’s virtual-obstacle geometry is reported as closer to the true map than the DL-based method’s geometry.
  • Trade-offs: Environment model-free methods are efficient with sparse data but scale poorly for massive, high-dimensional maps, whereas model-assisted methods require more data and heavier training.Model-assisted approaches are described as easier to scale geographically for large CKMs.
  • CKM utilization: CKM utilization is formulated through an optimization framework for a generic system containing transmitters, receivers, cooperative nodes, and non-cooperative nodes.The framework optimizes node decision variables over multiple channel coherence blocks to maximize system utility.
  • CSI acquisition: Conventional optimization is difficult because system utility depends strongly on current and future CSI that may be costly, inaccurate, or unavailable.Fast-changing channels reduce future-CSI prediction accuracy, while non-cooperative nodes cannot feasibly participate in conventional training.
  • CSI acquisition: CKM can support current CSI acquisition with light or no training, predict future CSI, and help obtain CSI for non-cooperative nodes.The paper presents this as an environment-aware alternative to conventional channel training and environment-unaware designs.

1) Conventional channel training based design:

Conventional communications acquire CSI through channel training, but the required signaling overhead grows with channel dimensions and network density. CKM instead infers current and future channel knowledge from locations and trajectories, enabling optimization without heavy training while allowing limited online refinement when estimates are inaccurate.

  • Training overhead: Channel training estimates each communication pair’s CSI using predetermined pilots, with overhead generally proportional to M_BS N_SC.The overhead becomes significant as the number of base-station antennas or subcarriers increases.
  • Training overhead: With K_1 transmitters and K_2 receivers, orthogonal-pilot training can require K_1K_2 times the single-pair overhead.This scaling is described as unacceptable for sufficiently dense future networks with massive numbers of devices.
  • CKM-based design: CKM infers channel knowledge from current node locations and predicted trajectories, then uses those estimates to optimize wireless communication designs without or with limited channel training.The inferred current and future channel matrices approximate the utility terms in the original design problem.
  • CKM-based design: The CKM-based utility objective can be evaluated without channel training, enabling training-free communications through subsequent optimization of the design variables.This corresponds to solving the approximate utility-maximization problem using CKM-derived channel estimates.
  • Refinement: Training-free designs may degrade when CKM or localization errors make inferred channels differ from ground truth, motivating limited online training for refinement.Online training can exploit coarse CKM-based estimates rather than starting without prior channel knowledge.
  • Extensions: CKM can also support predictive communication, interference management with non-cooperative nodes, network planning, link scheduling, localization, and sensing.These uses extend beyond training-free and light-training communication design.

B. Training-Free and Light-Training Communications

The paper frames training-free and light-training communication through CKM-enabled design, using hybrid analog-digital beamforming in mmWave massive MIMO as a representative case. Conventional methods face CSI-acquisition and training-overhead challenges, whereas CKM supplies location- and environment-based information for beamformer design.

  • Hybrid beamforming example: The representative system is a point-to-point mmWave massive MIMO link using hybrid analog-digital beamforming with fewer RF chains than antennas.The transmitter and receiver jointly design analog and digital beamformers over coherent blocks to maximize achievable rate or system utility.
  • Conventional approaches: Hybrid beamforming depends on CSI acquisition, which is difficult for large-dimensional massive MIMO channels because analog and digital beamformers are coupled.This motivates alternatives to complete channel estimation before beamformer design.
  • Conventional approaches: A large training duration compromises achievable rate, while beam sweeping incurs overhead proportional to the number of possible beam combinations.Beam sweeping avoids explicit full-channel estimation but may require many candidate beam pairs.
  • CKM-enabled approaches: CKM enables hybrid beamforming without training or with only light training by exploiting node locations and environmental information.The design depends on the CKM type, including CMM, CPM, CAM, and BIM.
  • CKM-enabled approaches: CMM and CPM support training-free hybrid beamforming, whereas CAM and BIM support light-training designs.The following subsection distinguishes these CKM types according to the channel information or beamforming information they provide.

2) CMM and CPM enabled training-free hybrid beamforming:

CMM and CPM provide channel information that enables training-free hybrid beamforming, while CAM and BIM support light-training designs using complementary large-scale environment information and limited additional training. In the reported comparison, CAM- and BIM-enabled schemes improve with antenna count and outperform location-based beam alignment by reducing training overhead.

  • CMM and CPM: CMM directly maps transmitter and receiver locations to channel matrices, minimizing computation but requiring high storage for many channel matrices.In the fixed-BS mmWave massive MIMO case, the estimated UE location is used to obtain the channel estimate.
  • CMM and CPM: CPM reduces CMM’s storage cost by providing location-specific multipath information, including significant-path count, power, phase, and angle-of-arrival/departure data.These path parameters are used to reconstruct the MIMO channel matrix at the estimated UE location.
  • Training-free beamforming: CMM- or CPM-derived channel matrices let the BS and UE design hybrid beamformers without extra channel training.Performance depends on CKM and location accuracy, while dynamic scatterers and localization errors can reduce channel-estimation accuracy.
  • Light-training beamforming: CAM supplies path angles while additional training estimates path gains, reducing the unknown channel parameters to the significant-path gains.The CAM-based design uses angle information to construct transmit and receive array-response matrices before learning the path gains.
  • Light-training beamforming: BIM maps transmitter and receiver locations to candidate transmit and receive beamforming vectors and is expected to approach exhaustive beam sweeping with lower overhead.It is described as an advanced beam-sweeping technique empowered by CKM.
  • Performance comparison: As BS antennas increase, channel-estimation-based effective rate decreases drastically, whereas CAM- and BIM-enabled rates improve monotonically and outperform location-based beam alignment.The comparison attributes the contrast to the training overhead of channel estimation and its reduction under CKM-enabled schemes.
  • Other applications: CKM-enabled training-free and light-training designs also extend to IRS reflective beamforming and wireless power-transfer energy beamforming.CMM and CPM are associated with training-free designs, while CAM and BIM support light-training designs in the IRS example.

C. Predictive Communication for Yet-to-Reach Locations

CKM enables communications and mobility decisions using channel knowledge predicted at locations before nodes arrive. The section illustrates this through UAV trajectory and placement designs, as well as dense-network resource management and localization applications.

  • CKM predicts wireless channels along controllable moving trajectories before nodes reach those locations, supporting resource management and mobility control.
  • 1) CKM-assisted single-UAV trajectory design: For cellular-connected UAVs, an SINR map combining channel gains and interfering-BS loading supports trajectory optimization that avoids blockage and strong interference.The design minimizes mission latency while ensuring a minimum SINR over the flight path.
  • 2) CKM-assisted multi-UAV placement design: For multi-UAV placement, derivative-free optimization addresses the lack of an analytic CKM objective and seeks high network utility under mutual interference.
  • 2) CKM-assisted multi-UAV placement design: In a two-UAV example, each UAV is placed where its desired GBS link is strong and its interference link to the other GBS is weak, enhancing the sum-rate.
  • D. Resource Management for Non-cooperative Nodes and Dense Networks: CKM also supports link scheduling for dense D2D networks and expands localization fingerprints by incorporating heterogeneous channel information from base stations and neighboring devices.

V. OPEN PROBLEMS

The paper identifies data sufficiency, dynamic environments, heterogeneous data, continual updates, privacy, orientation, and experimental validation as open problems for CKM-enabled communications. It also connects CKM with digital twins and semantic communication.

  • Data sufficiency: Accurate CKM construction requires theoretical guidance on how much location-specific channel data is sufficient.Spatial statistics are suggested as a foundation for modeling and predicting channel-knowledge distributions across areas.
  • Dynamic environments: Dynamic wireless environments require CKM methods that adapt to changing conditions and remain robust to channel variation.The paper suggests combining sensory information, vision, and machine learning.
  • Data heterogeneity: Heterogeneous data from smartphones, autonomous vehicles, and UAVs complicate CKM construction because formats, granularity, quality, and redundancy differ.The paper calls for construction and utilization methods that explicitly account for these sources.
  • Continuous updates: Sequential data arrival creates an online-update problem, for which incremental learning and Recursive Least Squares are proposed as computationally efficient approaches.These methods update an existing CKM without recomputing or retraining the entire map.
  • Privacy: Practical CKM deployment requires privacy-preserving techniques because location data can reveal routines, whereabouts, and social interactions.The paper identifies virtual location and related techniques as possible remedies.
  • Orientation: Device orientation can be added as a CKM dimension, producing a tensor that captures channel responses under different orientations.This is particularly relevant when UEs use antenna arrays.
  • Validation: Prototype experiments are still needed, although one BIM-based system achieved performance comparable to exhaustive sweeping over 4096 beam pairs.The same strategy produced higher received power than location-based beam alignment in the reported scenarios.
  • Future connections: CKM can serve as a digital twin of the wireless environment and may interact with semantic communication through shared environment knowledge.The paper presents both directions as topics for future research.
Loading 2309.07460v2…