PaperlayerView exact revision on arXiv

Source-linked AI summary

Channel Measurement and Ray-Tracing-Statistical Hybrid Modeling for Low-Terahertz Indoor Communications

Yi Chen, Yuanbo Li, Chong Han, Ziming Yu, Guangjian Wang

arXiv:2101.12436v1cs.IT

TL;DR

Indoor THz channel models require better evidence about propagation and multi-path behavior. This paper measures 130-143 GHz channels in a meeting room, clusters and matches measured MPCs with ray tracing, and develops a hybrid model. The hybrid model agrees better with measured PDAPs than statistical and 3GPP GSCM models.

  • Problem

    THz channel propagation and multi-path characteristics remain insufficiently characterized for accurate indoor channel modeling.

  • Method

    The paper combines wideband directional-antenna measurements, MPC clustering and matching, ray tracing, and statistical modeling for indoor THz channels.

  • Results

    The hybrid channel model has better agreement with measured PDAPs than the statistical model and 3GPP GSCM for all Tx-Rx positions.

  • Takeaways & Limitations

    The study supports a hybrid cluster-based model for representing indoor THz multi-path propagation in the meeting-room setting.

Abstract

from arXiv · show

TeraHertz (THz) communications are envisioned as a promising technology, owing to its unprecedented multi-GHz bandwidth. One fundamental challenge when moving to new spectrum is to understand the science of radio propagation and develop an accurate channel model. In this paper, a wideband channel measurement campaign between 130 GHz and 143 GHz is investigated in a typical meeting room. Directional antennas are utilized and rotated for resolving the multi-path components (MPCs) in the angular domain. With careful system calibration that eliminates system errors and antenna effects, a realistic power delay profile is developed. Furthermore, a combined MPC clustering and matching procedure with ray-tracing techniques is proposed to investigate the cluster behavior and wave propagation of THz signals. In light of the measurement results, physical parameters and insights in the THz indoor channel are comprehensively analyzed, including the line-of-sight path loss, power distributions, temporal and spatial features, and correlations among THz multi-path characteristics. Finally, a hybrid channel model that combines ray-tracing and statistical methods is developed for THz indoor communications. Numerical results demonstrate that the proposed hybrid channel model shows good agreement with the measurement and outperforms the conventional statistical and geometric-based stochastic channel model in terms of the temporal-spatial characteristics.

I. INTRODUCTION

The paper addresses missing understanding of 140 GHz indoor temporal-spatial propagation and develops measurement, clustering, and hybrid-modeling methods to characterize it. Measurements and simulations support a ray-tracing-statistical model that agrees well with measured multi-path behavior and outperforms conventional models.

  • The study targets incomplete characterization of THz propagation, including K-factor, sparsity, temporal and spatial dispersion, and correlations among multi-path features.
  • An MCD-based DBSCAN procedure clusters measured MPCs and matches them with ray-tracing simulations, outperforming conventional K-Power-Means on the measurement data.
  • The analysis examines path loss, reflection, multi-path sparsity, K-factor, delay and power spreads, angular spreads, and correlations among channel parameters.Power-delay-angular profiles reveal sparsity and the significance of wall reflections.
  • The proposed RT-statistical hybrid model combines ray tracing with statistical modeling and improves PDAP characterization over conventional statistical and 3GPP GSCM models.The model is reported to show good agreement with measured delay and angular spreads and improved PDAP performance.

II. CHANNEL MEASUREMENT CAMPAIGN

The campaign uses a calibrated 130-143 GHz VNA platform with directional horn antennas and ray tracing to measure and process indoor THz channels. Its 13 GHz bandwidth provides 76.9 ps time resolution, enabling separation of paths more than 2.3 cm apart.

  • The measurement system covers 130-143 GHz and uses VNA-based channel sounding with calibration to remove cable and RF-system effects.
  • Ray tracing and an MPC clustering-and-matching procedure are used to compare measured propagation with simulated MPCs.
  • 13 GHz bandwidth yields 76.9 ps time resolution and resolves two paths when their distance difference exceeds 2.3 cm.
  • The system records 1301 frequency points at 10 MHz spacing, giving a 100 ns maximum excess delay.
  • Directional horn antennas and motorized rotation units support angular measurements, with 30° transmitter and 10° receiver half-power beamwidths.

B. Meeting Room Environment and Measurement Deployment

Measurements are conducted in a furnished meeting room with two transmitter and multiple receiver deployments, while receiver scanning forms three-dimensional power-delay-angular profiles. The setup captures mainly desk, chair, and wall reflections within the measured elevation range.

  • Meeting room environment: The meeting room contains a central desk, eight chairs, two TVs, one glass wall, and three limestone walls.
  • Measurement limits: A 30 m maximum detectable path length permits recording reflected paths with at most three reflection orders.
  • Measurement deployment: The receiver scans azimuth from 0° to 360° with 10° spatial resolution, while the transmitter beam is directed toward the receiver.
  • Measurement deployment: The deployment uses two transmitter positions and twelve receiver positions across two measurement sets.
  • Propagation observations: Reflections from the ceiling and floor are 16 dB below the LoS path, so measured reflected paths mainly come from desks, chairs, and walls.
  • PDAP processing: The processing pipeline applies IDFT and combines delay and angular samples into three-dimensional and two-dimensional PDAPs.

D. MPCs Clustering and Matching with Ray-Tracing Simulator

The paper combines MCD-based DBSCAN clustering with ray-tracing MPC matching to recover physically meaningful THz propagation clusters while handling outliers and irregular cluster shapes.

  • Procedure: The procedure clusters measured and ray-traced MPCs, then matches their clusters to connect observed components with physical propagation paths.Ray-tracing MPCs are matched with measured MPCs after clustering.
  • Algorithm rationale: DBSCAN automatically selects cluster count, supports arbitrarily shaped clusters, and classifies isolated MPCs as noise.These properties address limitations of K-means-like methods that assign every MPC to a cluster.
  • Algorithm rationale: The method is suited to sparse THz channels but struggles when clusters are close together.The authors also note sensitivity to threshold selection in conventional MCD matching.
  • Procedure: MCD replaces Euclidean distance to account for circular AoA and jointly measure angular and temporal separation.The distance uses AoA and ToA components with delay scaling.
  • Comparison: Compared with KPM, the proposed method avoids merging sparsely separated MPCs and splitting the nearby LoS cluster.The comparison uses the same number of clusters for fairness and identifies matched clusters for physical interpretation.
  • Physical interpretation: Ray-tracing interpretation maps clusters to LoS, desk, chair, wall, and multiple-reflection paths, supporting physically grounded channel analysis.The procedure separates matched, non-matched, and outlier clusters; non-matched clusters can arise from simplified geometry or tracing configuration.

III. THZ CHANNEL CHARACTERIZATION AND ANALYSIS

The paper characterizes THz indoor channels through path loss, reflection behavior, and temporal-spatial multipath features. These measured properties are presented as guidelines for THz communication system design.

  • Scope: The characterization examines LoS path loss, MPC reflection properties, and temporal and spatial multipath features.The analysis is based on processed measurement data.
  • Scope: The revealed channel properties and physical parameters are intended as guidelines for THz communication system design.
  • Path-loss comparison: Figure 6 compares measured path loss with a fitted CI model and FSPL at 140 GHz.

A. Path Loss Model

The path-loss analysis fits a close-in model to 140 GHz meeting-room measurements and examines how reflection order changes reflected-path loss. Wall reflections generally carry more energy than obstacle reflections, with geometry and obstructions affecting the balance.

  • Path-loss model: The CI path-loss model uses the Tx–Rx distance, a 1 m reference distance, and shadow-fading variation around the fitted path-loss relation.The model is developed from measured received power, while FSPL is computed using Friis’ law.
  • Path-loss model: PLE = 1.75 and σD = 3.44 dB for the 140 GHz LoS channel in a meeting room.
  • Reflection loss: One reflection adds 6 dB to 9 dB of loss, consistent with the measured drywall relative dielectric constant ϵr = 6.4 at 140 GHz.
  • Reflection categories: Furniture can block wall-reflection paths and create obstacle-reflection paths, while receiver geometry changes the relative wall-reflection power.A screen near wall 4 lowers Rw for Rx2–Rx6 by blocking a first-order wall-reflection path.
  • Reflection categories: Wall-reflection MPC power exceeds obstacle-reflection MPC power at all receivers except Rx2 in measurement set 2.The exception is associated with strong non-line-of-sight paths reflected by nearby desks and chairs.

D. Temporal and Spatial Features of THz Multi-path Propagation

The measured 140-GHz indoor channel is sparse but strongly shaped by line-of-sight and critical reflected paths, producing substantial variation in temporal and spatial characteristics across receiver positions.

  • Temporal and spatial dispersion: The channel contains around ten clusters, while delay and angular spreads vary among receivers but less substantially than the K-factor.Small delay spread confines received power temporally, and small angular spread confines it spatially.
  • LoS blockage: LoS blockage makes NLoS multipath significant, with considerable energy contributed by reflected MPCs despite overall channel sparsity.The highly directional transmit beam can cause severe service interruption when the LoS path is blocked.
  • LoS dominance and multipath power: K-factor values range from 8.59 to 56.43 in measurement set 1 and from 6.04 to 326.94 in measurement set 2.The highest values occur at shorter Tx-Rx separations, where reflected paths travel substantially farther than the LoS path.
  • LoS dominance and multipath power: Reflected MPCs are weakened by longer propagation distance, reflection loss, and beam misalignment, yielding received power at least 30 dB below the LoS path for obstacle reflections.A wall reflection is 25 dB below LoS, while obstacle reflections incur additional free-space loss and antenna-gain attenuation.
  • Receiver-position dependence: Receivers with nearly identical geometry can have sharply different characteristics because strong or obstructed wall-reflection MPCs alter K-factor, delay spread, and reflected-power ratio.Rx7 has a delay spread of 11.152 ns and the smallest K-factor, whereas Rx6 has the smallest delay spread and largest Rw.
  • Statistical characterization: Log-normal fits show no noticeable difference between the two measurement sets, indicating statistically consistent temporal and spatial characteristic parameters.The fitted parameters include cluster count, K-factor, delay spread, angular spread, and Rw.

E. Intra-cluster Characteristics

Intra-cluster spreads are substantially narrower than overall channel spreads, while correlations reveal how separation, LoS dominance, and wall reflections shape temporal and spatial multipath behavior.

  • Cluster spreads: Mean CDS and CAS are 0.35 ns and 5.93°, compared with mean DS and AS of 4.23 ns and 34.12°.Thus, individual clusters occupy narrower temporal and angular regions than the aggregate channel.
  • Correlation analysis: Distance is negatively correlated with cluster count and K-factor, indicating greater channel sparsity and lower LoS dominance at longer separations.The measured explanation is that the LoS power weakens relative to NLoS MPCs as Tx-Rx separation increases.
  • Correlation analysis: Cluster count has weak correlation with K-factor, with correlation coefficient 0.12, and is uncorrelated with delay and angular spreads.This indicates that cluster count does not determine MPC power distribution or temporal and spatial dispersion.
  • Correlation analysis: The delay-spread and angular-spread correlation coefficient is 0.44, showing that temporal and spatial dispersion are related.The receiver comparison further illustrates that similar separation distances can produce substantially different multipath characteristics.
  • Correlation analysis: K-factor is negatively related to DS and AS, whereas Rw is positively correlated with distance and can make wall reflections dominate other NLoS MPCs.Strong LoS power confines received energy in temporal and angular domains, while distant receivers require careful treatment of wall-reflection power.

IV. HYBRID CHANNEL MODELING FOR THZ INDOOR PROPAGATION

The proposed RT-statistical hybrid model combines deterministic dominant paths from room geometry with statistically generated cluster subpaths and additional non-RT clusters, then validates the result against measurements and baseline models.

  • Hybrid model construction: The model uses ray tracing to generate one LoS path and several wall-reflection paths, each serving as the center of a cluster.The RT component requires the room dimensions and Tx and Rx positions.
  • Hybrid model construction: A statistical component supplements intra-cluster subpaths in RT clusters and generates additional obstacle-reflection and non-RT clusters.The spatial channel impulse response is formed from deterministic RT and statistical CIR components.
  • Measurement-driven scope: The model excludes elevation AoA and Tx AoD because the measurements scan only limited elevation AoA and do not scan AoD.The modeled variables include delay and azimuth AoA, with elevation composed and then eliminated from the model.
  • Statistical component: Statistical parameters are extracted from directional-antenna measurements, so the transmit antenna pattern is included in generated amplitudes rather than separately in the statistical CIR.The RT CIR retains the transmit antenna pattern explicitly.

1) Number of clusters:

The statistical model describes distance-dependent non-RT cluster counts and probabilistic subpath structure, with log-normal counts, Poisson arrivals, Von Mises angles, and power-law amplitudes.

  • Number of clusters: Far-separated Tx-Rx pairs contain fewer non-RT clusters because cluster count is linearly related to Tx-Rx distance.The number of clusters is fitted with a linear model and rounded using a ceiling function.
  • Subpath counts: The numbers of pre-cursor and post-cursor subpaths in RT and non-RT clusters follow log-normal distributions.The post-cursor count in a non-RT cluster is illustrated with a CDF and log-normal fit.
  • Arrival processes: Inter-cluster and intra-cluster MPC arrival intervals are independently exponentially distributed under Poisson arrival processes.Separate inter-cluster and intra-cluster arrival rates parameterize the two processes.
  • Angular distributions: Azimuth AoAs for inter-cluster MPCs and intra-cluster subpaths follow Von Mises distributions parameterized by location and concentration.The parameters are expressed in radians for the modeled AoA distributions.
  • Amplitude and phase: Inter-cluster and intra-cluster MPC amplitudes use separate coefficient-and-exponent representations, while each MPC phase is independently uniform.The coefficients and exponents distinguish inter-cluster behavior from intra-cluster subpath behavior.

C. Model Validation and Evaluation

The proposed hybrid model is validated against measurements and conventional baselines using delay/angular spreads and power-delay-angular profiles. It shows good agreement across temporal and angular characteristics and substantially better PDAP similarity than the comparison models.

  • Validation setup: The evaluation compares the proposed hybrid, conventional statistical, and 3GPP TR 38.901 models with measured channel data.The measured and simulated results are assessed using delay spread, angular spread, and PDAP-based metrics.
  • DS and AS validation: Good agreement in delay spread and angular spread indicates that the three models characterize temporal and angular small-scale fading.These quantities are used to compare model behavior in the temporal and angular domains.
  • PDAP validation: 3.65 dB, 4.22 dB and 5.73 dB are the mean RMSE values for the proposed RT-statistical, 3GPP GSCM, and conventional statistical models, respectively.The proposed model has the lowest RMSE, while the conventional statistical model has the highest.
  • PDAP validation: 0.51, 0.05, and 0.14 are the mean SSIM values for the proposed RT-statistical, conventional statistical, and 3GPP GSCM models, respectively.The proposed model’s SSIM is approximately 10 times higher than the conventional statistical model’s value.
  • Overall evaluation: The proposed hybrid model accurately captures power distributions in both temporal and angular domains and outperforms the conventional and 3GPP GSCM models.RMSE measures absolute PDAP error, while SSIM evaluates structural similarity between PDAPs.

D. Comparison and Discussion

The discussion attributes the hybrid model’s stronger PDAP agreement to ray tracing that preserves geometry-dependent spatial consistency while statistical modeling completes the multipath characteristics. This improves fidelity relative to conventional statistical and GSCM approaches, with added but limited computational complexity.

  • Comparison: The conventional statistical model performs poorly for PDAP because it uses transmitter-receiver separation without propagation-environment geometry and assigns ToA and AoA independently.Its RMSE and SSIM performance is described as unacceptable in the comparison.
  • Discussion: 3GPP GSCM preserves the measured line-of-sight path but randomly generates the scattering environment, leaving remaining multipath components inconsistent with measurements.It is more accurate than the conventional statistical model but still performs worse than the proposed hybrid model.
  • Complexity: The hybrid model requires transmitter-receiver positions and room dimensions for ray tracing, introducing additional computational complexity relative to the conventional statistical and GSCM models.The ray-tracing burden is limited because only a small number of line-of-sight and wall-reflection rays, with up to triple reflections, are traced.
  • Channel interpretation: LoS and wall-reflection paths dominate the meeting-room THz channel, producing high K-factor and Rw values and motivating the hybrid model’s deterministic component.Directional scanning and MPC processing support analysis of these spatial propagation characteristics.
  • Comparison: The proposed RT-statistical hybrid model achieves better agreement with measured PDAPs across all transmitter-receiver positions than the statistical and 3GPP GSCM models.The improvement is linked to the deterministic ray-tracing component.
  • Discussion: Ray tracing introduces spatial consistency and captures the most significant propagation paths, while the statistical component completes the multipath characteristics.This division combines geometry-dependent dominant paths with statistical modeling of the remaining characteristics.
Loading 2101.12436v1…