Pith. sign in

REVIEW 4 major objections 6 minor 43 references

A Mahalanobis-distance method that clusters and tracks multipath components at once yields smoother cluster evolution on 132 GHz channels.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 15:34 UTC pith:TT4OJI6L

load-bearing objection Solid simultaneous clustering-tracking method with clear gains on one sub-THz route; the D_th tuning to GCR/MSSD is real but does not erase the contribution. the 4 major comments →

arxiv 2607.04664 v1 pith:TT4OJI6L submitted 2026-07-06 eess.SP

A Simultaneous Clustering and Tracking Algorithm for Capturing Cluster-Level Spatial Consistency in 6G Wireless Channels

Jiaxin Lin , Pan Tang , Jianhua Zhang , Zhaowei Chang , Peijie Liu , Yufeng Qin , Ke Chen , Huixin Xu
show 1 more author
Byonghyo Shim
This is my paper
classification eess.SP
keywords spatial consistencyclustering and trackingMahalanobis distancemultipath componentschannel modelingsub-THz6Gcluster-level consistency
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Wireless channels change smoothly as radios move; at the cluster level that means multipath components with similar delay and angle should stay in the same groups so those groups can be tracked. Existing pipelines either cluster each snapshot alone then try to match the groups afterward, or they only carry forward cluster centers. This paper claims that associating each new multipath component to an existing cluster by Mahalanobis distance—using the full joint distribution of delay, angle, and transceiver location built from all previously assigned paths—does both jobs at once and produces more continuous tracks. On measurement-calibrated ray-tracing data at 132 GHz in an industrial setting, the method produces far fewer, longer cluster trajectories and lower successive-difference and gradient-change scores than a strong feature-matching baseline. If the claim holds, cluster-based 6G models (especially for massive MIMO, sensing, and sub-THz links) can be built from more reliable spatially consistent cluster parameters.

Core claim

On 132 GHz industrial ray-tracing channels, the Mahalanobis-distance simultaneous clustering and tracking (MD-SCT) algorithm associates new multipath components to existing clusters via the covariance of delay, angles, and positions, producing 17 continuous tracks of average length 11.894 m versus 104 fragmented tracks of 1.457 m for the multidimensional-feature-matching baseline, with gradient change rate reduced from 0.561 to 0.028 and lower mean-square successive differences on clustering indices and intra-cluster spreads.

What carries the argument

Mahalanobis-distance simultaneous clustering and tracking (MD-SCT): the distance of a new multipath vector from a cluster’s running mean and covariance (delay, azimuth/zenith angles, and TX/RX positions) decides assignment; assignment itself is the tracking step, with a threshold D_th for birth of new clusters from outliers.

Load-bearing premise

The single distance threshold (set to 400 after a sensitivity sweep that minimizes the same smoothness scores used as success metrics) plus a dense initialization segment generalizes so that the reported smoothness reflects true physical consistency rather than metric-tuned assignment on this route.

What would settle it

Re-run MD-SCT and the baseline on measured (not only ray-traced) double-directional multipath sequences at 132 GHz or another band, with D_th chosen without reference to the final GCR/MSSD scores; if the large gap in track length, GCR, and MSSD disappears or reverses, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes MD-SCT, a Mahalanobis-distance-based algorithm that assigns multipath components (MPCs) to existing clusters using the historical joint distribution of delay, angles, and transceiver positions (Algorithm 1, Eqs. (1)–(4)), thereby performing clustering and tracking simultaneously. After dense-region initialization with a conventional method (KPowerMeans, K1=10), new MPCs are accepted if D_Mah,i,k̃ < D_th and otherwise re-clustered as births. The method is applied to measurement-calibrated ray-tracing channels at 132 GHz in an industrial IoT scenario. Against the MFM tracking-after-clustering baseline, MD-SCT reports fewer, longer tracks (17 vs 104 clusters; avg. length 11.894 m vs 1.457 m), lower GCR (0.028 vs 0.561), and lower MSSD on DB, CH, and intra-cluster delay/angle standard deviations (Table I; Fig. 3). MSSD is introduced as a metric for successive consistency of clustering statistics.

Significance. Cluster-level spatial consistency is a recognized requirement for 6G channel models (massive MIMO, ISAC, THz). A simultaneous clustering-and-tracking rule that uses the full historical covariance of delay–angle–position, rather than only previous centroids, is a clear and useful algorithmic contribution. The algorithm is stated cleanly (Algorithm 1, Eqs. (1)–(4)), complexity reduction via Woodbury is noted, and the empirical comparison on a sub-THz IIoT route shows large continuity gains versus a published MFM baseline. Introducing MSSD as a sequence-level consistency metric is a modest but practical addition. If the superiority holds under non-circular threshold selection and broader validation, the work would strengthen spatial-consistency-aware cluster extraction for 6G modeling.

major comments (4)
  1. §III and Fig. 2: D_th is set to 400 because that value minimizes the same normalized GCR and global MSSD later used as primary success metrics in Table I. With essentially one free threshold, this couples hyperparameter selection to the reported gains and weakens the claim that lower GCR/MSSD reflect better recovery of physical cluster consistency rather than an enlarged acceptance region on this route. Please select D_th by a criterion independent of the final GCR/MSSD scores (e.g., held-out route segment, silhouette/DB on a validation window, or fixed quantile of the Mahalanobis distribution under the null), or report performance for a range of D_th without cherry-picking the optimum of the evaluation metrics.
  2. §III–IV: Validation is limited to a single measurement-calibrated ray-tracing route (132 GHz IIoT, one TX, one RX path, dense 0.01 m init then 0.1 m spacing). The central claim of improved cluster-level spatial consistency for 6G therefore rests on one geometry and noise-free (or low-noise) MPC parameters. At minimum, add (i) a second route/geometry or snapshot spacing, and/or (ii) a controlled noise study on delay/angle parameters, and show that the ranking vs MFM is preserved. Without that, transfer beyond this tuned synthetic route is not established.
  3. §I and §IV: The introduction surveys several joint clustering-and-tracking methods (e.g., Kalman-initialized KPowerMeans [35], [36]; previous-snapshot-referenced clustering [34]), yet the only quantitative baseline is MFM [30] (tracking-after-clustering). For the claim that MD-SCT improves on joint methods that “mainly rely on cluster centers,” please include at least one joint centroid/prediction-based baseline under the same data and metrics, or clearly reframe the contribution as improvement over tracking-after-clustering only.
  4. Table I / §IV: Average DB and CH are reported as comparable or slightly better for MD-SCT, but K1 and per-snapshot cluster counts for MFM are chosen via DB/CH, while MD-SCT’s long tracks accumulate historical covariance and suppress re-labeling. The large drop in cluster count (104→17) and GCR may partly reflect label persistence rather than improved within-snapshot clustering quality. Please report snapshot-wise cluster purity or association accuracy against the known ray-traced path identities (paths 1–5 in Fig. 1b/3a) so that continuity is separated from correctness of grouping.
minor comments (6)
  1. Abstract and §I: “conducte” → “conducted” (§III); “zPOS_i.RX” period vs comma inconsistency in Algorithm 1 input.
  2. Eq. (8): the modified GCR integrates the l1-norm of second derivatives; state polynomial order n used for g_k(r) and whether the same order is used for both methods.
  3. Fig. 2: axis labels “0.50 0.75 0.875 1 1.125” for D_th/D_ref_th are fine, but state explicitly that D_ref_th=400 is the chosen operating point, not an independent reference.
  4. §II-A: covariance regularization ϵI is mentioned as optional; state whether it was used in the 132 GHz experiments and the value of ϵ if so.
  5. Table I: units and scaling of MSSD columns (×10^2, ×10^{-3}, etc.) should be defined once in the caption for readability.
  6. References: ensure consistent capitalization of titles and that arXiv preprints cited as such are clearly marked if not yet peer-reviewed.

Circularity Check

1 steps flagged

D_th is chosen by minimizing the same GCR/MSSD scores later reported as proof of superiority, so Table I gains are partly metric-tuned rather than independent.

specific steps
  1. fitted input called prediction [§III (para. on D_th), Fig. 2, Table I / §IV]
    "Due to the affine-invariant property of the Mahalanobis distance, the only parameter that needs to be specified is the threshold D_th. For the considered 132 GHz IIoT case, D_th is empirically set to 400, which gives the lowest GCR and MSSD in the threshold-sensitivity analysis as shown in Fig. 2. ... The results demonstrate that the proposed algorithm yields smoother cluster evolution. ... MD-SCT ... GCR 0.028 ... lower MSSD values"

    D_th is the sole free parameter of Algorithm 1. It is selected by a sensitivity sweep that explicitly minimizes the identical GCR and global MSSD quantities later tabulated as the main success metrics against MFM. With no held-out geometry, noise realization, or independent validation set, the large reported reductions in those metrics are at least partly the direct consequence of choosing the acceptance threshold that optimizes them, rather than an independent demonstration that the Mahalanobis historical-covariance rule recovers physically smoother clusters.

full rationale

The MD-SCT assignment rule itself (Mahalanobis distance to the historical cluster covariance, Eq. 1) is not tautological and does not reduce to its inputs by definition; it is a legitimate simultaneous clustering-and-tracking procedure that can produce longer tracks. The circularity is confined to evaluation. Section III and Fig. 2 explicitly select the sole free threshold D_th = 400 because that value yields the lowest normalized GCR and global MSSD on the identical 132 GHz IIoT route. Table I and Section IV then present those same lower GCR (0.028 vs 0.561) and MSSD values as the principal evidence that MD-SCT captures better cluster-level spatial consistency than the MFM baseline. K1 is likewise chosen via DB/CH indices that later reappear in the metric suite. Both methods also optimize free parameters to GCR, so the ranking is not an out-of-sample test of physical consistency. This is classic fitted-input-called-prediction on the success metrics; it does not invalidate the algorithm but does make the quantitative superiority claim partially forced by construction on this single synthetic route. No self-definitional loop, load-bearing self-citation uniqueness theorem, or renamed known result is present. Score 5 reflects partial (evaluation-only) circularity rather than a fully circular derivation.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 2 invented entities

The central performance claim rests on standard clustering math, the domain premise that cluster-level spatial consistency is the right modeling target, and several paper-specific choices: empirical D_th, dense init segment, K1 from DB/CH, and validation on calibrated ray tracing rather than pure long-route measurements. No new physical entities are postulated; MD-SCT and MSSD are methodological constructs.

free parameters (4)
  • Mahalanobis assignment threshold D_th = 400 (D_ref_th)
    Only free scale-like parameter after affine invariance; set empirically to 400 for lowest GCR and MSSD on the studied route (Fig. 2, §III). Directly controls birth of new clusters vs continued tracking.
  • Initial cluster count K1 = 10
    Chosen as optimal under DB and CH on the dense initialization subset (§III); seeds all subsequent Mahalanobis associations.
  • Initialization spatial window = 100 snapshots, 3.00–3.99 m
    First 100 snapshots from 3.00–3.99 m at 0.01 m spacing used for conventional clustering and covariance estimation; quality of C_i,k depends on this choice.
  • Covariance regularization epsilon (optional)
    Paper allows C_reg = C + εI for numerical stability when samples are few; ε is an unstated free stabilizer if used.
axioms (6)
  • domain assumption Cluster-level spatial consistency—similar MPCs remaining in the same clusters as transceivers move—is a fundamental physical property that channel models should preserve.
    Stated in abstract and §I; motivates simultaneous clustering and tracking and the use of GCR/MSSD as success criteria.
  • ad hoc to paper Mahalanobis distance on delay, angles, and TX/RX position using the historical cluster mean and covariance correctly measures consistency with an existing multipath cluster.
    Core modeling choice of §II-A and Eq. (1); not derived from a uniqueness theorem, but motivated by correlation and scale invariance.
  • domain assumption Conventional clustering (KPowerMeans/DBSCAN) on a dense local subset yields reliable initial labels and covariance matrices.
    Algorithm 1 step 1; justified by high spatial similarity and prior THz Silhouette evaluation cited for KPowerMeans.
  • domain assumption Measurement-calibrated deterministic ray tracing at 132 GHz produces MPC parameters adequate to validate clustering/tracking algorithms along a continuous RX route.
    §III; dominant paths matched to measurements, then dense synthetic route used for algorithm comparison.
  • ad hoc to paper Lower MSSD of clustering statistics and lower integrated second-derivative GCR of fitted cluster-center trajectories indicate better cluster-level spatial consistency.
    §II-B definitions; MSSD is introduced here as a novel metric; GCR is adapted from prior tracking work.
  • standard math Standard linear algebra identities (sample mean/covariance, Woodbury update) hold for iterative Mahalanobis computation.
    Eqs. (2)–(4) and complexity discussion citing Woodbury [40].
invented entities (2)
  • MD-SCT algorithm no independent evidence
    purpose: Simultaneous multipath clustering and tracking via Mahalanobis association to historical cluster distributions, with outlier re-clustering for birth and inactivity for death.
    Primary methodological object of the paper (Algorithm 1); independent_evidence false because it is a procedure defined and evaluated only within this work’s pipeline.
  • MSSD metric for clustering-statistic sequences no independent evidence
    purpose: Quantify short-term variability of DB, CH, and intra-cluster parameter spreads across successive snapshots as a spatial-consistency score.
    Introduced in contributions and §II-B as novel for this use; classical successive-difference idea applied to clustering indices.

pith-pipeline@v1.1.0-grok45 · 16155 in / 3935 out tokens · 33145 ms · 2026-07-11T15:34:01.521940+00:00 · methodology

0 comments
read the original abstract

Spatial consistency is a fundamental physical property of wireless channels that reflects the smooth evolution of the channel between spatial locations. At the cluster level, it requires similar multipath components (MPCs) remain grouped into the same clusters as the transceivers move, enabling consistent cluster tracking. Cluster-level spatial consistency is essential for realistic cluster-based channel models, especially for potential 6G techniques such as massive MIMO, integrated sensing and communication, and terahertz (THz) communication. However, existing clustering and tracking methods do not fully exploit spatial correlations of MPCs. In tracking-after-clustering, clustering and tracking are decoupled, while joint clustering-and-tracking mainly relies on cluster centers from the previous snapshot. In this work, we propose a Mahalanobis-distance-based simultaneous clustering and tracking (MD-SCT) algorithm to capture the joint distribution of clustered MPCs in the delay, angular and spatial domains. Under Mahalanobis distance, MPCs in successive snapshots are associated with existing clusters, thereby inherently tracking while clustering. The algorithm is further applied in the sub-THz band. Performance is evaluated using mean square successive difference and gradient change rate. The results demonstrate that the proposed algorithm yields smoother cluster evolution. This improves the reliability of clustered channels for spatial consistency modeling in 6G.

Figures

Figures reproduced from arXiv: 2607.04664 by Byonghyo Shim, Huixin Xu, Jianhua Zhang, Jiaxin Lin, Ke Chen, Pan Tang, Peijie Liu, Yufeng Qin, Zhaowei Chang.

Figure 1
Figure 1. Figure 1: Measurement photo and simulation layout at 132 GHz for the industrial Internet of Things (IIoT) scenario. Dominant [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sensitivity analysis of Dth/Dref th in terms of (a) normal￾ized global gradient change rate (GCR) and (b) global mean squared successive difference (MSSD), where Dref th = 400. IV. PERFORMANCE VALIDATION In this section, we perform the clustering and tracking on the channel in Section III. For comparison, we also test the multidimensional feature metric-based (MFM) cluster￾tracking method [30]. In the MFM … view at source ↗
Figure 3
Figure 3. Figure 3: AOA-AOD of MPCs over TX-RX distance at 132 GHz. Five paths are marked in (a) (1: LOS; 2–5: first-order reflections). [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 2 canonical work pages

  1. [1]

    A. F. Molish,Wireless communications. John Wiley & Sons, 2012, vol. 34

    2012

  2. [2]

    Channel measurement, modeling, and simulation for 6G: A survey and tutorial,

    J. Zhang, J. Lin, P. Tanget al., “Channel measurement, modeling, and simulation for 6G: A survey and tutorial,” 2023, arXiv:2305.16616

    Pith/arXiv arXiv 2023

  3. [3]

    Towards 6G hyper-connectivity: Vision, challenges, and key enabling technologies,

    H. Lee, B. Lee, H. Yanget al., “Towards 6G hyper-connectivity: Vision, challenges, and key enabling technologies,”J. Commun. Netw., vol. 25, no. 3, pp. 344–354, May 2023

    2023

  4. [4]

    Channel nonstationarity and consis- tency for beyond 5G and 6G: A survey,

    X. Cheng, Z. Huang, and L. Bai, “Channel nonstationarity and consis- tency for beyond 5G and 6G: A survey,”IEEE Commun. Surveys Tuts., vol. 24, no. 3, pp. 1634–1669, Jun. 2022

    2022

  5. [5]

    The method to implement 5G channel model with spatial consistency,

    G. Yu, L. Tian, J. Zhanget al., “The method to implement 5G channel model with spatial consistency,” inProc. IEEE/CIC Int. Conf. Commun. China (ICCC), Aug. 2018, pp. 736–740

    2018

  6. [6]

    Sub-6 GHz to mmWave for 5G- advanced and beyond: Channel measurements, characteristics and impact on system performance,

    H. Miao, J. Zhang, P. Tanget al., “Sub-6 GHz to mmWave for 5G- advanced and beyond: Channel measurements, characteristics and impact on system performance,”IEEE J. Sel. Areas Commun., vol. 41, no. 6, pp. 1945–1960, 2023

    1945

  7. [7]

    Clustering in 3D MIMO channel: Measurement-based results and improvements,

    P. Tang, J. Zhang, Y . Sunet al., “Clustering in 3D MIMO channel: Measurement-based results and improvements,” inProc. IEEE 92nd Veh. Technol. Conf. (VTC–Fall), Sep. 2015, pp. 1–6

    2015

  8. [8]

    Measurement-based validation of the 3GPP spatial consistency procedures,

    W. Sloane, M. Shafi, C. Gentileet al., “Measurement-based validation of the 3GPP spatial consistency procedures,”IEEE Trans. Veh. Technol., vol. 73, no. 4, pp. 4787–4800, Jan. 2024

    2024

  9. [9]

    Characterization of spatial consistency of cluster channels in urban environments at 24 and 60 GHz,

    N. Suzuki, H. Tsukada, R. Takahashiet al., “Characterization of spatial consistency of cluster channels in urban environments at 24 and 60 GHz,”IEEE Antennas Wirel. Propag. Lett., vol. 23, no. 5, pp. 1583– 1587, Feb. 2024

    2024

  10. [10]

    A framework of Mahalanobis-distance metric with supervised learning for clustering multipath components in MIMO channel analysis,

    Y . Chen, C. Han, J. Heet al., “A framework of Mahalanobis-distance metric with supervised learning for clustering multipath components in MIMO channel analysis,”IEEE Trans. Antennas Propag., vol. 70, no. 6, pp. 4069–4081, Feb. 2022

    2022

  11. [11]

    Clustering analysis in the wireless propagation channel with a variational Gaussian mixture model,

    Y . Li, J. Zhang, Z. Maet al., “Clustering analysis in the wireless propagation channel with a variational Gaussian mixture model,”IEEE Trans. Big Data, vol. 6, no. 2, pp. 223–232, May 2020

    2020

  12. [12]

    Clustering enabled wireless channel modeling using big data algorithms,

    R. He, B. Ai, A. F. Molischet al., “Clustering enabled wireless channel modeling using big data algorithms,”IEEE Commun. Mag., vol. 56, no. 5, pp. 177–183, Feb. 2018

    2018

  13. [13]

    Spatial channel model for MIMO simulations,

    “Spatial channel model for MIMO simulations,” 3GPP, TR 25.996, 2003

    2003

  14. [14]

    IST-4-027756 WINNER II D1. 1.2 v1. 2 WINNER II channel models,

    P. Ky ¨osti, J. Meinil ¨a, L. Hentil ¨aet al., “IST-4-027756 WINNER II D1. 1.2 v1. 2 WINNER II channel models,”Inf. Soc. Technol, vol. 11, 2007

    2007

  15. [15]

    The COST 2100 MIMO channel model,

    L. Liu, C. Oestges, J. Poutanenet al., “The COST 2100 MIMO channel model,”IEEE Wirel. Commun., vol. 19, no. 6, pp. 92–99, dec 2012

    2012

  16. [16]

    Guidelines for evaluation of radio interface technologies for IMT-2020,

    “Guidelines for evaluation of radio interface technologies for IMT-2020,” ITU-R, M. 2412, Oct. 2017

    2020

  17. [17]

    Toward 6G with terahertz com- munications: Understanding the propagation channels,

    X. Cai, X. Cheng, and F. Tufvesson, “Toward 6G with terahertz com- munications: Understanding the propagation channels,”IEEE Commun. Mag., vol. 62, no. 2, pp. 32–38, Feb. 2024

    2024

  18. [18]

    Far-field to near-field: Experimental studies of MIMO channel characterization and modeling in the 6 GHz band,

    H. Miao, J. Zhang, P. Tanget al., “Far-field to near-field: Experimental studies of MIMO channel characterization and modeling in the 6 GHz band,”IEEE J. Sel. Areas Commun., vol. 43, no. 11, pp. 3889–3902, 2025

    2025

  19. [19]

    Achieving wireless cable testing for MIMO devices with a novel condition number reduction method,

    H. Wang, J. Zhang, Z. Huet al., “Achieving wireless cable testing for MIMO devices with a novel condition number reduction method,” IEEE Trans. Antennas Propag., pp. 1–1, 2025, Early Access, DOI: 10.1109/TAP.2025.3618761

    work page doi:10.1109/tap.2025.3618761 2025

  20. [20]

    A unified RCS modeling of typical targets for 3GPP ISAC channel standardization and experimental analysis,

    Y . Zhang, J. Zhang, H. Gonget al., “A unified RCS modeling of typical targets for 3GPP ISAC channel standardization and experimental analysis,”IEEE J. Sel. Areas Commun., pp. 1–1, 2025, Early Access, DOI: 10.1109/JSAC.2025.3608732

    work page doi:10.1109/jsac.2025.3608732 2025

  21. [21]

    Fast and accurate terahertz beam man- agement via frequency-dependent beamforming,

    S. Kim, J. Park, J. Moonet al., “Fast and accurate terahertz beam man- agement via frequency-dependent beamforming,”IEEE Trans. Wireless Commun., vol. 23, no. 3, pp. 1699–1712, early access, Jul. 2023

    2023

  22. [22]

    An empirical study on near-field, spatial non-stationarity, and beam misalignment characteristics of THz XL- MIMO channels at 132 GHz,

    H. Xu, P. Tang, J. Zhanget al., “An empirical study on near-field, spatial non-stationarity, and beam misalignment characteristics of THz XL- MIMO channels at 132 GHz,” inIEEE Int. Conf. Commun. Workshops (ICC Workshops), Jun. 2024, pp. 744–749

    2024

  23. [23]

    Deterministic ray tracing: A promising approach to THz channel modeling in 6G deployment scenarios,

    J. Zhang, J. Lin, P. Tanget al., “Deterministic ray tracing: A promising approach to THz channel modeling in 6G deployment scenarios,”IEEE Commun. Mag., vol. 62, no. 2, pp. 48–54, Feb. 2024

    2024

  24. [24]

    Wireless communications and applications above 100 GHz: Opportunities and challenges for 6G and beyond,

    T. S. Rappaport, Y . Xing, O. Kanhereet al., “Wireless communications and applications above 100 GHz: Opportunities and challenges for 6G and beyond,”IEEE Access, vol. 7, pp. 78 729–78 757, Jun. 2019

    2019

  25. [25]

    Can wireless environment information decrease pilot overhead: A channel prediction example,

    L. Shi, J. Zhang, L. Yuet al., “Can wireless environment information decrease pilot overhead: A channel prediction example,”IEEE Wireless Commun. Lett., vol. 14, no. 3, pp. 861–865, Jan. 2025

    2025

  26. [26]

    Digital twin channel for 6G: Concepts, architectures and potential applications,

    H. Wang, J. Zhang, G. Nieet al., “Digital twin channel for 6G: Concepts, architectures and potential applications,”IEEE Commun. Mag., vol. 63, no. 3, pp. 24–30, Nov. 2024

    2024

  27. [27]

    A framework for automatic clustering of parametric MIMO channel data including path powers,

    N. Czink, P. Cera, J. Saloet al., “A framework for automatic clustering of parametric MIMO channel data including path powers,” inProc. IEEE 64th Veh. Technol. Conf. (VTC-Fall), Sep. 2006, pp. 1–5

    2006

  28. [28]

    A density-based algorithm for discovering clusters in large spatial databases with noise,

    M. Ester, H.-P. Kriegel, J. Sanderet al., “A density-based algorithm for discovering clusters in large spatial databases with noise,” inProc. 2nd Int. Conf. Knowl. Discovery Data Mining (KDD), vol. 96, no. 34, 1996, pp. 226–231

    1996

  29. [29]

    A novel automatic cluster tracking algorithm,

    N. Czink and C. Mecklenbrauker, “A novel automatic cluster tracking algorithm,” inProc. IEEE 17th Annu. Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC). IEEE, Sep. 2006, pp. 1–5

    2006

  30. [30]

    A multidimensional feature metric-based cluster-tracking algorithm and its application to time-varying millimeter- wave channels,

    B. Zhu, F. Du, Q. Liet al., “A multidimensional feature metric-based cluster-tracking algorithm and its application to time-varying millimeter- wave channels,”IEEE Trans. Antennas Propagat., vol. 72, no. 11, pp. 8910–8914, Sep. 2024

    2024

  31. [31]

    Multipath clustering and cluster tracking for geometry-based stochastic channel modeling,

    P. Hanpinitsak, K. Saito, J.-i. Takada, M. Kim, and L. Materum, “Multipath clustering and cluster tracking for geometry-based stochastic channel modeling,”IEEE Trans. Antennas Propag., vol. 65, no. 11, pp. 6015–6028, Sep. 2017

    2017

  32. [32]

    A power-angle-spectrum based clustering and tracking algorithm for time-varying radio channels,

    C. Huang, R. He, Z. Zhong, B. Ai, Y .-A. Geng, Z. Zhong, Q. Li, K. Haneda, and C. Oestges, “A power-angle-spectrum based clustering and tracking algorithm for time-varying radio channels,”IEEE Trans. Veh. Technol., vol. 68, no. 1, pp. 291–305, Oct. 2019

    2019

  33. [33]

    A novel tracking-based multipath component clustering algorithm,

    C. Huang, R. He, Z. Zhong, Y .-A. Geng, Q. Li, and Z. Zhong, “A novel tracking-based multipath component clustering algorithm,”IEEE Antennas Wirel. Propag. Lett., vol. 16, pp. 2679–2683, Aug. 2017

    2017

  34. [34]

    A framework of automatic clustering and tracking for time-variant multipath components,

    Q. Wang, B. Ai, R. Heet al., “A framework of automatic clustering and tracking for time-variant multipath components,”IEEE Commun. Lett., vol. 21, no. 4, pp. 953–956, 2016

    2016

  35. [35]

    Tracking time-variant cluster parameters in MIMO channel measurements,

    N. Czink, R. Tian, S. Wyneet al., “Tracking time-variant cluster parameters in MIMO channel measurements,” inProc. 2nd Int. Conf. Commun. Netw. China (CHINACOM). IEEE, Aug. 2007, pp. 1147– 1151

    2007

  36. [36]

    Tracking based multipath clustering in vehicle- to-infrastructure channels,

    J. Gedschold, C. Schneider, M. K ¨aske, R. S. Thom ¨a, G. Del Galdo, M. Boban, and J. Luo, “Tracking based multipath clustering in vehicle- to-infrastructure channels,” inProc. IEEE 29th Annu. Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC), Sep. 2018, pp. 1–5

    2018

  37. [37]

    A cluster separation measure,

    D. L. Davies and D. W. Bouldin, “A cluster separation measure,”IEEE Trans. Pattern Anal. Mach. Intell., vol. PAMI-1, no. 2, pp. 224–227, Apr. 1979

    1979

  38. [38]

    A dendrite method for cluster analysis,

    T. Cali ´nski and J. Harabasz, “A dendrite method for cluster analysis,” Commun. Statist., Theory Methods, vol. 3, no. 1, pp. 1–27, 1974

    1974

  39. [39]

    Effect of distance measures on K- nearest neighbour classifier,

    V . Kalra, I. Kashyap, and H. Kaur, “Effect of distance measures on K- nearest neighbour classifier,” inProc. 2nd Int. Conf. Comput. Sci., Eng. Appl. (ICCSEA), Sep. 2022, pp. 1–7

    2022

  40. [40]

    M. A. Woodbury,Inverting Modified Matrices, ser. Statistical Research Group Memorandum Reports. Princeton, NJ, USA: Princeton Univer- sity Press, 1950, vol. 42

    1950

  41. [41]

    J. H. Zar,Biostatistical analysis, 5th ed. Prentice Hall, 2009

    2009

  42. [42]

    Dual-polarized channel measurements and modeling at 132 GHz in an indoor factory,

    P. Liu, P. Tang, L. Tianet al., “Dual-polarized channel measurements and modeling at 132 GHz in an indoor factory,” inProc. IEEE Globecom Workshops (GC Wkshps), Dec. 2023, pp. 1463–1468

    2023

  43. [43]

    Comparison of clustering techniques using an indoor measurement at 300 GHz,

    A. Ghosh, R. Takahashi, and M. Kim, “Comparison of clustering techniques using an indoor measurement at 300 GHz,”IEEE Trans. THz Sci. Technol., vol. 13, no. 6, pp. 678–687, Sep. 2023

    2023