Pith. sign in

REVIEW 4 major objections 5 minor 49 references

MudiNet: Task-guided Disentangled Representation Learning for 5G Indoor Multipath-assisted Positioning

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Suppressing diffuse multipath cuts 5G indoor positioning error to 0.15 m.

desk verdict New architecture, but the experiments never simulate a single diffuse reflector — the paper's central claim is untested. read the letter →

arxiv 2506.04024 v1 pith:7J34NIA5 submitted 2025-06-04 eess.SP

Ye Tian , Xueting Xu , Ao Peng This is my paper
classification eess.SP
keywords multipath-assistedpositioningdiffusereflectiondisentangledrepresentationlearningvariationalinferencechannelimpulseresponseindoorlocalization5Gsemi-supervised
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper sets out to establish that, in a 5G indoor multipath channel, the diffuse-reflection part of the channel impulse response is statistically independent of the specular-reflection part, and that this separability can be learned and used to suppress the diffuse part during positioning. The authors build MudiNet, a semi-supervised network that takes multiple channel impulse responses over time, uses self-attention to estimate location-independent environmental features, uses an MLP branch for user-position features, and forces the two kinds of latent variables apart with a variational evidence lower bound. The reported consequence is that single-antenna SISO positioning reaches 0.15 m mean error at SNR = 20 dB and 3.56 m at SNR = -10 dB, beating the strongest baseline by 0.06 m and 0.49 m respectively, with the largest gains occurring where multipath aliasing is worst. A reader should care because the approach removes the need for virtual-anchor association, ray tracing, and explicit geometric parameter estimation.

What carries the argument

The carrying mechanism is the diffuse-specular reflection disentanglable (DSRD) channel model, which splits the CIR into $\mathrm{CIR}_s(t,\tau)=\sum_i a_i(t)\delta(t-\tau_i)$ and $\mathrm{CIR}_d(t,\tau)=\int_D b_j(t) S(t-\tau_j)\,d\tau$, where scatterers in the region $D$ follow a Gaussian density $f_d(x,y)$. The load-bearing identity is the coverage statement of Proposition 2, Equation (13): $\{s_{1,2}(1:T,\tau)\}\approx D$, meaning the arcs cut out of the diffuse region by confocal delay ellipses, accumulated over a trajectory, cover the whole region and therefore wash out the dependence of diffuse delay-power observations on time and user position. That identity converts diffuse multipath from a position-dependent nuisance into a stationary nuisance, which justifies the variational latent-variable model with self-attention over the time axis for global environmental features and an MLP for position features.

What would settle it

Compute, for each simulated trajectory, the union of the arcs $s_{1,2}(t,\tau)$ over all $t$ and $\tau$ and compare its area with the area of $D$; if a substantial part of $D$ is never swept, Proposition 2 fails and the reported gains should not be reproducible. A direct experimental test would compare two trajectory sets, one that covers $D$ and one that avoids large parts of it, and check whether the learned $z_d$ and $z_u$ remain independent (for example, by measuring mutual information) and whether the positioning error gap persists.

Watch

Extended reading notes

Core claim

The central claim is that the aliasing multipath caused by diffuse reflectors and the multipath caused by specular reflectors are independent and separable, so a latent variable $z_d$ describing diffuse scattering can be removed without losing information needed to locate the user. The paper formalizes this as two propositions: given the user position, the specular and diffuse CIRs are conditionally independent (Proposition 1), and over multiple CIR observations the diffuse region $D$ is swept by the delay-ellipse arcs, making $z_d$ independent of both time and position (Proposition 2). MudiNet operationalizes this by maximizing an ELBO in which KL divergences disentangle $z_d$, $z_s$, and $z_u$, with prior uncertainties ordered $\epsilon_u^2 > \epsilon_d^2 > \epsilon_s^2$, and a supervised position loss on $\hat{p}=f_{\mathrm{pos}}(z_u,z_s)$ ensures the remaining latents carry positioning information. The paper reports that suppressing $z_d$ yields the accuracy figures above, with the advantage over baselines growing as SNR drops.

Load-bearing premise

The load-bearing premise is that, over a user's 110-second random walk, the arcs cut by delay ellipses cover essentially the whole diffuse reflector region $D$, making the diffuse observation independent of time and user position; the paper does not derive or check this coverage for the simulated trajectories.

Editorial extensions

If this is right

  • Diffuse multipath can be treated as a learnable nuisance variable and discarded, so single-antenna positioning no longer requires ray tracing or explicit association of each reflected path to a virtual anchor.
  • The method's margin over the best baseline grows from 0.06 m at 20 dB to 0.49 m at -10 dB, meaning the value of disentanglement is highest precisely in the low-SNR, heavy-aliasing conditions found deep indoors.
  • Multi-time CIR observations are not a convenience but a requirement: disentanglement relies on the trajectory sweeping out the diffuse region, so single-snapshot inputs cannot support the claimed separation.
  • Ordering the latent-variable prior uncertainties as $\epsilon_u^2 > \epsilon_d^2 > \epsilon_s^2$ encodes the physical fact that specular reflection is nearly deterministic, giving a general recipe for task-guided disentanglement.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Not explored in the paper: the same nuisance-variable treatment of diffuse multipath could carry over to other radio sensing tasks, including channel charting, mapping, and SLAM, where diffuse reflectors also distort observations.
  • Not explored in the paper: real environments contain multiple diffuse regions with different roughnesses and non-Gaussian scatterer densities, so the coverage condition of Eq. (13) would need to hold separately for each region for the independence claim to survive.
  • Because the low-SNR regime is simulated by lowering transmit power rather than adding receiver noise, the 3.56 m figure at -10 dB is tied to a particular aliasing model; measured 5G CIRs would show whether the gain transfers to real receivers.
  • A design consequence the authors do not draw: instead of random walks, trajectories could be planned to deliberately criss-cross the diffuse region, guaranteeing the coverage on which the disentanglement rests.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes MudiNet, a semi-supervised variational-inference framework that learns a disentangled latent representation of 5G channel impulse responses (CIRs) for indoor multipath-assisted positioning. The authors model diffuse reflectors as random scatterers and specular reflectors as deterministic ideal surfaces, and they argue (Propositions 1 and 2) that the corresponding CIR components are conditionally independent and separable. The network uses self-attention over multi-time CIRs, three latent variables (diffuse environment, specular environment, and user position), and a task-guided loss that combines an ELBO with a position regression loss. Simulation results on a single simulated indoor scenario report mean localization errors from 0.15 m at SNR=20 dB to 3.56 m at SNR=-10 dB, outperforming MLP, CNN, LSTM, Deepfi, and Transformer baselines, with the largest gains at low SNR.

Significance. If the central claim were established, the paper would contribute a useful deep-learning architecture for indoor positioning that explicitly separates diffuse-reflector multipath from specular multipath and suppresses the former. The idea of using multi-time CIR observations to estimate environmental structure and position-related features is plausible, and the reported accuracy gains over strong baselines are nontrivial. However, the significance is currently undermined by a fundamental mismatch between the claimed phenomenon (diffuse reflectors) and the experiments (which never include diffuse scatterers), and by theoretical steps that are either circular or rest on unverified assumptions. The paper is clearly written and the tables and figures are generally readable, but the empirical evidence does not yet support the advertised robustness claim.

major comments (4)
  1. [Section IV-A, Fig. 5 and Table II] The simulation environment contains only ideal smooth reflective surfaces; no diffuse reflector is placed in the scenario. The only mechanism intended to mimic diffuse multipath is the transmission-power-control scheme described around Eq. (31), which adjusts the relative power of specular MPCs under low SNR. This proxy is not validated against the statistical model of Sec. II (Eq. (7), random scatterer positions, ellipsoidal delay arcs). Therefore, the gains reported in Table II at SNR=-10 dB (3.56 m vs. 4.05 m for MLP) could stem from generic robustness to noise, from the multi-time input format, or from the self-attention mechanism, rather than from any disentanglement and suppression of diffuse and specular latent variables. This is a load-bearing gap because the paper's central contribution is specifically robustness against diffuse reflectors.
  2. [Section II-B, Proposition 2, Eq. (13) and Eq. (14)] The proof of Proposition 2 hinges on Eq. (13), which asserts that the arcs swept by the delay ellipses over T observations approximately cover the whole diffuse region D. No coverage condition is derived or checked for the randomly generated 110-second trajectories used in the simulation. Even if full coverage were granted, Eq. (14) does not imply that the diffuse delay-power observation is independent of time and UE position, because the integrand in Eq. (11) contains r_t and r_r, the distances from the scatterer to the transmitter and receiver, which depend on the UE position. Thus the conclusion p(z_d, z_u) = p(z_d) p(z_u) is not established by the given argument.
  3. [Section II-B, Proposition 1, Eq. (12)] Proposition 1 essentially builds the claimed conditional independence into the signal model rather than proving it. In Eq. (2), CIR_s is a deterministic set of specular MPCs given the UE and reflector geometry, while CIR_d is generated by a random scatterer distribution; the two components are then asserted to be conditionally independent in Eq. (12). No argument is given that the common dependence on the UE position, the transmit power, and the propagation environment does not induce statistical dependence between the two components. Furthermore, the word 'separable' in the conclusion does not follow from conditional independence; separability of the latent variables in the learned representation is a stronger, algorithm-dependent property that is not demonstrated by the proof.
  4. [Section III-D, Eq. (26) and Eq. (29)] The disentanglement claim is partly circular. The latent variables are forced to be independent by the choice of Gaussian priors with diagonal covariance (Eq. (26)) and by the KL divergence terms in the ELBO (Eq. (23)), and the position labels guide z_u and z_s through L_pos (Eq. (28)). The reported test-set accuracy is an independent measurement, but it cannot by itself show that the separation is specifically diffuse-versus-specular; no ablation is provided that removes the disentanglement constraints, fixes z_d, or otherwise isolates the effect attributed to diffuse suppression. Without such an ablation, the performance gain could be a generic benefit of variational regularization or of the multi-time input representation.
minor comments (5)
  1. [Abstract and Section I] There are several grammatical and typographical issues, e.g., 'multipath component (MPC) are no longer regarded' and 'an localization' in the Introduction; a careful proofreading pass is needed.
  2. [Section IV-B] The description of the baseline splitting of multi-time CIRs into single-time samples is somewhat confusing; it would help to state explicitly that the baseline methods receive one CIR snapshot per sample while MudiNet receives the full multi-time sequence, and to clarify how the training/test split is ensured to be fair.
  3. [Section II-B] In the proof of Proposition 1, the phrase 'As shown in Figure' leaves the reference incomplete; the figure is presumably Fig. 2, but this should be stated.
  4. [Section IV-A] The sentence 'We use the FP power at the start time P_FP(t=0) of each trajectory as the signal power' is clear, but Eq. (31) should explicitly define all symbols (e.g., G_ss) in the equation caption or in the text immediately following it.
  5. [Section III-A] The notation for the latent variable model is introduced informally; in particular, the distinction between z_s and z_d as 'environment-related' variables and z_u as 'position-related' is plausible but should be tied more precisely to the conditional independence statements of Propositions 1 and 2.

Circularity Check

1 steps flagged · score 5.0 of 10

Proposition 1's separability theorem restates the model's definitions; empirical accuracy is independently measured but does not actually test diffuse reflectors.

  1. self definitional [Section II-B, Proposition 1 and Eq. (12)]
    "This study differentiates between diffuse reflectors and specular reflectors in the spatial model, treating the specular reflection components as deterministic multipath, while considering the diffuse reflection components as uncertain, indistinguishable multipath... Under the assumption of WSS, given a UE position at time step t0, the CIR d(t, τ) from diffuse reflection is conditionally independent of the CIR s(t, τ) from specular reflection: p((τ d, βd), (τs, βs))|p(t0)) = p((τ d, βd)|p(t0))p((τs, βs)|p(t0)) (12)."

    The proof assigns specular reflections a deterministic mirror-geometry model (Eqs. 5-6) and diffuse reflections an independent random-scatterer model f_d(x,y) with no coupling to specular reflectors; Eqs. (8)-(11) integrate only over D. Equation (12), the claimed conditional independence, is then the factorization of two separately constructed random processes. It is not derived from measured data or from independent physical laws; it is a restatement of the model's construction. The first contribution ('demonstrating that ... independent and separable') is therefore self-definitional: the conclusion is inserted by the way the two components are defined.

full rationale

The paper's central empirical claim (MudiNet accuracy at various SNRs) is supported by a train/test split on a fixed simulation, so the accuracy numbers are independent measurements and not circular. The main circularity is in the theoretical foundation: Proposition 1 'proves' that diffuse and specular MPC contributions are conditionally independent by first defining specular reflection as a deterministic mirror-geometry function and diffuse reflection as an independent Gaussian-random-scatterer process, then reading off the factorization in Eq. (12). That is a self-definitional restatement, not a demonstration from independent physics, and it is load-bearing because the paper's first contribution and the design of the three-factor latent space (z_d, z_s, z_u) rest on it. Proposition 2 further assumes unverified coverage of the diffuse region (Eq. 13) to conclude p(z_d,z_u)=p(z_d)p(z_u), a conclusion that the ELBO and the factorized encoder (Eqs. 20, 23) also build in by construction; this is an unsupported assumption rather than a separate circular step. The simulation (Section IV-A) explicitly contains only ideal smooth reflectors (Fig. 5), so the claimed robustness specifically against diffuse reflectors is not directly tested; the low-SNR power-control proxy changes MPC power ratios, which is a plausible stand-in but not a validation of the physical disentanglement. The self-citation [20] is used only for background claims and is not load-bearing. Overall, the theoretical separability claim reduces by construction, while the localization accuracy comparison remains an independent measurement, yielding a partial circularity score of 5.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central derivation rests on the signal model's conditional independence and the coverage assumption in Eq. (13), both of which are asserted rather than established. The Gaussian scatterer model and the power-control SNR proxy add further domain assumptions that are not independently validated.

free parameters (4)
  • Latent prior variances epsilon_u^2, epsilon_d^2, epsilon_s^2 = Not reported; ordered as epsilon_u^2 > epsilon_d^2 > epsilon_s^2
    Section III-D: chosen by hand to encourage disentanglement; exact values not given. The ordering is a modeling choice that influences how much position information is pushed into z_u versus z_s and z_d.
  • Latent dimension l = Not reported
    Used in Eq. (26) for the latent space; no value or search procedure is provided.
  • Self-attention dimension d = Not reported
    Used in Eqs. (16)-(17) for attention projections; no value is given.
  • Number of layers and neurons = Not reported
    The paper says MLP and Transformer baselines use the same layer and neuron counts as the proposed method, but the actual counts are not stated.
assumptions (6)
  • domain assumption The indoor channel is wide-sense stationary during observations.
    Section II-A invokes WSS to justify treating specular randomness as noise; this underpins the conditional independence statement.
  • domain assumption Diffuse scatterers follow a Gaussian spatial distribution fd around a central point (Eq. 7).
    Used to derive diffuse power integrals in Eqs. (8)-(11); no measurement justifies the Gaussian form.
  • ad hoc to paper Given UE position, specular and diffuse CIR components are conditionally independent (Eq. 12).
    This factorization is the conclusion of Proposition 1 but is effectively built into the signal model of Eqs. (1)-(2), making the proof circular.
  • ad hoc to paper Over T observations, the family of delay ellipses covers the diffuse region D (Eq. 13).
    Proposition 2 depends on this coverage assumption; no bound on T or trajectory is given.
  • domain assumption Latent variables zs, zd, zu are Gaussian with ordered variances, and the reparameterization trick applies (Eqs. 26-27).
    Standard VAE assumptions; the variance ordering is a hand-selected inductive bias for disentanglement.
  • ad hoc to paper Transmission-power SNR control reproduces the statistical effect of diffuse multipath aliasing.
    Section IV-A uses this to generate all test data; no comparison with a ray-traced diffuse reflector model is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MudiNet: Task-guided Disentangled Representation Learning for 5G Indoor Multipath-assisted Positioning." pith.science (2026). https://pith.science/paper/7J34NIA5

@misc{pith2026250604024,
  author       = {Pith},
  title        = {Pith review of: MudiNet: Task-guided Disentangled Representation Learning for 5G Indoor Multipath-assisted Positioning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7J34NIA5}},
  note         = {Machine review of arXiv:2506.04024}
}
read the original abstract

In the fifth-generation communication system (5G), multipath-assisted positioning (MAP) has emerged as a promising approach. With the enhancement of signal resolution, multipath component (MPC) are no longer regarded as noise but rather as valuable information that can contribute to positioning. However, existing research often treats reflective surfaces as ideal reflectors, while being powerless in the face of indistinguishable multipath caused by diffuse reflectors. This study approaches diffuse reflectors from the perspective of uncertainty, investigating the statistical distribution characteristics of indoor diffuse and specular reflectors. Based on these insights, a task-guided disentangled representation learning method leveraging multi-time channel impulse response (CIR) observations is designed to directly map CIRs to positions, while mitigating the adverse effects of components that contribute minimally to localization accuracy (e.g., diffuse multipath).In this semi-supervised learning framework, a global feature extraction architecture based on self-attention is proposed to capture location-independent wireless environmental information, while an MLP is employed to extract the time-varying features related to user equipment (UE) positions. Variational inference based on a latent variable model (LVM) is applied to separate independent features within the CIR, with position labels guiding the LVM to express components more beneficial for localization. Additionally, we provide a feasibility proof for the separability of diffuse and specular environmental features in CIRs. Simulation results demonstrate that the proposed method achieves higher localization accuracy compared to conventional search-based localization methods, with enhanced robustness against indistinguishable multipath from diffuse reflectors.

Figures

Figures reproduced from arXiv: 2506.04024 by the authors.

Figure 1
Figure 1. The left side represents the diffuse MPCs, where the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The illustration of diffuse reflector distribution. The [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Structure of Proposed Multipath Disentangled Net [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Structure of Self-attention Layer environmental variables. Specifically, as shown in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Simulation Scenarios Illustration. Representing Sce [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Two different methods for controlling SNR under bandwidth of 100Mhz: by adding additive Gaussian white noise [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Error Cumulative Distribution Function for Different [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Mean Position Error under Different SNR Conditions [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Improvements brought by the proposed method under different levels of multipath aliasing are compared, and the [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 43 canonical work pages

  1. [1]

    Accurate indoor positioning using temporal–spatial constraints based on Wi-Fi fine time measurements,

    W. Shao, H. Luo, F. Zhao, H. Tian, S. Yan, and A. Crivello, “Accurate indoor positioning using temporal–spatial constraints based on Wi-Fi fine time measurements,” IEEE Internet of Things Journal , vol. 7, no. 11, pp. 11 006–11 019, 2020

    work page 2020

  2. [2]

    A novel 3-D indoor localization algorithm based on BLE and multiple sensors,

    Y . Yu, R. Chen, L. Chen, X. Zheng, D. Wu, W. Li, and Y . Wu, “A novel 3-D indoor localization algorithm based on BLE and multiple sensors,” IEEE Internet of Things Journal , vol. 8, no. 11, pp. 9359–9372, 2021

    work page 2021

  3. [3]

    Recent advances in floor positioning based on smartphone,

    Q. Wang, M. Fu, J. Wang, H. Luo, L. Sun, Z. Ma, W. Li, C. Zhang, R. Huang, X. Li et al. , “Recent advances in floor positioning based on smartphone,” Measurement, vol. 214, p. 112813, 2023

    work page 2023

  4. [4]

    A new TOA localization and synchronization system with virtually synchronized periodic asymmetric ranging network,

    S. Zhao, X.-P. Zhang, X. Cui, and M. Lu, “A new TOA localization and synchronization system with virtually synchronized periodic asymmetric ranging network,” IEEE Internet of Things Journal , vol. 8, no. 11, pp. 9030–9044, 2021

    work page 2021

  5. [5]

    Integrated sensing and communications: Toward dual-functional wire- less networks for 6G and beyond,

    F. Liu, Y . Cui, C. Masouros, J. Xu, T. X. Han, Y . C. Eldar, and S. Buzzi, “Integrated sensing and communications: Toward dual-functional wire- less networks for 6G and beyond,” IEEE journal on selected areas in communications, vol. 40, no. 6, pp. 1728–1767, 2022

    work page 2022

  6. [6]

    Survey of cellular mobile radio localization methods: From 1G to 5G,

    J. A. del Peral-Rosado, R. Raulefs, J. A. L ´opez-Salcedo, and G. Seco- Granados, “Survey of cellular mobile radio localization methods: From 1G to 5G,” IEEE Communications Surveys & Tutorials , vol. 20, no. 2, pp. 1124–1148, 2017

    work page 2017

  7. [7]

    Positioning under multipath environments in wireless network: Survey, design and opportunities,

    L. Xie, Y . Zhang, C. Zhao, C. Zhang, Z. Mu, and X. Yang, “Positioning under multipath environments in wireless network: Survey, design and opportunities,” IEEE Network , 2024

    work page 2024

  8. [8]

    Multipath assisted positioning with simultaneous localization and mapping,

    C. Gentner, T. Jost, W. Wang, S. Zhang, A. Dammann, and U.-C. Fiebig, “Multipath assisted positioning with simultaneous localization and mapping,” IEEE Transactions on Wireless Communications , vol. 15, no. 9, pp. 6104–6117, 2016

    work page 2016

Show all 49 references
  1. [9]

    Single-anchor ultra-wideband localization sys- tem using wrapped PDoA,

    F. Ge and Y . Shen, “Single-anchor ultra-wideband localization sys- tem using wrapped PDoA,” IEEE Transactions on Mobile Computing , vol. 21, no. 12, pp. 4609–4623, 2021

    2021

  2. [10]

    Mmwave 6D radio localization with a snapshot observation from a single BS,

    M. A. Nazari, G. Seco-Granados, P. Johannisson, and H. Wymeersch, “Mmwave 6D radio localization with a snapshot observation from a single BS,” IEEE Trans. V eh. Technol., vol. 72, no. 7, pp. 8914–8928, Jul. 2023

    2023

  3. [11]

    Multipath tracking with LTE signals for accurate TOA estimation in the application of indoor positioning,

    Z. Liu, L. Chen, X. Zhou, N. Shen, and R. Chen, “Multipath tracking with LTE signals for accurate TOA estimation in the application of indoor positioning,” Geo-spatial Information Science , vol. 26, no. 1, pp. 31–43, 2023

    2023

  4. [12]

    Machine learn- ing for time-of-arrival estimation with 5G signals in indoor positioning,

    Z. Liu, L. Chen, X. Zhou, Z. Jiao, G. Guo, and R. Chen, “Machine learn- ing for time-of-arrival estimation with 5G signals in indoor positioning,” IEEE Internet of Things Journal , vol. 10, no. 11, pp. 9782–9795, 2023

    2023

  5. [13]

    AoA- based positioning for aerial intelligent reflecting surface-aided wireless communications: An angle-domain approach,

    T. Zhou, K. Xu, Z. Shen, W. Xie, D. Zhang, and J. Xu, “AoA- based positioning for aerial intelligent reflecting surface-aided wireless communications: An angle-domain approach,” IEEE Wireless Commu- nications Letters , vol. 11, no. 4, pp. 761–765, 2022

    2022

  6. [14]

    Multipath triangulation: Decimeter-level wifi localization and orientation with a single unaided receiver,

    E. Soltanaghaei, A. Kalyanaraman, and K. Whitehouse, “Multipath triangulation: Decimeter-level wifi localization and orientation with a single unaided receiver,” in Proceedings of the 16th annual international conference on mobile systems, applications, and services , 2018, pp...

    2018

  7. [15]

    Massive MIMO-based localization and mapping exploiting phase in- formation of multipath components,

    X. Li, E. Leitinger, M. Oskarsson, K. ˚Astr¨om, and F. Tufvesson, “Massive MIMO-based localization and mapping exploiting phase in- formation of multipath components,” IEEE transactions on wireless communications, vol. 18, no. 9, pp. 4254–4267, 2019

    2019

  8. [16]

    Exploiting multipath information for integrated localization and sensing via PHD filtering,

    Y . Du, H. Zhao, Y . Liu, X. Yu, and Y . Shen, “Exploiting multipath information for integrated localization and sensing via PHD filtering,” IEEE Transactions on V ehicular Technology, 2024

    2024

  9. [17]

    UWB positioning system based on LSTM classification with mitigated NLOS effects,

    D.-H. Kim, A. Farhad, and J.-Y . Pyun, “UWB positioning system based on LSTM classification with mitigated NLOS effects,” IEEE Internet of Things Journal , vol. 10, no. 2, pp. 1822–1835, 2022

    2022

  10. [18]

    Multipath-assisted single- anchor localization via deep variational learning,

    T. Wang, Y . Li, J. Liu, K. Hu, and Y . Shen, “Multipath-assisted single- anchor localization via deep variational learning,” IEEE Transactions on Wireless Communications, 2024

    2024

  11. [19]

    A variational learning approach for concurrent distance estimation and environmental identification,

    Y . Li, S. Mazuelas, and Y . Shen, “A variational learning approach for concurrent distance estimation and environmental identification,” IEEE Transactions on Wireless Communications , vol. 22, no. 9, pp. 6252– 6266, 2023

    2023

  12. [20]

    Multi-state con- straint multipath-assisted positioning and mismatch alleviation,

    X. Xu, A. Peng, X. Hong, Y . Zhang, and X.-P. Zhang, “Multi-state con- straint multipath-assisted positioning and mismatch alleviation,” IEEE Internet of Things Journal , 2023

    2023

  13. [21]

    Ray-tracing-assisted fingerprinting based on channel impulse response measurement for indoor positioning,

    P.-H. Tseng, Y .-C. Chan, Y .-J. Lin, D.-B. Lin, N. Wu, and T.-M. Wang, “Ray-tracing-assisted fingerprinting based on channel impulse response measurement for indoor positioning,” IEEE Transactions on Instrumentation and Measurement , vol. 66, no. 5, pp. 1032–1045, 2017

    2017

  14. [22]

    Toward 5G NR high-precision indoor positioning via channel frequency response: A new paradigm and dataset generation method,

    K. Gao, H. Wang, H. Lv, and W. Liu, “Toward 5G NR high-precision indoor positioning via channel frequency response: A new paradigm and dataset generation method,” IEEE Journal on Selected Areas in Communications, vol. 40, no. 7, pp. 2233–2247, 2022

    2022

  15. [23]

    CSI-based fingerprinting for indoor localization: A deep learning approach,

    X. Wang, L. Gao, S. Mao, and S. Pandey, “CSI-based fingerprinting for indoor localization: A deep learning approach,” IEEE transactions on vehicular technology, vol. 66, no. 1, pp. 763–776, 2016

    2016

  16. [24]

    iPos- 5G: Indoor positioning via commercial 5G NR CSI,

    Y . Ruan, L. Chen, X. Zhou, Z. Liu, X. Liu, G. Guo, and R. Chen, “iPos- 5G: Indoor positioning via commercial 5G NR CSI,” IEEE Internet of Things Journal , vol. 10, no. 10, pp. 8718–8733, 2022. 12

    2022

  17. [25]

    Hi-loc: Hybrid indoor localization via enhanced 5G NR CSI,

    Y . Ruan, L. Chen, X. Zhou, G. Guo, and R. Chen, “Hi-loc: Hybrid indoor localization via enhanced 5G NR CSI,” IEEE Transactions on Instrumentation and Measurement , vol. 71, pp. 1–15, 2022

    2022

  18. [26]

    UWB channel impulse responses for positioning in complex environments: A detailed feature analysis,

    S. Kram, M. Stahlke, T. Feigl, J. Seitz, and J. Thielecke, “UWB channel impulse responses for positioning in complex environments: A detailed feature analysis,” Sensors, vol. 19, no. 24, p. 5547, 2019

    2019

  19. [27]

    UWB system for indoor positioning and tracking with arbitrary target orientation, optimal anchor location, and adaptive NLOS mitigation,

    Y .-Y . Chen, S.-P. Huang, T.-W. Wu, W.-T. Tsai, C.-Y . Liou, and S.-G. Mao, “UWB system for indoor positioning and tracking with arbitrary target orientation, optimal anchor location, and adaptive NLOS mitigation,” IEEE Transactions on V ehicular Technology, vol. 69, no. 9, p...

    2020

  20. [28]

    High- accuracy localization for assisted living: 5G systems will turn multipath channels from foe to friend,

    K. Witrisal, P. Meissner, E. Leitinger, Y . Shen, C. Gustafson, F. Tufves- son, K. Haneda, D. Dardari, A. F. Molisch, A. Conti et al. , “High- accuracy localization for assisted living: 5G systems will turn multipath channels from foe to friend,” IEEE Signal Processing Magazin...

    2016

  21. [29]

    Evaluation of position-related information in multipath components for indoor positioning,

    E. Leitinger, P. Meissner, C. R ¨udisser, G. Dumphart, and K. Witrisal, “Evaluation of position-related information in multipath components for indoor positioning,” IEEE Journal on Selected Areas in communications , vol. 33, no. 11, pp. 2313–2328, 2015

    2015

  22. [30]

    Multipath-based SLAM for non-ideal reflective surfaces exploiting multiple-measurement data association,

    L. Wielandner, A. Venus, T. Wilding, and E. Leitinger, “Multipath-based SLAM for non-ideal reflective surfaces exploiting multiple-measurement data association,” arXiv preprint arXiv:2304.05680 , 2023

    2023 arXiv

  23. [31]

    Accuracy bounds for array-based positioning in dense multipath channels,

    T. Wilding, S. Grebien, U. M ¨uhlmann, and K. Witrisal, “Accuracy bounds for array-based positioning in dense multipath channels,” Sen- sors, vol. 18, no. 12, p. 4249, 2018

    2018

  24. [32]

    Angle-based positioning estimation lever- aging diffuse scattering paths in millimeter-wave MIMO systems,

    L. Guo, T. Lv, and J. Zeng, “Angle-based positioning estimation lever- aging diffuse scattering paths in millimeter-wave MIMO systems,” IEEE Sensors Journal, 2024

    2024

  25. [33]

    5G positioning and mapping with diffuse multipath,

    F. Wen, J. Kulmer, K. Witrisal, and H. Wymeersch, “5G positioning and mapping with diffuse multipath,” IEEE Transactions on Wireless Communications, vol. 20, no. 2, pp. 1164–1174, 2020

    2020

  26. [34]

    5G SLAM using the clustering and assignment approach with diffuse multipath,

    Y . Ge, F. Wen, H. Kim, M. Zhu, F. Jiang, S. Kim, L. Svensson, and H. Wymeersch, “5G SLAM using the clustering and assignment approach with diffuse multipath,” Sensors, vol. 20, no. 16, p. 4656, 2020

    2020

  27. [35]

    A 3-D non- stationary wideband geometry-based channel model for MIMO vehicle- to-vehicle communications in tunnel environments,

    H. Jiang, Z. Zhang, L. Wu, J. Dang, and G. Gui, “A 3-D non- stationary wideband geometry-based channel model for MIMO vehicle- to-vehicle communications in tunnel environments,” IEEE Transactions on V ehicular Technology, vol. 68, no. 7, pp. 6257–6271, 2019

    2019

  28. [36]

    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 Communications Surveys & Tutorials, vol. 24, no. 3, pp. 1634–1669, 2022

    2022

  29. [37]

    A 3-D nonstationary wideband V2V GBSM with UPAs for massive MIMO wireless communication systems,

    L. Bai, Z. Huang, H. Du, and X. Cheng, “A 3-D nonstationary wideband V2V GBSM with UPAs for massive MIMO wireless communication systems,” IEEE Internet of Things Journal , vol. 8, no. 24, pp. 17 622– 17 638, 2021

    2021

  30. [38]

    Learning latent represen- tations for style control and transfer in end-to-end speech synthesis,

    Y .-J. Zhang, S. Pan, L. He, and Z.-H. Ling, “Learning latent represen- tations for style control and transfer in end-to-end speech synthesis,” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2019, pp. 6945–6949

    2019

  31. [39]

    Self-supervised vq-vae for one-shot music style transfer,

    O. C ´ıfka, A. Ozerov, U. S ¸ims ¸ekli, and G. Richard, “Self-supervised vq-vae for one-shot music style transfer,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2021, pp. 96–100

    2021

  32. [40]

    Multiple style transfer via variational autoencoder,

    Z.-S. Liu, V . Kalogeiton, and M.-P. Cani, “Multiple style transfer via variational autoencoder,” in 2021 IEEE International Conference on Image Processing (ICIP) . IEEE, 2021, pp. 2413–2417

    2021

  33. [41]

    A belief propagation algorithm for multipath-based SLAM,

    E. Leitinger, F. Meyer, F. Hlawatsch, K. Witrisal, F. Tufvesson, and M. Z. Win, “A belief propagation algorithm for multipath-based SLAM,” IEEE transactions on wireless communications , vol. 18, no. 12, pp. 5613– 5629, 2019

    2019

  34. [42]

    Indoor environment learning via RF-mapping,

    R. Amiri, S. Yerramalli, T. Yoo, M. Hirzallah, M. Zorgui, R. Prakash, and X. Zhang, “Indoor environment learning via RF-mapping,” IEEE Journal on Selected Areas in Communications , vol. 41, no. 6, pp. 1859– 1872, 2023

    2023

  35. [43]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017

    2017

  36. [44]

    Auto-encoding variational bayes,

    D. P. Kingma, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  37. [45]

    An overview of 5G advanced evolution in 3GPP release 18,

    X. Lin, “An overview of 5G advanced evolution in 3GPP release 18,” IEEE Communications Standards Magazine , vol. 6, no. 3, pp. 77–83, 2022

    2022

  38. [46]

    Mlp-mixer: An all-mlp architecture for vision,

    I. O. Tolstikhin, N. Houlsby, A. Kolesnikov, L. Beyer, X. Zhai, T. Un- terthiner, J. Yung, A. Steiner, D. Keysers, J. Uszkoreitet al., “Mlp-mixer: An all-mlp architecture for vision,” Advances in neural information processing systems, vol. 34, pp. 24 261–24 272, 2021

    2021

  39. [47]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

    2014 arXiv

  40. [48]

    Long short-term memory,

    A. Graves and A. Graves, “Long short-term memory,” Supervised sequence labelling with recurrent neural networks , pp. 37–45, 2012

    2012

  41. [49]

    Adam: A method for stochastic optimization,

    D. P. Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.