pith. sign in

arxiv: 2511.11391 · v3 · submitted 2025-11-14 · 💻 cs.LG

SPOT: Single-Shot Positioning via Trainable Near-Field Rainbow Beamforming

Yeyue Cai , Jianhua Mo , Meixia Tao This is my paper

Pith reviewed 2026-05-17 21:57 UTC · model grok-4.3

classification 💻 cs.LG
keywords single-shot positioningrainbow beamformingphase-time arraydeep learningnear-field localizationtrue-time delaywireless sensingoverhead reduction
0
0 comments X

The pith

Trainable phase-time arrays create rainbow beams that recover user position from one transmission's peak power and frequency index.

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

The paper shows that making the phase shifter and true-time delay values trainable lets a network design frequency-dependent beams optimized for localization rather than communication. After a single downlink transmission, the user reports only the strongest quantized power and the subcarrier index where it occurred. A small fully connected network then maps that pair directly to the user's angle and range. The result is an order-of-magnitude drop in overhead together with lower two-dimensional positioning error than both analytical rainbow designs and prior learning-based schemes.

Core claim

An end-to-end deep learning scheme jointly optimizes the phase shifter and true-time delay coefficients of a phase-time array to synthesize near-field rainbow beams, then recovers angle-range coordinates from the maximum quantized received power and its subcarrier index after one transmission, producing lower positioning error at roughly one-tenth the overhead of existing analytical and learning-based methods.

What carries the argument

The trainable phase shifter and true-time delay coefficients that generate task-oriented near-field rainbow beams, combined with a lightweight fully connected module that decodes position from the peak power feedback and its subcarrier index.

If this is right

  • Localization requires only one downlink transmission instead of repeated measurements.
  • Overhead drops by roughly an order of magnitude relative to prior schemes.
  • Two-dimensional positioning error stays lower than both analytical rainbow beam designs and other learned methods.
  • The approach applies directly to wideband near-field scenarios using phase-time arrays.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same trainable-beam idea could be extended to joint velocity estimation by adding Doppler-sensitive feedback.
  • In dense networks the reduced pilot count would lower overall latency for tracking many users.
  • Hardware impairments such as phase noise in the true-time delays would need explicit modeling to preserve the reported accuracy gains.

Load-bearing premise

Optimizing the beam coefficients end-to-end will produce rainbow beams whose single strongest power measurement and frequency index contain enough information for a simple network to recover accurate angle and range.

What would settle it

A controlled near-field wideband experiment in which the single-transmission positioning error of the trained beams exceeds that of a multi-transmission analytical baseline under identical hardware constraints would disprove the overhead and accuracy claims.

Figures

Figures reproduced from arXiv: 2511.11391 by Jianhua Mo, Meixia Tao, Yeyue Cai.

Figure 2
Figure 2. Figure 2: System architecture of the proposed SPOT scheme. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Maximum received power of rainbow beam (dBm). [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Distance sensing performance of CBS approach. (a) [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: 2D RMSE under different power quantization bit length. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
read the original abstract

Phase-time arrays, which integrate phase shifters (PSs) and true-time delays (TTDs), have emerged as a cost-effective architecture for generating frequency-dependent rainbow beams in wideband sensing and localization. This paper proposes an end-to-end deep learning-based scheme that simultaneously designs the rainbow beams and estimates user positions. Treating the PS and TTD coefficients as trainable variables allows the network to synthesize task-oriented beams that maximize localization accuracy. A lightweight fully connected module then recovers the user's angle-range coordinates from its feedback of the maximum quantized received power and its corresponding subcarrier index after a single downlink transmission. Compared with existing analytical and learning-based schemes, the proposed method reduces overhead by an order of magnitude and delivers consistently lower two-dimensional positioning error.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The paper proposes SPOT, an end-to-end deep learning framework for single-shot 2D user positioning in near-field scenarios. Phase shifter (PS) and true-time delay (TTD) coefficients in a phase-time array are treated as trainable variables and optimized jointly with a lightweight fully connected decoder. The resulting frequency-dependent rainbow beams are designed so that the subcarrier index of the maximum received power, together with its quantized value, encodes the user's angle-range coordinates after one downlink transmission. The work claims an order-of-magnitude reduction in overhead and consistently lower positioning error relative to existing analytical and learning-based baselines.

Significance. If the central performance claims hold under realistic channel conditions and noise, the approach would represent a meaningful advance in low-overhead wideband localization by turning hardware degrees of freedom into task-specific beams via end-to-end training. The single-shot nature and minimal feedback (one index plus one quantized scalar) address a practical bottleneck in multi-shot methods. The trainable rainbow-beam design is a concrete example of hardware-aware, data-driven optimization that could generalize to other sensing tasks.

major comments (2)
  1. [Method description / simulation setup] The central claim that a single (argmax subcarrier index + quantized power) pair suffices for unique 2D position recovery rests on the learned beam pattern having no significant sidelobes or position-grid ambiguities. This assumption is load-bearing for the order-of-magnitude overhead reduction; without explicit verification (e.g., via a uniqueness plot or collision count over the discretized angle-range grid), the mapping could fail under small perturbations.
  2. [Numerical results / decoder architecture] The recovery module is described as a lightweight fully connected network that inverts the mapping from only the maximum quantized power and its subcarrier index. If the paper reports results only under idealized channel models without model mismatch or hardware impairments, the consistently lower error versus multi-shot baselines cannot yet be considered robust (see also the absence of error bars or ablation on quantization levels).
minor comments (3)
  1. [Feedback model] Clarify the exact quantization bit-width used for the received power and whether it is fixed or also learned; this directly affects the information available to the decoder.
  2. [Numerical results] Add a brief comparison table that includes both overhead (number of transmissions or pilots) and achieved RMSE for all baselines under identical array size, bandwidth, and SNR conditions.
  3. [Training procedure] Specify the training loss (positioning MSE? CRLB-weighted?) and whether any regularization is applied to encourage beam patterns with low sidelobes.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We address each of the major comments below, providing clarifications and indicating the revisions we will make to strengthen the paper.

read point-by-point responses

  1. Referee: [Method description / simulation setup] The central claim that a single (argmax subcarrier index + quantized power) pair suffices for unique 2D position recovery rests on the learned beam pattern having no significant sidelobes or position-grid ambiguities. This assumption is load-bearing for the order-of-magnitude overhead reduction; without explicit verification (e.g., via a uniqueness plot or collision count over the discretized angle-range grid), the mapping could fail under small perturbations.

    Authors: We agree that explicit verification of the uniqueness of the position-to-feedback mapping would strengthen the central claim. The end-to-end training objective directly minimizes positioning error, which implicitly encourages beam patterns with reduced ambiguities and sidelobes. Nevertheless, to address this concern directly, we will add to the revised manuscript an analysis (e.g., a uniqueness plot or collision count) over the discretized angle-range grid for the learned beams. This will demonstrate that the single (argmax subcarrier index + quantized power) pair enables unique recovery with negligible collisions under the considered setup. revision: yes

  2. Referee: [Numerical results / decoder architecture] The recovery module is described as a lightweight fully connected network that inverts the mapping from only the maximum quantized power and its subcarrier index. If the paper reports results only under idealized channel models without model mismatch or hardware impairments, the consistently lower error versus multi-shot baselines cannot yet be considered robust (see also the absence of error bars or ablation on quantization levels).

    Authors: We thank the referee for this important point. Our current simulations are performed under a near-field channel model with additive noise, but we acknowledge that they do not yet include hardware impairments or model mismatch. To improve robustness assessment, we will include error bars in the numerical results to show variability across random seeds or channel realizations. Additionally, we will provide an ablation study on the quantization levels of the received power to evaluate sensitivity. We will also add a discussion on potential limitations regarding hardware impairments and outline future work to incorporate them. revision: partial

Circularity Check

0 steps flagged

End-to-end trainable rainbow beamforming and single-shot recovery show no circular derivation

full rationale

The paper describes an end-to-end deep learning scheme that treats PS and TTD coefficients as trainable variables optimized directly against a positioning loss to synthesize task-oriented beams, followed by a lightweight FC module that recovers angle-range from the single-shot max quantized power and subcarrier index. No equations or steps reduce a claimed prediction or uniqueness result to its own fitted inputs by construction, nor rely on self-citations that import uniqueness theorems or ansatzes from prior author work. The central claim rests on supervised training for the localization task rather than any self-definitional loop or renamed known result, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that joint optimization of beam coefficients via back-propagation will produce beams that are optimal for the positioning task and that the quantized power-plus-index feedback is sufficient to invert for 2D coordinates.

free parameters (1)
  • PS and TTD coefficients
    Treated as trainable variables inside the neural network; their values are learned from data rather than derived analytically.
axioms (1)
  • domain assumption The near-field channel model and frequency-dependent beam pattern can be accurately simulated for training.
    Invoked implicitly when the network is trained on synthetic data.

pith-pipeline@v0.9.0 · 5426 in / 1288 out tokens · 28094 ms · 2026-05-17T21:57:23.380209+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

10 extracted references · 10 canonical work pages

  1. [1]

    Six Integration Avenues for ISAC in 6G and Beyond,

    N. Gonz ´alez-Prelcic, D. Tagliaferri, M. F. Keskin, H. Wymeersch, and L. Song, “Six Integration Avenues for ISAC in 6G and Beyond,”IEEE Vehicular Technology Magazine, vol. 20, no. 1, pp. 18–39, Mar. 2025

    work page 2025

  2. [2]

    Joint phase-time arrays: A paradigm for frequency-dependent analog beamforming in 6G,

    V . V . Ratnam, J. Mo, A. Alammouri, B. L. Ng, J. Zhang, and A. F. Molisch, “Joint phase-time arrays: A paradigm for frequency-dependent analog beamforming in 6G,”IEEE Access, vol. 10, pp. 73 364–73 377, 2022

    work page 2022

  3. [3]

    Frequency-Dependent Beamforming for Hy- brid Near-Far Field Communications Through Joint Phase-Time Arrays,

    Y . Cai, M. Tao, and S. Sun, “Frequency-Dependent Beamforming for Hy- brid Near-Far Field Communications Through Joint Phase-Time Arrays,” in2025 IEEE Wireless Communications and Networking Conference (WCNC), 2025, pp. 1–6

    work page 2025

  4. [4]

    Hybrid Near/Far-Field Frequency- Dependent Beamforming via Joint Phase-Time Arrays,

    Y . Cai, M. Tao, J. Mo, and S. Sun, “Hybrid Near/Far-Field Frequency- Dependent Beamforming via Joint Phase-Time Arrays,”arXiv preprint arXiv:2501.15207, 2025

    work page arXiv 2025

  5. [5]

    Beamforming with Joint Phase and Time Array: System Design, Prototyping and Performance,

    J. Mo, A. AlAmmouri, S. Dong, Y . Nam, W.-S. Choi, G. Xu, and J. C. Zhang, “Beamforming with Joint Phase and Time Array: System Design, Prototyping and Performance,” in2024 58th Asilomar Conference on Signals, Systems, and Computers. IEEE, 2024, pp. 875–881

    work page 2024

  6. [6]

    Integrated sensing and communications with joint beam-squint and beam-split for mmWave/THz massive MIMO,

    F. Gao, L. Xu, and S. Ma, “Integrated sensing and communications with joint beam-squint and beam-split for mmWave/THz massive MIMO,” IEEE Transactions on Communications, vol. 71, no. 5, pp. 2963–2976, 2023

    work page 2023

  7. [7]

    Beam Squint Assisted User Localization in Near-Field Integrated Sensing and Communications Systems,

    H. Luo, F. Gao, W. Yuan, and S. Zhang, “Beam Squint Assisted User Localization in Near-Field Integrated Sensing and Communications Systems,”IEEE Transactions on Wireless Communications, vol. 23, no. 5, pp. 4504–4517, 2024

    work page 2024

  8. [8]

    Deep Learning- Based Near-Field User Localization With Beam Squint in Wideband XL-MIMO Systems,

    H. Lei, J. Zhang, H. Xiao, D. W. K. Ng, and B. Ai, “Deep Learning- Based Near-Field User Localization With Beam Squint in Wideband XL-MIMO Systems,”IEEE Transactions on Wireless Communications, vol. 24, no. 2, pp. 1568–1583, 2025

    work page 2025

  9. [9]

    Near-Field Wideband Beam Training Based on Distance-Dependent Beam Split,

    T. Zheng, M. Cui, Z. Wu, and L. Dai, “Near-Field Wideband Beam Training Based on Distance-Dependent Beam Split,”IEEE Transactions on Wireless Communications, vol. 24, no. 2, pp. 1278–1292, 2025

    work page 2025

  10. [10]

    Deep Learning Based Near-Field Positioning in True- Time-Delay Array Systems,

    R. Klus, J. Talvitie, I. Pehlivan, M. C. Ilter, L. Klus, D. Cabric, and M. Valkama, “Deep Learning Based Near-Field Positioning in True- Time-Delay Array Systems,” in2025 IEEE 26th International Workshop on Signal Processing and Artificial Intelligence for Wireless Communi- cations (SPAWC), 2025, pp. 1–5

    work page 2025