pith. sign in

arxiv: 2606.05053 · v1 · pith:UVVBEJQ4new · submitted 2026-06-03 · 📡 eess.SP

Deep Learning Based Multi-Step Channel Prediction for Adaptive Underwater Acoustic OFDM Systems

Tian Tian , Ying Zhang , Agastya Raj , Fei-Yun Wu , Marco Ruffini This is my paper

Pith reviewed 2026-06-28 04:36 UTC · model grok-4.3

classification 📡 eess.SP
keywords underwater acoustic communicationsOFDMchannel predictionTransformeradaptive modulationdeep learningCSI forecastingbit error rate
0
0 comments X

The pith

A Transformer model with shared parameters predicts multiple future states of underwater acoustic channels to drive adaptive modulation and power allocation in OFDM systems.

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

The paper introduces PatchCSI-T, a Transformer-based model designed for multistep prediction of channel state information in underwater acoustic OFDM communications. It incorporates feature-independent modeling and parameter sharing to produce forecasts that support a greedy adaptive scheme for choosing modulation and power levels. On real-world UWA channel datasets, this combination yields measurable gains in end-to-end bit error rate and spectral efficiency while keeping latency low. A sympathetic reader would care because underwater channels change quickly due to motion and multipath, so timely forecasts can prevent the use of outdated channel knowledge that otherwise degrades performance. The work therefore centers on turning accurate future channel estimates into concrete transmission adaptations.

Core claim

The authors develop PatchCSI-T, a Transformer-based multistep channel prediction model with feature-independent modeling and parameter sharing, which, combined with a greedy adaptive modulation and power allocation scheme, enables accurate low-latency CSI forecasting and improves end-to-end BER and spectral efficiency on real-world UWA channel datasets.

What carries the argument

PatchCSI-T, a Transformer-based multistep channel prediction model with feature-independent modeling and parameter sharing that generates future CSI values for input to the adaptive allocation scheme.

If this is right

  • Accurate low-latency CSI forecasting becomes feasible for underwater acoustic OFDM.
  • End-to-end bit error rate decreases when the forecasts feed the greedy allocation scheme.
  • Spectral efficiency rises on the same real-world channel datasets.
  • The combined prediction-plus-adaptation framework operates without requiring frequent pilot overhead.

Where Pith is reading between the lines

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

  • The same prediction approach might reduce the need for frequent channel sounding in other rapidly varying wireless environments.
  • Lower prediction latency could support higher vehicle speeds or more frequent adaptation in mobile UWA links.
  • Replacing the greedy allocator with an optimization-based one might extract still larger gains from the same forecasts.

Load-bearing premise

The multi-step forecasts produced by PatchCSI-T remain accurate enough on real underwater channels to improve the decisions made by the greedy adaptive modulation and power allocation scheme.

What would settle it

Running the full adaptive OFDM system on the real-world UWA datasets with PatchCSI-T predictions yields no BER reduction and no spectral-efficiency gain relative to using the most recent measured CSI without prediction.

Figures

Figures reproduced from arXiv: 2606.05053 by Agastya Raj, Fei-Yun Wu, Marco Ruffini, Tian Tian, Ying Zhang.

Figure 1
Figure 1. Figure 1: System framework and channel prediction model. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Time-varying channel dataset: (a) CIRs, (b) CFRs. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: summarizes CFR predictions over horizons Tpred P t2, 4, 8, 16, 32, 64, 128u, including NMSE, per-epoch training time (Tepoch) and inference time (Tpred). PatchCSI-T con￾sistently achieves higher accuracy and faster inference. At Tpred “ 32, it attains -14.76 dB NMSE, a 1.4-4.9 dB NMSE gain over the baselines, with 14.7 ms inference („6ˆ faster than CNN1d-LSTM/BiGRU and ą 50ˆ faster than MTL￾LSTM). Notably,… view at source ↗
Figure 4
Figure 4. Figure 4: Multi-head attention weight matrices and prediction curves for selected subcarrier components. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Performance of the adaptive OFDM scheme under ground-truth CFR at SNR=5 dB: (a) number of active subcarriers [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: BER distribution under real and predicted CFRs. [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Spectral efficiency for the adaptive OFDM system. [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
read the original abstract

We develop an adaptive OFDM framework for underwater acoustic communications based on PatchCSI-T, a Transformer-based multistep channel prediction model with feature-independent modeling and parameter sharing. Combined with a greedy adaptive modulation and power allocation scheme, the proposed approach enables accurate, low-latency CSI forecasting and improves end-to-end BER and spectral efficiency on real-world UWA channel datasets.

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

3 major / 2 minor

Summary. The paper proposes PatchCSI-T, a Transformer-based multi-step channel prediction model using feature-independent modeling and parameter sharing, for adaptive underwater acoustic OFDM systems. Combined with a greedy adaptive modulation and power allocation scheme, it claims to enable accurate low-latency CSI forecasting and to improve end-to-end BER and spectral efficiency on real-world UWA channel datasets.

Significance. If the claimed BER and SE gains are rigorously validated, the work would be significant for practical adaptive communications in highly non-stationary UWA channels, where multi-step forecasting could enable better resource allocation than myopic schemes relying on outdated CSI.

major comments (3)
  1. [Abstract] Abstract: the central claim of improved end-to-end BER and spectral efficiency is asserted without any quantitative results, baselines, error bars, or validation details, preventing assessment of whether the data support the claim.
  2. [Results / Experiments] The manuscript does not isolate the contribution of PatchCSI-T multi-step forecasts to the reported gains; an ablation replacing the forecasts with perfect CSI, last-known CSI, or a naïve predictor when driving the identical greedy allocator on the same real traces is required, because moderate prediction error on non-stationary channels can cause the myopic allocator to select suboptimal constellations or powers.
  3. [Proposed Method / §3] The feature-independent modeling and parameter sharing in PatchCSI-T are presented as enabling accurate forecasts, yet no quantitative multi-step prediction metrics (e.g., NMSE at different horizons) or comparisons to standard predictors (LSTM, AR) are supplied to substantiate that these design choices remain useful inputs to the greedy scheme.
minor comments (2)
  1. Notation for the channel prediction horizon and the greedy allocation objective should be defined consistently between the model description and the performance evaluation sections.
  2. Figure captions should explicitly state the number of real-world traces, the prediction horizon used, and the exact baselines plotted.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and indicate the planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of improved end-to-end BER and spectral efficiency is asserted without any quantitative results, baselines, error bars, or validation details, preventing assessment of whether the data support the claim.

    Authors: We agree that the abstract should contain quantitative support for the central claims. In the revised manuscript we will update the abstract to report specific BER and spectral-efficiency gains (with baselines and error bars) obtained on the real UWA datasets. revision: yes

  2. Referee: [Results / Experiments] The manuscript does not isolate the contribution of PatchCSI-T multi-step forecasts to the reported gains; an ablation replacing the forecasts with perfect CSI, last-known CSI, or a naïve predictor when driving the identical greedy allocator on the same real traces is required, because moderate prediction error on non-stationary channels can cause the myopic allocator to select suboptimal constellations or powers.

    Authors: We accept the need for explicit isolation of the predictor's contribution. The revised manuscript will include the requested ablation experiments that replace PatchCSI-T forecasts with perfect CSI, last-known CSI, and a naïve predictor while keeping the identical greedy allocator and the same real traces. revision: yes

  3. Referee: [Proposed Method / §3] The feature-independent modeling and parameter sharing in PatchCSI-T are presented as enabling accurate forecasts, yet no quantitative multi-step prediction metrics (e.g., NMSE at different horizons) or comparisons to standard predictors (LSTM, AR) are supplied to substantiate that these design choices remain useful inputs to the greedy scheme.

    Authors: We will add the missing quantitative evidence. The revised §3 and results section will report multi-step NMSE at multiple horizons together with direct comparisons of PatchCSI-T against LSTM and AR predictors, confirming that the design choices improve the inputs supplied to the greedy allocator. revision: yes

Circularity Check

0 steps flagged

No significant circularity; claims rest on empirical evaluation rather than self-referential derivation

full rationale

The abstract and provided text describe a Transformer-based model (PatchCSI-T) trained for multi-step CSI prediction, then combined with a greedy allocator for BER/SE gains on real UWA datasets. No equations, fitted parameters, or self-citations are shown that reduce a claimed prediction or uniqueness result to the inputs by construction. The central claim is an empirical performance improvement, which is externally falsifiable on the stated datasets and does not invoke self-definitional loops, fitted-input-as-prediction, or load-bearing self-citation chains. A score of 2 reflects the normal minor self-citation tolerance without load-bearing reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5583 in / 933 out tokens · 26087 ms · 2026-06-28T04:36:30.768237+00:00 · methodology

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

20 extracted references · 6 canonical work pages

  1. [1]

    2024 , volume=

    Tian, Tian and Raj, Agastya and Xavier, Bruno Missi and Zhang, Ying and Wu, Fei-Yun and Yang, Kunde , journal=IEEE_J_WCOM, title=. 2024 , volume=

    2024

  2. [2]

    Hu, Xin and Huo, Yiming and Dong, Xiaodai and Wu, Fei-Yun and Huang, Aiping , issn =. Channel

  3. [3]

    Liu, Lei and Cai, Lin and Ma, Lu and Qiao, Gang , issn =. Channel

  4. [4]

    Distributed optimization and statistical learning via the alternating direction method of multipliers

    Boyd, Stephen , year =. Distributed. FNT in Machine Learning , volume =. doi:10.1561/2200000016 , langid =

    work page doi:10.1561/2200000016

  5. [5]

    2014 , volume =

    Parikh, Neal and Boyd, Stephen , year =. Proximal. OPT , volume =. doi:10.1561/2400000003 , abstract =

    work page doi:10.1561/2400000003

  6. [6]

    2014 , doi =

    Statistical Characterization of a Class of Underwater Acoustic Communication Channels , author =. 2014 , doi =

    2014

  7. [7]

    and Yu, Wenxian , year =

    Wen, Fei and Liu, Peilin and Liu, Yipeng and Qiu, Robert C. and Yu, Wenxian , year =. Robust

  8. [8]

    2019 , month = may, journal =

    Inexact Alternating Direction Methods of Multipliers for Separable Convex Optimization , author =. 2019 , month = may, journal =

    2019

  9. [9]

    arxiv , langid =:arXiv:1704.06209 , publisher =

    Wohlberg, Brendt , year =. arxiv , langid =:arXiv:1704.06209 , publisher =

    Pith/arXiv arXiv

  10. [10]

    2021 , journal =

    Acceleration. 2021 , journal =. doi:10.1561/2400000036 , urldate =. arxiv , langid =:2101.09545 , primaryclass =

    work page doi:10.1561/2400000036 2021

  11. [11]

    Beck, Amir and Teboulle, Marc , year =. A. SIAM Journal on Imaging Sciences , volume =. doi:10.1137/080716542 , urldate =

    work page doi:10.1137/080716542

  12. [12]

    Tony and Wang, Lie , year =

    Cai, T. Tony and Wang, Lie , year =. Orthogonal. IEEE Transactions on Information Theory , volume =. doi:10.1109/TIT.2011.2146090 , urldate =

    work page doi:10.1109/tit.2011.2146090 2011

  13. [13]

    Zhang, Yijian and Dall'Anese, Emiliano and Hong, Mingyi , year =. Online. IEEE Transactions on Signal and Information Processing over Networks , volume =. doi:10.1109/TSIPN.2021.3051292 , keywords =

    work page doi:10.1109/tsipn.2021.3051292 2021

  14. [14]

    and Proakis, John G

    Radosevic, Andreja and Ahmed, Rameez and Duman, Tolga M. and Proakis, John G. and Stojanovic, Milica , year = 2014, month = apr, journal = IEEE_J_OE, volume =. Adaptive

    2014

  15. [15]

    Efficiency

    Huang, Lihuan and Zhang, Qunfei and Zhang, Lifan and Shi, Juan and Zhang, Lingling , year = 2020, month = aug, journal = IEEE_J_COML, volume =. Efficiency

    2020

  16. [16]

    Underwater Acoustic Communication Channels:

    Stojanovic, Milica and Preisig, James , year = 2009, month = jan, journal = IEEE_M_COM, volume =. Underwater Acoustic Communication Channels:

    2009

  17. [17]

    Federated

    Zhao, Hao and Ji, Fei and Li, Qiang and Guan, Quansheng and Wang, Shuai and Wen, Miaowen , year = 2022, month = apr, journal =. Federated

    2022

  18. [18]

    Han, Lu and Ye, Han-Jia and Zhan, De-Chuan , year = 2024, month = nov, journal = IEEE_J_KDE, volume =. The

    2024

  19. [19]

    Zhu, Zhengliang and Tong, Feng and Zhou, Yuehai and Zhang, Ziqiao and Zhang, Fumin , year = 2023, month = sep, journal =. Deep

    2023

  20. [20]

    and Sinthong, Phanwadee and Kalagnanam, Jayant , year = 2023, month = mar, number =

    Nie, Yuqi and Nguyen, Nam H. and Sinthong, Phanwadee and Kalagnanam, Jayant , year = 2023, month = mar, number =. A. arXiv , langid =:2211.14730 , primaryclass =

    Pith/arXiv arXiv 2023