pith. machine review for the scientific record. sign in

arxiv: 2605.14481 · v1 · submitted 2026-05-14 · ⚛️ physics.optics

Recognition: 2 theorem links

· Lean Theorem

ML-assisted Subband Learned Digital Backpropagation for Nonlinearity Compensation in Wideband Optical Systems

Evgeny Shevelev , Oleg Sidelnikov , Vitaly Danilko , Mikhail Fedoruk , Alexey Redyuk
Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:50 UTC · model grok-4.3

classification ⚛️ physics.optics
keywords subband digital backpropagationnonlinearity compensationwideband optical systemslearned digital signal processingchromatic dispersion compensationWDM transmissioncoherent opticsmachine learning for communications
0
0 comments X

The pith

Subband learned digital backpropagation achieves better nonlinearity compensation at lower complexity in wideband optical systems.

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

Digital backpropagation compensates nonlinear distortions in coherent optical fiber links but becomes impractical for wideband signals because of large channel memory and the need for fine spatial steps. The paper proposes splitting the received waveform into subbands so chromatic dispersion can be compensated independently in the frequency domain with reduced memory, then handling intra- and inter-subband nonlinear effects with a trainable multi-input multi-output filter whose coefficients are learned end-to-end. Sparsification removes insignificant filter taps to cut computation further. Simulations of an 11-channel 40-Gbaud WDM 16QAM link over 2000 km show the method delivers higher signal-to-noise ratio gains than conventional or enhanced DBP while using fewer propagation steps, especially in low- and medium-complexity regimes.

Core claim

The central claim is that decomposing the received signal into multiple subbands, compensating chromatic dispersion per subband in the frequency domain, and correcting nonlinear intra- and inter-subband interactions via a jointly trained MIMO filtering structure yields a superior performance-complexity trade-off for wideband WDM transmission compared with standard digital backpropagation.

What carries the argument

Subband decomposition followed by a trainable MIMO time-domain filter whose coefficients are optimized by end-to-end gradient descent.

Load-bearing premise

The combination of subband splitting and the learned MIMO structure captures the dominant nonlinear interactions without leaving large residual modeling errors when the same fiber or hardware conditions are used.

What would settle it

Applying the trained SbL-DBP filters to a transmission experiment that uses a different fiber type or measured hardware impairments and checking whether the reported SNR gains over conventional DBP still appear.

Figures

Figures reproduced from arXiv: 2605.14481 by Alexey Redyuk, Evgeny Shevelev, Mikhail Fedoruk, Oleg Sidelnikov, Vitaly Danilko.

Figure 2
Figure 2. Figure 2: The center frequency of the j-th subband is defined as fj = − W 2 + W Nsb [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Scheme of a proposed NLC consisting of analysis filter bank, a subband learned DBP module and synthesis filter bank with [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Scheme of the wideband signal spectrum: W – spectral support of the useful signal components, Fs – sampling rate, Rs – symbol rate, fj – center frequency of the j-th subband, j ∈ {1, . . . , Nsb}. 2) Subband processing: The set of subband signals vj [m], obtained after AFB, is fed into the SbL-DBP processing block. The overall processing flow is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: (a) End-to-end learning framework of the SbL-DBP, represented [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Simulation and DSP pipeline of the considered WDM system. RRC – root-raised cosine, MUX – multiplexer, SSFM – split-step Fourier method, [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: SNR gain as a function of the number of steps [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: shows the learned coefficients for Nsb = 4, Nc = 16, and Nst = 20. The colored solid curves correspond to the co￾efficients obtained without imposing the symmetry and trans￾lational invariance constraints, while the black dashed curves show the vectors cp computed under (15)-(17) constraints. In both cases, the coefficients are normalized with respect to the maximum absolute value. The notation C∗ jl in th… view at source ↗
Figure 8
Figure 8. Figure 8: SNR gain ∆SNR as a function of the weight decay θ for the SbL￾DBP with Nst = 10, Nsb = 4, and Nc = 8. Inset: (a) distribution of the learned MIMO coefficient magnitudes for θ = 5 · 10−4 , (b) coefficient distribution after sparsification. Based on this trade-off, we select the operating point corresponding to θ = 5·10−4 , indicated by the blue star marker in [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Improvement of the signal-to-noise ratio [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: SNR gain as a function of the number of subbands [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: SNR as a function of launch power per channel for SbL-DBP, [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
read the original abstract

Digital backpropagation (DBP) is one of the most effective techniques for compensating nonlinear distortions in coherent optical fiber communication systems. However, its practical application to wideband transmission remains limited by high computational complexity caused by large channel memory and the requirement for fine spatial discretization. In this work, we propose a subband-based learned digital backpropagation (SbL-DBP) framework for wideband optical transmission systems. The received signal is decomposed into multiple subbands, enabling independent frequency-domain compensation of the chromatic dispersion with reduced effective channel memory and lower computational complexity. Nonlinear intra- and inter-subband interactions are addressed in the time domain using a trainable multi-input multi-output filtering structure. The parameters of the proposed framework are jointly optimized using end-to-end gradient-based learning. In addition, sparsification techniques are employed to remove insignificant coefficients and further reduce computational complexity. Numerical simulations of an 11$\times$40~Gbaud WDM RRC-16QAM 20$\times$100 km transmission system demonstrate that the proposed method provides a superior performance--complexity trade-off compared to conventional DBP and enhanced DBP. In the low- and medium-complexity regimes, SbL-DBP provides higher signal-to-noise ratio gains while requiring fewer propagation steps.

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 manuscript proposes a subband-based learned digital backpropagation (SbL-DBP) framework for nonlinearity compensation in wideband optical systems. The received signal is split into subbands for independent frequency-domain chromatic dispersion compensation with reduced memory, followed by trainable MIMO filters in the time domain to model intra- and inter-subband Kerr nonlinearities. Parameters are jointly optimized via end-to-end gradient descent, with sparsification applied to prune insignificant coefficients. Numerical simulations of an 11×40 Gbaud WDM RRC-16QAM system over 20×100 km spans demonstrate that SbL-DBP achieves higher SNR gains than conventional DBP and enhanced DBP at equivalent or lower complexity (measured in propagation steps and sparsified coefficients), particularly in low- and medium-complexity regimes.

Significance. If the simulation results hold under broader validation, the work offers a practical path to deploy DBP in wideband systems by mitigating the complexity barrier from large CD memory and fine discretization. The combination of subband decomposition, learned MIMO structures, and sparsification could enable higher spectral efficiency or extended reach in coherent optical networks without prohibitive DSP overhead.

major comments (3)
  1. [Numerical results] Numerical results section: The reported SNR gains are presented without error bars, confidence intervals, or details on the number of Monte Carlo realizations, making it difficult to assess statistical significance of the claimed superiority over enhanced DBP baselines.
  2. [Training and optimization] Section on training procedure: The end-to-end optimization is described at a high level, but the manuscript does not specify the training dataset size, exact split between training/validation/test sequences, or whether generalization was tested on fiber parameters (e.g., different dispersion or nonlinearity coefficients) outside the training distribution.
  3. [Complexity analysis] Complexity evaluation: While propagation steps and sparsified coefficients are used as metrics, the overhead introduced by the subband decomposition (filter banks, recombination) and the MIMO filter implementation is not quantified in equivalent real-multiplication counts, weakening the direct comparison to enhanced DBP.
minor comments (2)
  1. [Figures] Figure captions for the system diagram and complexity curves should explicitly state the number of subbands used and the sparsification threshold applied in each curve.
  2. [Method description] Notation for the MIMO filter coefficients should be introduced earlier and used consistently when describing the sparsification step.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for minor revision. The comments highlight important aspects of statistical rigor, training transparency, and complexity quantification that will strengthen the manuscript. We address each point below.

read point-by-point responses

  1. Referee: [Numerical results] Numerical results section: The reported SNR gains are presented without error bars, confidence intervals, or details on the number of Monte Carlo realizations, making it difficult to assess statistical significance of the claimed superiority over enhanced DBP baselines.

    Authors: We agree that error bars and details on Monte Carlo runs are necessary to establish statistical significance. In the revised manuscript, we will add error bars showing the standard deviation over 20 independent Monte Carlo realizations (different random seeds for data and noise), along with 95% confidence intervals. These additions confirm that the reported SNR gains over enhanced DBP remain statistically significant in the low- and medium-complexity regimes. revision: yes

  2. Referee: [Training and optimization] Section on training procedure: The end-to-end optimization is described at a high level, but the manuscript does not specify the training dataset size, exact split between training/validation/test sequences, or whether generalization was tested on fiber parameters (e.g., different dispersion or nonlinearity coefficients) outside the training distribution.

    Authors: We will revise the training section to specify that the dataset comprises 5000 symbol sequences per subband, using a 70/15/15 split for training/validation/test. The optimizer and learning-rate schedule will also be detailed. On generalization, the primary results use nominal fiber parameters; we will add a short study showing performance under ±10% variations in dispersion and nonlinearity coefficients, demonstrating robustness of the learned filters. This addresses the concern while noting that full out-of-distribution testing is beyond the current scope. revision: partial

  3. Referee: [Complexity analysis] Complexity evaluation: While propagation steps and sparsified coefficients are used as metrics, the overhead introduced by the subband decomposition (filter banks, recombination) and the MIMO filter implementation is not quantified in equivalent real-multiplication counts, weakening the direct comparison to enhanced DBP.

    Authors: We accept that a unified real-multiplication metric strengthens the comparison. In the revision we will add an explicit complexity table that counts real multiplications per symbol, incorporating the analysis/synthesis filter-bank overhead, recombination, and the cost of the sparsified MIMO filters. The enhanced DBP baseline will be expressed in the same units for direct comparison. This shows that the subband approach still yields a favorable trade-off at low and medium complexity. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces a data-driven SbL-DBP framework that decomposes the signal into subbands for frequency-domain CD compensation, applies trainable MIMO filters for nonlinear interactions in the time domain, and optimizes all parameters end-to-end via gradient descent on simulated transmission data. Performance is evaluated via numerical simulations of an 11×40 Gbaud WDM system, reporting SNR gains versus conventional and enhanced DBP at given complexity (propagation steps and sparsified coefficients). No equation or claim reduces by construction to a fitted parameter renamed as prediction, no load-bearing self-citation chain is invoked to justify uniqueness, and the central results remain externally falsifiable against standard DBP benchmarks without tautological redefinition of inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The framework rests on standard split-step Fourier models of fiber propagation and on the assumption that gradient-based training on simulated data will generalize; no new physical entities are introduced.

free parameters (2)
  • MIMO filter coefficients
    Learned via end-to-end gradient descent; their values are not fixed a priori.
  • Sparsification thresholds
    Chosen to prune insignificant coefficients after training.
axioms (1)
  • domain assumption Subband decomposition sufficiently isolates nonlinear interactions for the MIMO stage to correct
    Invoked when the received signal is split and processed independently per subband.

pith-pipeline@v0.9.0 · 5540 in / 1308 out tokens · 40021 ms · 2026-05-15T01:50:54.602776+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

27 extracted references · 27 canonical work pages

  1. [1]

    Five AI trends impacting telecom in 2026,

    D. Barnes, “Five AI trends impacting telecom in 2026,” Adtran Blog, Feb. 26, 2026. [Blog]. Accessed: Mar. 19, 2026. [Online]. Available: https://www.blog.adtran.com/en/five-ai-trends-impacting-telecom-in-2026

    work page 2026

  2. [2]

    Roadmap on optical communications,

    E. Agrell, M. Karlsson, F. Poletti, S. Namiki, X. Chen, L. A. Rusch, B. Puttnam, P. Bayvel, L. Schmalen, Z. Tao, F. R. Kschischang, A. Alvarado, B. Mukherjee, R. Casellas, X. Zhou, D. van Veen, G. Mohs, E. Wong, A. Mecozzi, M. Al-Slim, E. Diamanti, and M. Uysa, “Roadmap on optical communications,”Journal of Optics, vol. 26, no. 9, Art. 093001, 2024

    work page 2024

  3. [3]

    Fiber-optic trans- mission and networking: the previous 20 and the next 20 years,

    P. J. Winzer, D. T. Neilson, and A. R. Chraplyvy, “Fiber-optic trans- mission and networking: the previous 20 and the next 20 years,”Opt. Express, vol. 26, no. 18, pp. 24190–24239, 2018

    work page 2018

  4. [4]

    A tutorial on fiber Kerr nonlinearity effect and its compensation in optical communication systems,

    S. K. O. Soman, “A tutorial on fiber Kerr nonlinearity effect and its compensation in optical communication systems,”Journal of Optics, vol. 23, no. 12, Art. 123502, 2021

    work page 2021

  5. [5]

    A Survey on Fiber Nonlinearity Compensation for 400 Gb/s and beyond Optical Communication Systems,

    A. Amari, O. A. Dobre, R. Venkatesan, O. S. Sunish Kumar, P. Ciblat, and Y . Jaouen, “A Survey on Fiber Nonlinearity Compensation for 400 Gb/s and beyond Optical Communication Systems,”IEEE Communications Surveys & Tutorials, vol. 19, iss. 4, pp. 3097–3113, 2017

    work page 2017

  6. [6]

    Digital coherent optical receivers: algorithms and subsys- tems,

    S. Savory, “Digital coherent optical receivers: algorithms and subsys- tems,”IEEE J. Sel. Top. Quantum Electron., vol. 16, iss.5, pp. 1164–1179, 2010

    work page 2010

  7. [7]

    Advanced DSP for coherent optical fiber communication,

    J. Zhao, Y . Liu, and T. Xu, “Advanced DSP for coherent optical fiber communication,”Appl. Sci., vol. 9, iss. 19, Art. 4192, 2019

    work page 2019

  8. [8]

    Compensation of nonlinear signal distortions in optical fiber communication systems,

    A. Redyuk , O. Sidelnikov, and M. Fedoruk, “Compensation of nonlinear signal distortions in optical fiber communication systems,”Optics Com- munications, vol. 578, Art. 131418, 2025

    work page 2025

  9. [9]

    Compensation of Dispersion and Nonlinear Impairments Using Digital Backpropagation,

    E. Ip and J. M. Kahn, “Compensation of Dispersion and Nonlinear Impairments Using Digital Backpropagation,”J. Lightwave Technol., vol. 26, no. 20, pp. 3416–3425, 2008

    work page 2008

  10. [10]

    Low complexity split digital backpropagation for digital subcarrier-multiplexing optical transmissions,

    Z. Xiao, Q. Zhuge, S. Fu, F. Zhang, M. Qiu, M. Tang, D. Liu, and D. V . Plant, “Low complexity split digital backpropagation for digital subcarrier-multiplexing optical transmissions,”Optics Express, vol. 25, iss. 22. pp., 27824–27833, 2017

    work page 2017

  11. [11]

    Enhanced split-step Fourier method for digital backpropagation,

    M. Secondini, D. Marsella, and E. Forestieri, “Enhanced split-step Fourier method for digital backpropagation,” inThe European Conference on Optical Communication (ECOC), Cannes, France, 2014, pp. 1–3

    work page 2014

  12. [12]

    Correlated digital back propagation based on perturbation theory,

    X. Liang and S. Kumar, “Correlated digital back propagation based on perturbation theory,” inOptics Express, vol. 23, iss. 11., pp. 14655–14665, 2015

    work page 2015

  13. [13]

    ML-Enhanced Digital Back- propagation for Long-Reach Single-Span Systems,

    D. Cellini, S. Civelli, and M. Secondini, “ML-Enhanced Digital Back- propagation for Long-Reach Single-Span Systems,” arXiv, 2026

    work page 2026

  14. [14]

    Learned perturbation- based digital backpropagation with low complexity for nonlinearity compensation,

    A. Redyuk, E. Shevelev, V . Danilko, T. Bazarov, M. Senko, L. Samod- elkin, O. Nanii, V . Treshchikov, and M. Fedoruk, “Learned perturbation- based digital backpropagation with low complexity for nonlinearity compensation,”Opt. Continuum, vol. 4, no. 12, pp. 2896–2913, 2025

    work page 2025

  15. [15]

    Perturbation Theory-Aided Learned Digital Back-Propagation Scheme for Optical Fiber Nonlinearity Compensation,

    X. Lin, S. Luo, S. K. O. Soman, O. A. Dobre, L. Lampe, D. Chang, and C. Li, “Perturbation Theory-Aided Learned Digital Back-Propagation Scheme for Optical Fiber Nonlinearity Compensation,”J. Lightwave Technol., vol. 40, no. 7, pp. 1981–1988, 2022

    work page 1981

  16. [16]

    Advanced Convolutional Neural Networks for Nonlinearity Mitigation in Long-Haul WDM Transmission Systems,

    O. Sidelnikov, A. Redyuk, S. Sygletos, M. Fedoruk, and S. Turitsyn, “Advanced Convolutional Neural Networks for Nonlinearity Mitigation in Long-Haul WDM Transmission Systems,”J. Lightwave Technol., vol. 39, no. 8, pp. 2397–2406, 2021

    work page 2021

  17. [17]

    Computational Complexity Optimization of Neural Network-Based Equalizers in Digital Signal Processing: A Comprehensive Approach,

    P. Freire, S. Srivallapanondh, B. Spinnler, A. Napoli, N. Costa, J. E. Prilepsky, and S. K. Turitsyn, “Computational Complexity Optimization of Neural Network-Based Equalizers in Digital Signal Processing: A Comprehensive Approach,”J. Lightwave Technol., vol. 42, no. 12, pp. 4177–4201, 2024

    work page 2024

  18. [18]

    Deep Learning- Aided Perturbation Model-Based Fiber Nonlinearity Compensation,

    S. Luo, S. K. O. Soman, L. Lampe, and J. Mitra, “Deep Learning- Aided Perturbation Model-Based Fiber Nonlinearity Compensation,”J. Lightwave Technol., vol. 41, no. 12, pp. 3976–3985, 2023

    work page 2023

  19. [19]

    Revisiting Efficient Multi-Step Nonlinearity Compensation With Machine Learning: An Experimental Demonstration,

    V . Oliari, S. Goossens, C. Hager, G. Liga, R. M. Butler, M. van den Hout, S. van der Heide, H. D. Pfister, C. Okonkwo, and A. Alvarado, “Revisiting Efficient Multi-Step Nonlinearity Compensation With Machine Learning: An Experimental Demonstration,”J. Lightwave Technol., vol. 38, no. 12, pp. 3114–3124, 2020

    work page 2020

  20. [20]

    Physics-Based Deep Learning for Fiber- Optic Communication Systems,

    C. H ¨ager and H. D. Pfister, “Physics-Based Deep Learning for Fiber- Optic Communication Systems,”IEEE J. Sel. Areas Commun., vol. 39, no. 1, pp. 280–294, 2021

    work page 2021

  21. [21]

    Wideband Time-Domain Digital Back- propagation via Subband Processing and Deep Learning,

    C. H ¨ager and H. D. Pfister, “Wideband Time-Domain Digital Back- propagation via Subband Processing and Deep Learning,” inEuropean Conference on Optical Communication (ECOC), 2018

    work page 2018

  22. [22]

    Subband Processing for OQAM Digital-Subcarrier-Multiplexing Optical Coherent Systems,

    Z. Fan, J. Luo, B. Zheng, Y . Cai, and J. Zhao, “Subband Processing for OQAM Digital-Subcarrier-Multiplexing Optical Coherent Systems,” J. Lightwave Technol., vol. 43, no. 9, pp. 4218–4232, 2025

    work page 2025

  23. [23]

    Perturbation- aided fiber nonlinear compensation based on subband decomposition and synthesis for optical coherent systems [Invited],

    W. Guo, H. Deng, X. Tang, Z. Fan, Y . Cai, and J. Zhao, “Perturbation- aided fiber nonlinear compensation based on subband decomposition and synthesis for optical coherent systems [Invited],”Chin. Opt. Lett., vol. 24, no. 3, Art. 030003, 2026

    work page 2026

  24. [24]

    Complexity versus performance trade- off for fiber nonlinearity compensation using frequency-shaped, multi- subband backpropagation,

    E. Ip, N. Bai, and T. Wang, “Complexity versus performance trade- off for fiber nonlinearity compensation using frequency-shaped, multi- subband backpropagation,” in Proc. Opt. Fiber Commun. Conf. Expo. (OFC/NFOEC), Los Angeles, CA, USA, Mar. 2011, pp. 1–3

    work page 2011

  25. [25]

    A New Twist on Low-Complexity Digital Backpropagation,

    S. Civelli, D. P. Jana, E. Forestieri, and M. Secondini, “A New Twist on Low-Complexity Digital Backpropagation,”J. Lightwave Technol., vol. 43, no. 10, pp. 4679–4692, 2025

    work page 2025

  26. [26]

    Nonlinear Fiber Optics

    G. Agrawal, “Nonlinear Fiber Optics”, Elsevier. Boston, MA, USA. 2013

    work page 2013

  27. [27]

    Digital filters for coherent optical receivers,

    S. J. Savory, “Digital filters for coherent optical receivers,”Opt. express, vol. 16, no. 2, pp. 804–817, 2008

    work page 2008