pith. sign in

arxiv: 2606.27042 · v1 · pith:U2AKK7UVnew · submitted 2026-06-25 · 📡 eess.SP · physics.optics

Low Complexity Kolmogorov-Arnold Network-based DPD for Analog RoF Fronthaul

Carlos Daniel Fontes da Silva , Tianyu Jiang , Lu Zhang , Vjaceslavs Bobrovs , Xianbin Yu , Xiaodan Pang , Oskars Ozolins , Edson Porto da Silva This is my paper

Pith reviewed 2026-06-26 02:21 UTC · model grok-4.3

classification 📡 eess.SP physics.optics
keywords digital predistortionKolmogorov-Arnold networksanalog radio-over-fibernonlinear distortioncomputational complexityACLREVMsymbolic models
0
0 comments X

The pith

A symbolized Kolmogorov-Arnold Network digital predistortion model delivers neural-network level performance for analog radio-over-fiber at near-polynomial computational cost.

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

The paper introduces an envelope time-delay Kolmogorov-Arnold Network (ETDKAN) for digital predistortion in analog RoF fronthaul to reduce nonlinear distortions. By applying KAN symbolization, it creates a symbolic version (symbETDKAN) that keeps ACLR and EVM performance close to multilayer perceptron models. At the same time, its complexity stays near that of traditional memory polynomials. Experimental tests confirm a 4-5 dB ACLR improvement, showing the approach works in real hardware. This matters because it could allow more efficient, interpretable linearization without the high compute demands of deep neural networks.

Core claim

The ETDKAN model incorporates physical constraints of RF nonlinear devices and, through KAN symbolization, achieves a significant reduction in computational complexity while improving interpretability. The symbolic ETDKAN attains ACLR and EVM performance comparable to neural network-based models, while maintaining a computational complexity close to that of memory polynomials, with experimental validation showing 4-5 dB ACLR reduction.

What carries the argument

The envelope time-delay KAN (ETDKAN) with subsequent symbolization to produce symbETDKAN, which embeds RF device constraints into a spline-based network structure for low-complexity predistortion.

Load-bearing premise

The KAN symbolization step preserves the physical constraints of RF nonlinear devices without requiring post-hoc tuning that would increase complexity back toward MLP levels.

What would settle it

A direct comparison experiment where symbETDKAN is applied to the same A-RoF setup but fails to achieve the reported ACLR reduction or requires complexity adjustments exceeding memory polynomial levels would falsify the claim.

Figures

Figures reproduced from arXiv: 2606.27042 by Carlos Daniel Fontes da Silva, Edson Porto da Silva, Lu Zhang, Oskars Ozolins, Tianyu Jiang, Vjaceslavs Bobrovs, Xianbin Yu, Xiaodan Pang.

Figure 1
Figure 1. Figure 1: KAN diagram of a network representing a function [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: In the symbolic model, an activation function ϕ(x) is approximated by an expression of the form ϕˆ(x) = cf(ax + b)+d, where a, b, c, and d are the affine parameters, and f(·) an analytical function. Initially, the values c = 1 and d = 0 are fixed, and through a grid search, the values of a and b that maximize the coefficient of determination r 2 [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Diagram of proposed DPD based on envelope time-delay Kolmogorov- [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Indirect Learning Architecture (ILA) for DPD training. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: System model considered in this paper. (a) Block diagram for the A-RoF transmission system model; (b) Signal processing assumed for the OFDM [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Gain and phase shift characteristic curves of PA as a function of input [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Pareto fronts from multi-objective optimization of DPD’s hyperparameters for transmission metrics and complexity. (a) EVM vs NFLOP; (b) ACLR [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Symbolic model obtained for symbETDKAN-DPD. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: (a) Block diagram and (b) photo of experimental setup used for DPD [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Spectrum of the received signals without and with DPD. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗
read the original abstract

This paper proposes and demonstrates experimentally for the first time a Kolmogorov-Arnold Network (KAN)-based digital predistortion (DPD) model, named envelope time-delay KAN (ETDKAN), for mitigating nonlinear distortions in analog radio-over-fiber (A-RoF) systems. The ETDKAN model incorporates physical constraints of radio-frequency (RF) nonlinear devices and, through KAN symbolization, achieves a significant reduction in computational complexity while improving interpretability. The proposed model is numerically implemented and optimized alongside multilayer perceptron (MLP) and memory-polynomial-based DPDs. Results show that the resulting symbolic ETDKAN (symbETDKAN) attains ACLR and EVM performance comparable to neural network-based models, while maintaining a computational complexity close to that of memory polynomials. Experimental validation using an A-RoF system confirms the practical feasibility of the proposed approach, which resulted in a 4-5 dB reduction in ACLR in the analyzed scenario.

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 / 0 minor

Summary. The manuscript proposes the first Kolmogorov-Arnold Network (KAN)-based digital predistortion (DPD) model, termed envelope time-delay KAN (ETDKAN), for mitigating nonlinear distortions in analog radio-over-fiber (A-RoF) fronthaul. The model incorporates RF device physical constraints; KAN symbolization is used to produce a symbolic variant (symbETDKAN) that is claimed to match multilayer-perceptron (MLP) ACLR/EVM performance while retaining computational complexity comparable to memory-polynomial (MP) baselines. Numerical optimization against MLP and MP models plus experimental A-RoF validation are reported, with a claimed 4-5 dB ACLR improvement.

Significance. If the complexity reduction after symbolization is verified by explicit operation counts, this would constitute a meaningful advance: the first experimental KAN DPD demonstration that bridges the interpretability and low-complexity advantages of polynomials with the modeling power of neural networks, directly relevant to real-time fronthaul hardware constraints.

major comments (2)
  1. [Abstract] Abstract: the central claim that symbETDKAN 'maintains a computational complexity close to that of memory polynomials' after KAN symbolization is unsupported by any tabulated operation counts, basis-function counts, or before/after complexity metrics; without these data the reduction cannot be confirmed to survive symbolization and the comparison to MLP remains unverifiable.
  2. [Abstract] Abstract (experimental validation paragraph): the reported 4-5 dB ACLR reduction is presented without accompanying details on measurement uncertainty, number of independent trials, or statistical tests, which is load-bearing for the claim of practical feasibility and parity with neural-network models.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below, indicating the revisions we will incorporate to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that symbETDKAN 'maintains a computational complexity close to that of memory polynomials' after KAN symbolization is unsupported by any tabulated operation counts, basis-function counts, or before/after complexity metrics; without these data the reduction cannot be confirmed to survive symbolization and the comparison to MLP remains unverifiable.

    Authors: We agree that the abstract would benefit from explicit supporting metrics. The full manuscript includes complexity comparisons between ETDKAN, symbETDKAN, MLP, and memory-polynomial models. In the revision we will add a dedicated table (or subsection) reporting operation counts, basis-function counts, and before/after metrics after symbolization. We will also revise the abstract to include a concise quantitative statement referencing these results, thereby making the complexity claim directly verifiable. revision: yes

  2. Referee: [Abstract] Abstract (experimental validation paragraph): the reported 4-5 dB ACLR reduction is presented without accompanying details on measurement uncertainty, number of independent trials, or statistical tests, which is load-bearing for the claim of practical feasibility and parity with neural-network models.

    Authors: The reported 4-5 dB ACLR improvement was obtained from the described A-RoF experimental campaign. We will revise both the abstract and the experimental-results section to supply the requested supporting information: measurement uncertainty estimates, the number of independent trials, and any statistical tests performed. These additions will strengthen the evidence for practical feasibility and performance parity. revision: yes

Circularity Check

0 steps flagged

No circularity: model proposal and experimental validation are independent of fitted inputs

full rationale

The paper proposes ETDKAN as a new architecture that incorporates RF physical constraints by design and uses KAN symbolization for complexity reduction. Performance claims rest on numerical optimization and experimental A-RoF measurements compared to external MLP and memory-polynomial baselines. No equations reduce a prediction to a fitted parameter by construction, no self-citation chain justifies a uniqueness claim, and no ansatz is smuggled via prior work. The central result (comparable ACLR/EVM at near-memory-polynomial complexity) is externally falsifiable via the reported measurements and does not collapse to input data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities; the ETDKAN structure itself is introduced as a modeling choice whose internal parameters would be fitted during training.

pith-pipeline@v0.9.1-grok · 5739 in / 1096 out tokens · 39552 ms · 2026-06-26T02:21:18.544960+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

18 extracted references · 1 canonical work pages

  1. [1]

    6G Wireless Communications Networks: A Com- prehensive Survey,

    M. Alsabahet al., “6G Wireless Communications Networks: A Com- prehensive Survey,”IEEE Access, vol. 9, pp. 148 191–148 243, 2021

    2021

  2. [2]

    A Comprehensive Survey on Millimeter Wave Communications for Fifth-Generation Wireless Networks: Feasibility and Challenges,

    A. N. Uwaechia and N. M. Mahyuddin, “A Comprehensive Survey on Millimeter Wave Communications for Fifth-Generation Wireless Networks: Feasibility and Challenges,”IEEE Access, vol. 8, pp. 62 367– 62 414, 2020

    2020

  3. [3]

    Radio-Over-Fiber Technology: Present and Future,

    C. Lim and A. Nirmalathas, “Radio-Over-Fiber Technology: Present and Future,”Journal of Lightwave Technology, vol. 39, no. 4, pp. 881–888, 2021

    2021

  4. [4]

    A Generalized Memory Polynomial Model for Digital Predistortion of RF Power Amplifiers,

    D. R. Morganet al., “A Generalized Memory Polynomial Model for Digital Predistortion of RF Power Amplifiers,”IEEE Transactions on Signal Processing, vol. 54, no. 10, pp. 3852–3860, 2006

    2006

  5. [5]

    Augmented Real-Valued Time-Delay Neural Network for Compensation of Distortions and Impairments in Wireless Trans- mitters,

    D. Wanget al., “Augmented Real-Valued Time-Delay Neural Network for Compensation of Distortions and Impairments in Wireless Trans- mitters,”IEEE Transactions on Neural Networks and Learning Systems, vol. 30, no. 1, pp. 242–254, 2019

    2019

  6. [6]

    Efficient Digital Predistortion Using Sparse Neural Network,

    M. Tanio, N. Ishii, and N. Kamiya, “Efficient Digital Predistortion Using Sparse Neural Network,”IEEE Access, vol. 8, pp. 117 841–117 852, 2020

    2020

  7. [7]

    Base-band Derived V olterra Series for Power Amplifier Modeling,

    E. G. Limaet al., “Base-band Derived V olterra Series for Power Amplifier Modeling,” in2009 IEEE MTT-S International Microwave Symposium Digest. Boston, MA, USA: IEEE, 2009

    2009

  8. [8]

    KAN: Kolmogorov–Arnold Networks,

    Z. Liuet al., “KAN: Kolmogorov–Arnold Networks,” inThe Thirteenth International Conference on Learning Representations, 2025. [Online]. Available: https://openreview.net/forum?id=Ozo7qJ5vZi

    2025

  9. [9]

    Real-Valued Time Delay Kolmogorov-Arnold Network for Digital Predistortion of RF Power Amplifiers,

    J. Chenet al., “Real-Valued Time Delay Kolmogorov-Arnold Network for Digital Predistortion of RF Power Amplifiers,” in2025 International Conference on Microwave and Millimeter Wave Technology (ICMMT). Xi’an, China: IEEE, 2025

    2025

  10. [10]

    Linearization of RF Power Amplifiers in Wideband Communication Systems by Adaptive Indirect Learning Using RPEM Algorithm,

    D. H. Leet al., “Linearization of RF Power Amplifiers in Wideband Communication Systems by Adaptive Indirect Learning Using RPEM Algorithm,”Mobile Networks and Applications, vol. 25, pp. 1988–1997, 2020

    1988

  11. [11]

    Novel Machine Learning Linearization Scheme for 6G A-RoF Systems,

    L. A. M. Pereiraet al., “Novel Machine Learning Linearization Scheme for 6G A-RoF Systems,”Journal of Lightwave Technology, vol. 41, no. 23, pp. 7245–7252, 2023

    2023

  12. [12]

    Seimetz,High-Order Modulation for Optical Fiber Transmission, ser

    M. Seimetz,High-Order Modulation for Optical Fiber Transmission, ser. Springer Series in Optical Sciences. Berlin, Heidelberg: Springer, 2009

    2009

  13. [13]

    An Analytical Model for Performance Estimation in Modern High-Capacity IMDD Systems,

    G. Rizzelliet al., “An Analytical Model for Performance Estimation in Modern High-Capacity IMDD Systems,”J. Lightw. Technol., vol. 42, no. 5, pp. 1443–1452, Mar. 2024. JOURNAL OF LIGHTW A VE TECHNOLOGY , VOL. XX, NO. X, FEBRUARY 2026 9

    2024

  14. [14]

    Realistic power amplifier model for the new radio evaluation,

    “Realistic power amplifier model for the new radio evaluation,” 3GPP TSG-RAN WG4 Meeting #79, Tech. Rep., 2016

    2016

  15. [15]

    Optuna: A next-generation hyperparameter optimization framework,

    T. Akibaet al., “Optuna: A next-generation hyperparameter optimization framework,” inProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2019

    2019

  16. [16]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszkeet al., “Pytorch: An imperative style, high-performance deep learning library,” inAdvances in Neural Information Processing Systems

  17. [17]

    8024–8035

    Curran Associates, Inc., 2019, pp. 8024–8035

    2019

  18. [18]

    KAN or MLP: A fairer comparison,

    R. Yu, W. Yu, and X. Wang, “KAN or MLP: A fairer comparison,” inarXiv preprint arXiv:2407.16674, 2024. [Online]. Available: https://arxiv.org/abs/2407.16674

    work page arXiv 2024