pith. sign in

arxiv: 2605.22886 · v1 · pith:VVCZS5TLnew · submitted 2026-05-21 · 💻 cs.IT · cs.LG· cs.NI· math.IT

Resilience Characterization of AI-Native Wireless Receivers via Persistent Homology

Christo Kurisummoottil Thomas , Emilio Calvanese Strinati This is my paper

Pith reviewed 2026-05-25 02:56 UTC · model grok-4.3

classification 💻 cs.IT cs.LGcs.NImath.IT
keywords Topological Resilience Indexpersistent homologyAI-native wireless receiversOFDMdistributional shiftresilience metricchannel adaptation
0
0 comments X

The pith

A topological resilience index detects channel shifts in AI wireless receivers with over one OFDM symbol warning lead.

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

The paper introduces the Topological Resilience Index to quantify how well deep learning wireless receivers maintain performance when channel conditions change, an issue conventional bit error rates do not capture in real time. TRI integrates three dimensions drawn from persistent homology: the topological persistence of loss landscape sublevel sets for model mismatch, geometric drift in channel impulse response distributions, and the spectral gap of the channel manifold normalized by curvature. Theoretical results establish that the index stays bounded, increases monotonically with performance loss, and varies in a Lipschitz-stable manner under Wasserstein perturbations of the channel distribution. Simulations of an OFDM receiver across ten ITU-R environment transitions at multiple shift rates show TRI issuing earlier alerts than gradient-norm or validation-loss baselines, with the gradient-norm method giving no advance notice in any case. The same index then directs burst re-adaptation that cuts post-shift error rates by 80 percent relative to continued operation without adjustment.

Core claim

TRI quantifies the structural stability of a neural network receiver's parameter space during online adaptation to non-stationary channels through three complementary dimensions: validation-loss resilience via topological persistence of loss-landscape sublevel sets, CIR distribution shift tracking geometric drift from a calibration reference, and channel manifold topology given by the spectral gap of the Gaussian kernel matrix normalized by the Olivier-Ricci curvature norm. The index is shown to be bounded, monotonic under performance degradation, and Lipschitz-stable with respect to Wasserstein perturbations of channel distributions.

What carries the argument

Topological Resilience Index (TRI) grounded in persistent homology, combining loss-landscape persistence, CIR distribution shift, and normalized spectral gap of the channel manifold.

If this is right

  • TRI supplies a consistent mean warning lead of more than one OFDM symbol across all ten tested inter-environment transitions at three shift rates.
  • The gradient-norm baseline supplies zero warning lead in every scenario.
  • TRI-guided burst re-adaptation reduces post-shift BER by 80 percent relative to no adaptation within 200 OFDM symbols.
  • TRI remains bounded and monotonic as receiver performance degrades.

Where Pith is reading between the lines

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

  • The same topological construction could monitor parameter-space stability in other online adaptive systems that face distribution shifts.
  • Embedding TRI computation inside receiver hardware might enable proactive rather than reactive adaptation loops.
  • Persistent-homology tools may generalize to robustness analysis for other high-dimensional parameter spaces in communications.

Load-bearing premise

The three dimensions of TRI together capture resilience to distributional shifts in a manner that is monotonic and Lipschitz-stable in Wasserstein distance.

What would settle it

An OFDM receiver simulation across the same ten ITU-R transitions where TRI fails to deliver a mean warning lead exceeding one symbol or where TRI-guided adaptation fails to reduce post-shift BER by 80 percent within 200 symbols.

Figures

Figures reproduced from arXiv: 2605.22886 by Christo Kurisummoottil Thomas, Emilio Calvanese Strinati.

Figure 1
Figure 1. Figure 1: shows TRI and its components during a UMa→RMa transition at SNR= 15 dB, λ = 1.0. During pre-shift steady state (t < t∗ ), TRI≈ 0.92, with all three components near their reference values. Persistence diagrams contain few long-lived H0 features well above the diagonal ( [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Persistence diagrams and cross-scenario TRI. (a) Pre-shift and (b) During shift. D. Effect of Shift Rate and BER Performance [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Post adaptation TRI heatmap. TABLE II TRI WARNING LEAD AND POST-ADAPTATION TRI ACROSS ALL 10 CHANNEL SHIFT SCENARIOS (SNR = 15 DB; λ = 1.0). BER AND GRADIENT-NORM BASELINES ACHIEVE ZERO LEAD IN EVERY SCENARIO. Scenario TRI Lead (sym.) Post-adapt. TRI UMa → RMa 1.3 ± 1.1 0.65 UMi → InF-DH 0.7 ± 1.0 0.65 InH → UMa 1.0 ± 0.8 0.65 RMa → UMi 0.9 ± 1.2 0.65 UMa → InH 1.6 ± 1.4 0.36 UMi → RMa 0.8 ± 0.9 0.65 InF-D… view at source ↗
read the original abstract

AI-native wireless receivers based on deep learning exhibit remarkable performance under stationary channel conditions, yet their resilience to distributional shifts remains poorly characterized by conventional metrics such as bit error rate (BER). To overcome these limitations, this paper proposes a novel real-time metric, the Topological Resilience Index (TRI), grounded in persistent homology and persistence exponents. TRI quantifies the structural stability of a neural network receiver's parameter space during online adaptation to non-stationary channels. Specifically, TRI captures resilience through three complementary dimensions: (i) validation-loss resilience measuring model-channel mismatch, grounded in the topological persistence of loss-landscape sublevel sets; (ii) channel impulse response (CIR) distribution shift, tracking geometric drift of CIR vectors from the calibration reference distribution; and (iii) channel manifold topology, quantified by the spectral gap of the Gaussian kernel matrix normalized by the Olivier-Ricci curvature norm. We establish theoretical guarantees showing that TRI is bounded, monotonic under performance degradation, and Lipschitz-stable with respect to perturbations in channel distributions measured in Wasserstein distance. Simulation results for an OFDM deep-learning receiver adapting across ten ITU-R inter-environment transitions at three shift rates demonstrate that TRI provides a consistent mean warning lead of more than one OFDM symbol over gradient-norm and validation-loss baselines, whereas the gradient-norm baseline achieves zero lead in every scenario. Furthermore, the proposed TRI-guided burst re-adaptation reduces post-shift BER by 80% relative to no adaptation within 200 OFDM symbols.

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 / 2 minor

Summary. The paper introduces the Topological Resilience Index (TRI) for characterizing the resilience of AI-native wireless receivers to distributional shifts in channel conditions. TRI is defined using persistent homology on loss-landscape sublevel sets, CIR distribution shift, and channel manifold topology via spectral gap normalized by Olivier-Ricci curvature. The manuscript claims that TRI is bounded, monotonic under performance degradation, and Lipschitz-stable with respect to Wasserstein distance perturbations in channel distributions. Simulations on an OFDM deep-learning receiver across ten ITU-R transitions at three shift rates show that TRI provides a mean warning lead of more than one OFDM symbol over baselines, with gradient-norm achieving zero lead, and TRI-guided adaptation reduces post-shift BER by 80% within 200 OFDM symbols.

Significance. If the theoretical guarantees hold and the empirical results are reproducible, this work could provide a valuable new tool for monitoring and adapting deep learning based receivers in non-stationary wireless environments, potentially improving reliability in practical deployments. The use of persistent homology for this purpose is novel in the wireless communications context.

major comments (2)
  1. Abstract: the claims that TRI is bounded, monotonic under performance degradation, and Lipschitz-stable w.r.t. Wasserstein distance on channel distributions are asserted without derivation, proof sketch, or reference to any theorem/equation; this is load-bearing because the simulation superiority (warning lead and BER reduction) is attributed to these properties.
  2. The section establishing theoretical guarantees: the Lipschitz-stability of the combined TRI map (persistence of loss sublevel sets + CIR shift + spectral-gap/Olivier-Ricci normalization) is stated as holding but no explicit bound or regularity conditions on the loss landscape or kernel matrix are supplied, leaving the attribution of the >1 OFDM symbol lead unverified.
minor comments (2)
  1. Simulation results lack reported error bars, number of Monte Carlo runs, exclusion criteria for the ten ITU-R transitions, and statistical tests for the mean warning lead and 80% BER reduction claims.
  2. Notation for the three TRI components is introduced in the abstract but would benefit from explicit equations defining each dimension and their aggregation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and for recognizing the potential value of the Topological Resilience Index. We address the two major comments below and will revise the manuscript to strengthen the presentation of the theoretical claims.

read point-by-point responses

  1. Referee: Abstract: the claims that TRI is bounded, monotonic under performance degradation, and Lipschitz-stable w.r.t. Wasserstein distance on channel distributions are asserted without derivation, proof sketch, or reference to any theorem/equation; this is load-bearing because the simulation superiority (warning lead and BER reduction) is attributed to these properties.

    Authors: We agree that the abstract states these properties without explicit references or sketches. In the revised manuscript we will add a concise proof outline in the abstract or a footnote, citing standard results from persistent homology (e.g., stability theorems for persistence diagrams) and Wasserstein distance bounds on sublevel-set filtrations, together with a reference to the relevant section. This will make the attribution of the empirical gains to the stated properties explicit. revision: yes

  2. Referee: The section establishing theoretical guarantees: the Lipschitz-stability of the combined TRI map (persistence of loss sublevel sets + CIR shift + spectral-gap/Olivier-Ricci normalization) is stated as holding but no explicit bound or regularity conditions on the loss landscape or kernel matrix are supplied, leaving the attribution of the >1 OFDM symbol lead unverified.

    Authors: We acknowledge that the current text asserts Lipschitz stability of the composite TRI map without supplying an explicit constant or the required regularity assumptions on the loss function and the Gaussian kernel. In the revision we will expand the theoretical section to state the necessary conditions (Lipschitz continuity of the loss and bounded curvature of the kernel matrix) and derive a concrete stability bound in terms of the Wasserstein distance between channel distributions. This will allow direct verification of the claimed warning lead. revision: yes

Circularity Check

0 steps flagged

No significant circularity in TRI derivation chain.

full rationale

The abstract and provided excerpts define TRI via three explicit dimensions (persistence of loss sublevel sets, CIR distribution shift, spectral-gap normalized by Olivier-Ricci curvature) and assert separate theoretical guarantees of boundedness, monotonicity, and Lipschitz stability in Wasserstein distance. No equations, self-citations, or fitted parameters are exhibited that reduce any claimed guarantee or prediction back to the input data or to a prior self-referential normalization. The simulation results compare TRI against independent baselines (gradient-norm, validation-loss) without evidence that TRI values are constructed by construction from those baselines. The derivation therefore remains self-contained against external topological and geometric measures.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the introduction of the TRI metric and its asserted theoretical properties. No explicit free parameters are named in the abstract. The key axiom is the existence of the stated guarantees. The TRI itself is the primary invented entity.

axioms (1)
  • domain assumption TRI is bounded, monotonic under performance degradation, and Lipschitz-stable with respect to perturbations in channel distributions measured in Wasserstein distance.
    Stated directly in the abstract as established theoretical guarantees without further elaboration.
invented entities (1)
  • Topological Resilience Index (TRI) no independent evidence
    purpose: Quantifies structural stability of neural network receiver parameter space via three topological and geometric dimensions during online adaptation.
    Newly defined composite metric not present in prior literature cited by the abstract.

pith-pipeline@v0.9.0 · 5805 in / 1575 out tokens · 25352 ms · 2026-05-25T02:56:22.554352+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

27 extracted references · 27 canonical work pages

  1. [1]

    An introduction to deep learning for the physical layer,

    T. O’Shea and J. Hoydis, “An introduction to deep learning for the physical layer,”IEEE Transactions on Cognitive Communications and Networking, vol. 3, no. 4, pp. 563–575, dec 2017

    work page 2017

  2. [2]

    IMT-2030 framework recommendation,

    ITU-R, “IMT-2030 framework recommendation,” Tech. Rep. ITU-R M.2160, International Telecommunication Union Radiocommunication Sector (ITU-R), nov 2023

    work page 2030

  3. [3]

    A universal neural receiver that learns at the speed of wireless,

    L. Liu, L. Zheng, Y . Yi, and R. Calderbank, “A universal neural receiver that learns at the speed of wireless,”arXiv preprint arXiv:2602.15458, 2026. 10-1 100 101 Shift rate 6 0 1 2 3 4 5Warning lead (symbols) (a) Warning lead vs. shift rateTRIGrad. norm No adapt. SGD adapt.TRI-guided10-2 10-1 100BER (b) Post-adaptation BER (UMa!!RMa, SNR=15 dB)Pre-shift...

    work page arXiv 2026

  4. [4]

    Comeback kid: Resilience for mixed-critical wireless network resource management,

    R.J. Reifert, S. Roth, A.A. Ahmad, and A. Sezgin, “Comeback kid: Resilience for mixed-critical wireless network resource management,” IEEE Transactions on Vehicular Technology, vol. 72, no. 12, pp. 16177– 16194, 2023

    work page 2023

  5. [5]

    Design and analysis of resilient vehicular platoon systems over wireless networks,

    T. Shui and W. Saad, “Design and analysis of resilient vehicular platoon systems over wireless networks,” inGLOBECOM 2024–2024 IEEE Global Communications Conference, December 2024, pp. 5186–5192

    work page 2024

  6. [6]

    Resilient-by-design: A resilience framework for future wireless networks,

    N.H. Mahmood, S. Samarakoon, P. Porambage, M. Bennis, and M. Latva- Aho, “Resilient-by-design: A resilience framework for future wireless networks,”IEEE Communications Magazine, 2025

    work page 2025

  7. [7]

    Resilient radio access net- works: Ai and the unknown unknowns,

    B. Matthiesen, A. Dekorsy, and P. Popovski, “Resilient radio access net- works: Ai and the unknown unknowns,”arXiv preprint arXiv:2510.21587, 2025

    work page arXiv 2025

  8. [8]

    Deep transfer learning-based downlink channel prediction for FDD massive MIMO systems,

    Y . Yang, F. Gao, Z. Zhong, B. Ai, and A. Alkhateeb, “Deep transfer learning-based downlink channel prediction for FDD massive MIMO systems,”IEEE Transactions on Communications, vol. 68, no. 12, pp. 7485–7497, 2020

    work page 2020

  9. [9]

    Online meta-learning for hybrid model-based deep receivers,

    T. Raviv, S. Park, O. Simeone, Y . C. Eldar, and N. Shlezinger, “Online meta-learning for hybrid model-based deep receivers,”IEEE Transactions on Wireless Communications, vol. 22, no. 10, pp. 6415–6431, 2023

    work page 2023

  10. [10]

    An adaptive and robust deep learning framework for thz ultra-massive mimo channel estimation,

    W. Yu, Y . Shen, H. He, X. Yu, S. Song, J. Zhang, and K.B. Letaief, “An adaptive and robust deep learning framework for thz ultra-massive mimo channel estimation,”IEEE Journal of Selected Topics in Signal Processing, vol. 17, no. 4, pp. 761–776, 2023

    work page 2023

  11. [11]

    On the importance of gradients for detecting distributional shifts in the wild,

    R. Huang, A. Geng, and Y . Li, “On the importance of gradients for detecting distributional shifts in the wild,” inAdvances in Neural Information Processing Systems, 2021, vol. 34, pp. 677–689

    work page 2021

  12. [12]

    Maximum mean discrepancy on exponential windows for online change detection,

    F. Kalinke, M. Heyden, G. Gntuni, E. Fouché, and K. Böhm, “Maximum mean discrepancy on exponential windows for online change detection,” arXiv preprint arXiv:2205.12706, 2022

    work page arXiv 2022

  13. [13]

    Neural network-based cusum for online change-point detection,

    T. Gong, J. Lee, X. Cheng, and Y . Xie, “Neural network-based cusum for online change-point detection,”arXiv preprint arXiv:2210.17312, 2022

    work page arXiv 2022

  14. [14]

    H. K. Khalil,Nonlinear Systems, Prentice Hall, 3rd edition, 2002

    work page 2002

  15. [15]

    Introduction to online convex optimization,

    E. Hazan, “Introduction to online convex optimization,”Foundations and Trends in Optimization, vol. 2, no. 3–4, pp. 157–325, 2016

    work page 2016

  16. [16]

    A theory of learning from different domains,

    S. Ben-David, J. Blitzer, K. Crammer, A. Kulesza, F. Pereira, and J. W. Vaughan, “A theory of learning from different domains,”Machine Learning, vol. 79, no. 1, pp. 151–175, 2010

    work page 2010

  17. [17]

    Visualizing the loss landscape of neural nets,

    H. Li, Z. Xu, G. Taylor, C. Studer, and T. Goldstein, “Visualizing the loss landscape of neural nets,” inAdvances in Neural Information Processing Systems (NeurIPS), Montreal, Canada, dec 2018, pp. 6389–6399

    work page 2018

  18. [18]

    Topological signal processing over simplicial complexes,

    S. Barbarossa and S. Sardellitti, “Topological signal processing over simplicial complexes,”IEEE Transactions on Signal Processing, vol. 68, pp. 2992–3007, 2020

    work page 2020

  19. [19]

    Persistent homology—a survey,

    H. Edelsbrunner and J. Harer, “Persistent homology—a survey,” inCon- temporary Mathematics, vol. 453, pp. 257–282. American Mathematical Society, 2008

    work page 2008

  20. [20]

    Edelsbrunner and J

    H. Edelsbrunner and J. Harer,Computational Topology: An Introduction, American Mathematical Society, Providence, RI, 2010

    work page 2010

  21. [21]

    Study on channel model for frequencies from 0.5 to 100 GHz,

    3GPP, “Study on channel model for frequencies from 0.5 to 100 GHz,” Technical Report TR 38.901, 3rd Generation Partnership Project, June 2022

    work page 2022

  22. [22]

    Deep learning based communication over the air,

    S. Dörner, S. Cammerer, J. Hoydis, and S. ten Brink, “Deep learning based communication over the air,”IEEE Journal of Selected Topics in Signal Processing, vol. 12, no. 1, pp. 132–143, feb 2018

    work page 2018

  23. [23]

    Guidelines for evaluation of radio interface technologies for imt-2020,

    M. Series, “Guidelines for evaluation of radio interface technologies for imt-2020,” Tech. Rep. Report ITU-R M.2512, International Telecommunication Union (ITU), 2017

    work page 2020

  24. [24]

    Chazal, V

    F. Chazal, V . de Silva, M. Glisse, and S. Oudot,The Structure and Stability of Persistence Modules, Springer, Cham, 2016

    work page 2016

  25. [25]

    Ricci curvature of markov chains on metric spaces,

    Y . Ollivier, “Ricci curvature of markov chains on metric spaces,”Journal of Functional Analysis, vol. 256, no. 3, pp. 810–864, 2009

    work page 2009

  26. [26]

    Gudhi user and reference manual,

    The GUDHI Project, “Gudhi user and reference manual,” https://gudhi. inria.fr/, 2024

    work page 2024

  27. [27]

    Sionna: An open-source library for next-generation physical layer research,

    J. Hoydis, F. A. Aoudia, A. Valcarce, and H. Viswanathan, “Sionna: An open-source library for next-generation physical layer research,”arXiv preprint arXiv:2203.11854, 2022

    work page arXiv 2022