Q-FE: A Quantum-Native 6G Far-Edge Architecture Securing Industrial IoT Digital Twins via CSIDH-PQC and Asynchronous Federated Learning

Vincenzo Sammartino

arxiv: 2606.03611 · v1 · pith:4Y746IGUnew · submitted 2026-06-02 · 💻 cs.CR · cs.ET

Q-FE: A Quantum-Native 6G Far-Edge Architecture Securing Industrial IoT Digital Twins via CSIDH-PQC and Asynchronous Federated Learning

Vincenzo Sammartino This is my paper

Pith reviewed 2026-06-28 09:23 UTC · model grok-4.3

classification 💻 cs.CR cs.ET

keywords 6G networkspost-quantum cryptographydigital twinsfederated learningindustrial IoTCSIDHfar-edge computingURLLC

0 comments

The pith

Q-FE places micro-digital twins at 6G base stations, embeds CSIDH-512 keys in MAC frames, and runs asynchronous federated learning to secure industrial IoT digital twins with low overhead and quantum resistance.

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

The paper sets out a far-edge architecture that keeps digital-twin computation close to radio access points instead of routing it through distant clouds or MEC servers. It embeds short post-quantum keys directly into radio control frames and replaces synchronous model aggregation with a contract-governed asynchronous process. A sympathetic reader would care because the approach claims to satisfy sub-millisecond reliability targets and long-term cryptographic security at the same time on resource-limited industrial devices.

Core claim

Q-FE reduces MAC-layer overhead by 62% versus ML-KEM/Kyber-1024, holds P99.9 URLLC latency at 0.78 ms, speeds global-model convergence by 31% over synchronous federated learning, and resists quantum eavesdropping, model poisoning, and Sybil attacks through co-located micro-digital twins, cross-layer CSIDH-512 key exchange, and DAG smart-contract asynchronous aggregation.

What carries the argument

The cross-layer post-quantum key exchange module that places compact CSIDH-512 isogeny keys inside MAC control frames, paired with an asynchronous federated learning protocol run under lightweight DAG smart contracts at the edge.

If this is right

MAC-layer overhead falls 62% compared with ML-KEM while URLLC P99.9 latency remains 0.78 ms.
Global model convergence accelerates 31% relative to synchronous federated learning.
Micro-digital-twin handover migration finishes in 1.9 ms across 10,000 simulated events.
Each aggregation round runs in O(N log R) time.
The architecture blocks quantum eavesdropping, model poisoning, and Sybil attacks under the stated formal threat model.

Where Pith is reading between the lines

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

The same key-embedding technique might reduce radio load for other latency-critical wireless applications facing quantum threats.
Widespread use could move digital-twin workloads permanently from central clouds to radio-edge nodes.
Hardware prototypes would be required to confirm that the reported latency and overhead numbers survive outside simulation.
The approach may support larger fleets of mobile industrial robots without increasing spectrum demand.

Load-bearing premise

The NS-3 plus PySyft simulations accurately capture real 6G far-edge radio conditions, device constraints, and attack behaviors, and that CSIDH-512 keys fit inside MAC frames without fragmentation.

What would settle it

A physical testbed measurement showing that CSIDH-512 keys cause MAC-frame fragmentation or that P99.9 latency exceeds 0.78 ms under realistic 6G conditions would falsify the performance claims.

Figures

Figures reproduced from arXiv: 2606.03611 by Vincenzo Sammartino.

**Figure 2.** Figure 2: MAC-layer PQC overhead analysis (6G NR, µ=3, Tf =125 µs). (a) Public-key sizes (log scale): only CSIDH-512 (64 B) falls within the 256 B MAC CE budget. (b) Additional TTIs from fragmentation: Kyber-512 requires 3 extra TTIs (+375 µs), Kyber-1024 requires 6 (+750 µs); CSIDH-512 incurs zero. (c) Control-plane latency CDF vs. 1 ms URLLC budget: Q-FE achieves P99.9=0.78 ms; Kyber-1024 exceeds budget at P99.9=1… view at source ↗

**Figure 3.** Figure 3: AFL convergence analysis (N=100 nodes, 10% Byzantine, SWAT dataset, ε≈1.0, δ=10−5 ). (a) Accuracy vs. wall-clock time: Q-FE AFL+DAG-SC converges 31% faster than synchronous FedAvg. (b) Accuracy vs. aggregation round: Q-FE achieves superior sample efficiency versus async FedAvg without SC. (c) DAG-SC Byzantine filter: the Z-score mechanism rejects adversarial updates (red) while accepting honest contributio… view at source ↗

**Figure 4.** Figure 4: Security and energy analysis. (a) Classical vs. quantum security levels: RSA-2048 and ECDH are broken by Shor; CSIDH-512 retains 62-bit postquantum security (above NIST Cat. 1). (b) Model deviation bound vs. Byzantine fraction: the DAG-SC Z-score filter reduces effective Byzantine count by ≈90%, keeping deviation below the DP noise floor for f /N ≤ 0.25. (c) CSIDH-AKR power overhead vs. Tr (log-log): at T… view at source ↗

read the original abstract

Sixth-generation (6G) wireless networks will underpin ultra-dense Industrial IoT (IIoT) ecosystems in which resource-constrained Far-Edge devices -- autonomous mobile robots, industrial actuators, connected vehicles -- must simultaneously satisfy sub-millisecond latency, $10^{-7}$-class reliability, and decades-long cryptographic security. Current architectures delegate Digital Twin (DT) computation to centralised cloud or Mobile Edge Computing (MEC) servers, incurring prohibitive round-trip latency, and rely on classical public-key cryptography vulnerable to quantum attacks under the harvest-now, decrypt-later (HNDL) threat model. We propose Q-FE, a Quantum-Native 6G Far-Edge architecture integrating three co-designed components: (i) Micro-Digital Twins ($\mu$DTs) co-located with 6G base stations and high-capability endpoints; (ii) a Cross-Layer Post-Quantum Key Exchange module embedding CSIDH-512 isogeny key material directly within MAC-layer control frames, exploiting the scheme's uniquely compact keys ($\le 64$ bytes) to avoid packet fragmentation; and (iii) an Asynchronous Federated Learning (AFL) protocol governed by lightweight DAG smart contracts at MEC nodes, eliminating straggler bottlenecks and preventing model-poisoning and Sybil attacks without exposing raw data. End-to-end simulations (NS-3 + PySyft) demonstrate that Q-FE reduces MAC-layer overhead by 62% versus ML-KEM/Kyber-1024, maintains P99.9 URLLC latency at 0.78 ms, and accelerates global-model convergence by 31% over synchronous Federated Learning. Protocol complexity analysis confirms $O(N \log R)$ per aggregation round, and $\mu$DT handover migration completes in $1.9 \pm 0.3$ ms across $10^4$ simulated events. A formal threat model confirms resilience against quantum eavesdropping, model-poisoning, and Sybil attacks.

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.

Desk Editor's Note private letter to a colleague

Q-FE is a straightforward integration of CSIDH key embedding, micro digital twins, and async FL for 6G IIoT, but the headline performance numbers rest only on unvalidated NS-3/PySyft runs.

read the letter

The paper's core move is to co-locate small digital twins at the far edge, drop compact CSIDH-512 keys straight into MAC frames to skip fragmentation, and run asynchronous federated learning under DAG contracts to cut stragglers and some attacks. That combination is new as a single 6G proposal, and the choice of CSIDH for its small keys is a reasonable engineering fit for the latency constraints.

The simulations claim a 62% MAC overhead drop versus Kyber, 0.78 ms P99.9 latency, and 31% faster convergence. Those numbers are the main selling point, yet the abstract and stress-test note give no channel models, mobility traces, attack injection rates, or hardware cross-checks. Without those, the metrics stay simulator-dependent and hard to judge.

The threat model is described as formal but supplies no proof sketch or attack surface breakdown in the provided text. Protocol complexity is stated as O(N log R), which is plausible for the DAG setup but again unexpanded.

This is the kind of architecture paper that 6G security groups and industrial IoT teams might want to see, mainly for the concrete component choices rather than any new math. It is coherent on its own terms and engages the right prior work on isogeny crypto and async FL, so it clears the bar for a serious referee even though the simulation evidence needs substantial expansion before the numbers can be taken as transferable.

Referee Report

2 major / 1 minor

Summary. The paper proposes Q-FE, a quantum-native 6G far-edge architecture for Industrial IoT digital twins that co-locates micro-digital twins (μDTs) with base stations, embeds compact CSIDH-512 keys directly in MAC control frames to avoid fragmentation, and uses asynchronous federated learning governed by DAG smart contracts at MEC nodes. End-to-end NS-3 + PySyft simulations are reported to show 62% MAC-layer overhead reduction versus ML-KEM/Kyber-1024, P99.9 URLLC latency of 0.78 ms, 31% faster global-model convergence than synchronous FL, O(N log R) aggregation complexity, 1.9 ± 0.3 ms μDT handover, and resilience to quantum eavesdropping, model-poisoning, and Sybil attacks under a formal threat model.

Significance. If the simulation methodology and results prove robust, the work would offer a concrete cross-layer design for embedding post-quantum cryptography in 6G MAC frames while preserving URLLC constraints and adding asynchronous FL security primitives; the use of CSIDH's small key size is a targeted strength that could influence future 6G PQC integration studies.

major comments (2)

[Abstract] Abstract (and the simulation campaign it summarizes): the headline quantitative claims—62% MAC overhead reduction, 0.78 ms P99.9 latency, and 31% convergence acceleration—are produced exclusively by NS-3 + PySyft runs; no hardware measurements, no comparison against published 6G IIoT channel traces, and no sensitivity analysis to numerology, mobility, or attack-injection rates are referenced, so the transferability of these metrics cannot be assessed from the presented evidence.
[Abstract / threat-model discussion] The formal threat model is invoked to confirm resilience against quantum eavesdropping, model-poisoning, and Sybil attacks, yet the manuscript supplies no description of the model’s assumptions, adversary capabilities, or verification method; without this, the security claims remain ungrounded relative to the performance numbers.

minor comments (1)

The invented term “Micro-Digital Twins (μDTs)” is introduced without a crisp definition distinguishing it from standard edge DT placements; a short clarifying paragraph would help readers.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments on the simulation methodology and threat-model presentation. We address each major comment below, indicating where revisions will be incorporated.

read point-by-point responses

Referee: [Abstract] Abstract (and the simulation campaign it summarizes): the headline quantitative claims—62% MAC overhead reduction, 0.78 ms P99.9 latency, and 31% convergence acceleration—are produced exclusively by NS-3 + PySyft runs; no hardware measurements, no comparison against published 6G IIoT channel traces, and no sensitivity analysis to numerology, mobility, or attack-injection rates are referenced, so the transferability of these metrics cannot be assessed from the presented evidence.

Authors: The reported metrics are obtained exclusively from NS-3 + PySyft discrete-event simulations, consistent with the paper's focus on a cross-layer architectural proposal evaluated under reproducible 6G URLLC and IIoT conditions. We agree that explicit discussion of transferability would strengthen the work. In revision we will add a dedicated subsection on simulation assumptions (calibrated to 3GPP TR 38.901 and related 6G studies), parameter ranges, and sensitivity analysis covering numerology, mobility, and attack-injection rates. Hardware measurements and direct comparisons to published proprietary channel traces lie outside the scope of this simulation study. revision: partial
Referee: [Abstract / threat-model discussion] The formal threat model is invoked to confirm resilience against quantum eavesdropping, model-poisoning, and Sybil attacks, yet the manuscript supplies no description of the model’s assumptions, adversary capabilities, or verification method; without this, the security claims remain ungrounded relative to the performance numbers.

Authors: We agree that the threat-model description must be made explicit. Although a dedicated threat-model section exists in the full manuscript, its assumptions, adversary capabilities (quantum adversary under the harvest-now-decrypt-later model; malicious FL clients performing poisoning or Sybil attacks), and verification method (game-based proofs for CSIDH-PQC and formal analysis of the DAG-governed AFL protocol) are not sufficiently linked to the abstract claims. We will revise the manuscript to include a concise description of these elements in both the abstract discussion and the main security section. revision: yes

standing simulated objections not resolved

Hardware measurements and direct comparisons against published 6G IIoT channel traces, as the evaluation is simulation-based.

Circularity Check

0 steps flagged

No circularity; claims are direct simulation outputs with no internal derivation chain

full rationale

The provided abstract and description contain no equations, fitted parameters, or derivation steps. All headline metrics (62% overhead reduction, 0.78 ms latency, 31% convergence speedup) are stated as direct results of NS-3 + PySyft simulations rather than quantities computed from other quantities inside the paper. No self-definitional relations, fitted-input predictions, or load-bearing self-citations appear. The architecture is presented as a proposal whose performance is externally validated by simulation, leaving the derivation chain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

Only the abstract is available, so the ledger is limited to assumptions explicitly invoked there; the proposal depends on domain assumptions about CSIDH security and simulation fidelity rather than new free parameters or invented entities with independent evidence.

axioms (2)

domain assumption CSIDH-512 supplies keys compact enough (≤64 bytes) to embed in MAC-layer control frames without fragmentation while retaining post-quantum security
Stated in the description of the Cross-Layer Post-Quantum Key Exchange module.
domain assumption Asynchronous federated learning governed by DAG smart contracts prevents model-poisoning and Sybil attacks without exposing raw data
Invoked in the AFL protocol component.

invented entities (1)

Micro-Digital Twins (μDTs) no independent evidence
purpose: Co-located digital-twin instances that reduce round-trip latency for IIoT DT computation
Introduced as a core architectural component; no independent evidence or falsifiable prediction supplied beyond the proposal itself.

pith-pipeline@v0.9.1-grok · 5908 in / 1762 out tokens · 44735 ms · 2026-06-28T09:23:13.793496+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references

[1]

A vision of the 6G wireless systems: Applications, enabling technologies, and design aspects,

W. Saad, M. Bennis, and M. Chen, “A vision of the 6G wireless systems: Applications, enabling technologies, and design aspects,”IEEE Netw., vol. 34, no. 3, pp. 134–142, 2020

2020
[2]

Digital-twin-enabled 6G: Vision, ar- chitectural trends, and future directions,

L. U. Khan, W. Saad, Z. Hanet al., “Digital-twin-enabled 6G: Vision, ar- chitectural trends, and future directions,”IEEE Commun. Mag., vol. 60, no. 1, pp. 74–80, 2022

2022
[3]

Federated learning: Challenges, methods, and future directions,

T. Li, A. K. Sahu, A. Talwalkar, and V . Smith, “Federated learning: Challenges, methods, and future directions,”IEEE Signal Process. Mag., vol. 37, no. 3, pp. 50–60, 2020

2020
[4]

Cybersecurity in an era with quantum computers: Will we be ready?

M. Mosca, “Cybersecurity in an era with quantum computers: Will we be ready?”IEEE Secur. Priv., vol. 16, no. 5, pp. 38–41, 2018

2018
[5]

Module-lattice-based key-encapsulation mechanism standard,

National Institute of Standards and Technology, “Module-lattice-based key-encapsulation mechanism standard,” NIST, Tech. Rep. FIPS 203, 2024

2024
[6]

Post-quantum authentication in TLS 1.3: A performance study,

D. Sikeridis, P. Kampanakis, and M. Devetsikiotis, “Post-quantum authentication in TLS 1.3: A performance study,” inProc. ISOC NDSS, 2020

2020
[7]

CSIDH: An efficient post-quantum commutative group action,

W. Castryck, T. Lange, C. Martindale, L. Panny, and J. Renes, “CSIDH: An efficient post-quantum commutative group action,” inProc. ASI- ACRYPT, ser. Lecture Notes Comput. Sci., vol. 11274. Springer, 2018, pp. 395–427

2018
[8]

Machine learning with adversaries: Byzantine tolerant gradient descent,

P. Blanchard, E. M. E. Mhamdi, R. Guerraoui, and J. Stainer, “Machine learning with adversaries: Byzantine tolerant gradient descent,” inProc. NeurIPS, 2017, pp. 119–129

2017
[9]

NotLine: A non-intrusive automated platform to build a security digital twin,

F. Baiardi, V . Sammartino, and S. Ruggieri, “NotLine: A non-intrusive automated platform to build a security digital twin,” inProc. 28th IEEE/ACM Int. Symp. Distrib. Simul. Real-Time Appl. (DS-RT), 2025, pp. 1–8

2025
[10]

CTIDH: Faster constant-time CSIDH,

D. Cervantes-Vázquez, M. Chenu, J.-J. Chi-Domínguez, L. D. Feo, F. Rodríguez-Henríquez, and B. Smith, “CTIDH: Faster constant-time CSIDH,”IACR Trans. Cryptogr. Hardw. Embed. Syst., vol. 2021, no. 4, pp. 351–387, 2021

2021
[11]

An efficient key recovery attack on SIDH,

W. Castryck and T. Decru, “An efficient key recovery attack on SIDH,” inProc. EUROCRYPT, 2023

2023
[12]

Communication-efficient learning of deep networks from decentralized data,

B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” inProc. AISTATS, 2017, pp. 1273–1282

2017
[13]

Differentially private federated learning: A client level perspective,

R. C. Geyer, T. Klein, and M. Nabi, “Differentially private federated learning: A client level perspective,” inWorkshop on Privacy-Preserving Machine Learning (NeurIPS), 2017

2017
[14]

The tangle,

S. Popov, “The tangle,” IOTA Foundation White Paper, Tech. Rep., 2018, version 1.4.3

2018
[15]

Asynchronous decentralized parallel stochastic gradient descent,

X. Lian, C. Zhang, H. Zhanget al., “Asynchronous decentralized parallel stochastic gradient descent,” inProc. ICML, 2018. Vincenzo Sammartinois pursuing the National Ph.D. in Artificial Intelligence at the Università di Pisa, Italy, and is a Visiting Ph.D. Student at KAUST, Saudi Arabia, contributing to the Resilient- Guard project on decentralised TinyML...

2018

[1] [1]

A vision of the 6G wireless systems: Applications, enabling technologies, and design aspects,

W. Saad, M. Bennis, and M. Chen, “A vision of the 6G wireless systems: Applications, enabling technologies, and design aspects,”IEEE Netw., vol. 34, no. 3, pp. 134–142, 2020

2020

[2] [2]

Digital-twin-enabled 6G: Vision, ar- chitectural trends, and future directions,

L. U. Khan, W. Saad, Z. Hanet al., “Digital-twin-enabled 6G: Vision, ar- chitectural trends, and future directions,”IEEE Commun. Mag., vol. 60, no. 1, pp. 74–80, 2022

2022

[3] [3]

Federated learning: Challenges, methods, and future directions,

T. Li, A. K. Sahu, A. Talwalkar, and V . Smith, “Federated learning: Challenges, methods, and future directions,”IEEE Signal Process. Mag., vol. 37, no. 3, pp. 50–60, 2020

2020

[4] [4]

Cybersecurity in an era with quantum computers: Will we be ready?

M. Mosca, “Cybersecurity in an era with quantum computers: Will we be ready?”IEEE Secur. Priv., vol. 16, no. 5, pp. 38–41, 2018

2018

[5] [5]

Module-lattice-based key-encapsulation mechanism standard,

National Institute of Standards and Technology, “Module-lattice-based key-encapsulation mechanism standard,” NIST, Tech. Rep. FIPS 203, 2024

2024

[6] [6]

Post-quantum authentication in TLS 1.3: A performance study,

D. Sikeridis, P. Kampanakis, and M. Devetsikiotis, “Post-quantum authentication in TLS 1.3: A performance study,” inProc. ISOC NDSS, 2020

2020

[7] [7]

CSIDH: An efficient post-quantum commutative group action,

W. Castryck, T. Lange, C. Martindale, L. Panny, and J. Renes, “CSIDH: An efficient post-quantum commutative group action,” inProc. ASI- ACRYPT, ser. Lecture Notes Comput. Sci., vol. 11274. Springer, 2018, pp. 395–427

2018

[8] [8]

Machine learning with adversaries: Byzantine tolerant gradient descent,

P. Blanchard, E. M. E. Mhamdi, R. Guerraoui, and J. Stainer, “Machine learning with adversaries: Byzantine tolerant gradient descent,” inProc. NeurIPS, 2017, pp. 119–129

2017

[9] [9]

NotLine: A non-intrusive automated platform to build a security digital twin,

F. Baiardi, V . Sammartino, and S. Ruggieri, “NotLine: A non-intrusive automated platform to build a security digital twin,” inProc. 28th IEEE/ACM Int. Symp. Distrib. Simul. Real-Time Appl. (DS-RT), 2025, pp. 1–8

2025

[10] [10]

CTIDH: Faster constant-time CSIDH,

D. Cervantes-Vázquez, M. Chenu, J.-J. Chi-Domínguez, L. D. Feo, F. Rodríguez-Henríquez, and B. Smith, “CTIDH: Faster constant-time CSIDH,”IACR Trans. Cryptogr. Hardw. Embed. Syst., vol. 2021, no. 4, pp. 351–387, 2021

2021

[11] [11]

An efficient key recovery attack on SIDH,

W. Castryck and T. Decru, “An efficient key recovery attack on SIDH,” inProc. EUROCRYPT, 2023

2023

[12] [12]

Communication-efficient learning of deep networks from decentralized data,

B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” inProc. AISTATS, 2017, pp. 1273–1282

2017

[13] [13]

Differentially private federated learning: A client level perspective,

R. C. Geyer, T. Klein, and M. Nabi, “Differentially private federated learning: A client level perspective,” inWorkshop on Privacy-Preserving Machine Learning (NeurIPS), 2017

2017

[14] [14]

The tangle,

S. Popov, “The tangle,” IOTA Foundation White Paper, Tech. Rep., 2018, version 1.4.3

2018

[15] [15]

Asynchronous decentralized parallel stochastic gradient descent,

X. Lian, C. Zhang, H. Zhanget al., “Asynchronous decentralized parallel stochastic gradient descent,” inProc. ICML, 2018. Vincenzo Sammartinois pursuing the National Ph.D. in Artificial Intelligence at the Università di Pisa, Italy, and is a Visiting Ph.D. Student at KAUST, Saudi Arabia, contributing to the Resilient- Guard project on decentralised TinyML...

2018