PaperlayerView exact revision on arXiv

Source-linked AI summary

Securing Edge & Physical AI: Post-Quantum IoT Blockchain for Mission-Critical Cyber-Physical Agents

Kürşat Çakal, Suat Özdemir

arXiv:2610.33862v1cs.CRcs.ETcs.RO

TL;DR

Quantum-capable attacks threaten public-key mechanisms used in CPS and Edge AI, while migration to PQC carries costs for constrained systems. The paper presents PARS key management and an optional ledger for shared security state; in the evaluated configurations, consensus completed in under 91.05 seconds, subject to stated limitations and assumptions.

  • Problem

    Quantum-computing progress and post-quantum migration requirements create a security challenge for mission-critical systems using vulnerable cryptographic mechanisms.

    Source: §I · p. 3

  • Method

    PARS coordinates dynamic key reconciliation for edge and IoT instruments, using fresh mutual key chains and conveyed algorithm identifiers.

    Source: B. Private Authentication for… · p. 21, §VI · p. 41

  • Results

    Under 91.05 seconds: every evaluated combination of PQC algorithm, node-set size, and ZKP hardness completed consensus.

    Source: §VI · p. 42

  • Takeaways & Limitations

    The reported measurements characterize PARS-ESL feasibility across consensus, communication, memory, and failure metrics in the evaluated settings.

    Source: §VI · pp. 42–43

    QHT-KA cannot provide an unconditional time lock: early reveal remains possible for an authorized or sufficiently large colluding set.

    Source: 7) Conditional security claim and… · p. 29

Abstract

from arXiv · show

Distributed Cyber-Physical Systems (CPS) and Edge AI networks increasingly depend on public-key authentication and key-establishment mechanisms that are vulnerable to cryptographically relevant Quantum Computers (QC). Shor's algorithm threatens widely deployed asymmetric primitives, while Grover's algorithm reduces the effective security margins of symmetric and hash-based mechanisms. Although Post-Quantum Cryptography (PQC) provides a migration path, its larger public keys, ciphertexts, and signatures introduce additional computation, memory, communication, storage, and packet-fragmentation costs that can constrain dynamic multi-agent systems. In this article, we introduce rotational secrecy, an architectural objective that limits temporal key exposure through authenticated transitions between cryptographic epochs. We present Private Authentication for Rotational Secrecy (PARS), a reference mechanism that coordinates policy-triggered key establishment, rotation, and reconciliation among authorized Physical AI agents. For deployments that need distributed agreement on security state, we further develop the PARS-Enabled Edge Security Ledger (PARS-ESL), an optional permissioned architecture maintained by authorized validators or capable edge gateways. PARS-ESL does not store high-volume operational transactions. Instead, it records authenticated state for key epochs, group membership, revocation, algorithm-suite policies, challenge validation, and recovery checkpoints. As a result, resource-constrained agents can take part as lightweight clients, and low-latency operational traffic stays outside the ledger consensus path, protected by the currently active symmetric epoch keys. Our results show that ledger-assisted post-quantum key-epoch management is feasible for selected Edge AI and CPS security-control workloads under the stated trust, compromise, connectivity, and resource assumptions.

I. INTRODUCTION

Quantum-driven threats and the resource costs of post-quantum cryptography make security migration a cross-layer challenge for Edge AI and CPS. The paper presents PARS for authenticated key-epoch evolution and an optional ledger for coordinating security state, while bounding their suitability by deployment assumptions.

I. INTRODUCTION

  • The optional permissioned ESL coordinates shared epochs, membership, revocation, algorithm policy, and validation state without recording high-rate application traffic.Its suitability depends on trust, audit requirements, connectivity, validator assumptions, and resource budgets.Source: §I · pp. 3, 5
  • Post-quantum migration is urgent as quantum threats advance, while larger PQC signatures can fragment packets and strain storage on embedded nodes.The article also cites finalized NIST standards and national-security migration requirements as concrete drivers for action.Source: §I · pp. 1, 3, 4
  • Prior work leaves gaps in limiting application-history growth, distinguishing key replacement from secrecy guarantees, supporting constrained endpoints, and accounting for endpoint quantum-hardware requirements.The article identifies these as four central limitations in surveyed post-quantum blockchain work for CPS security.Source: §I · p. 4
  • Rotational secrecy limits temporal key exposure through authenticated state transitions, while PARS coordinates time-, action-, and group-triggered key updates among authorized peers.The intended exposure reduction depends on fresh secret contributions, secure erasure, replay protection, authenticated transitions, and consistent membership management; rotation alone does not guarantee forward secrecy or post-compromise recovery.Source: §I · p. 4
  • The paper combines quantum-threat assessment, PQC selection, hybrid KEM–KDF–AEAD construction, authentication, key evolution, distributed trust, and migration planning in a cross-layer methodology.It also maps adversarial capabilities to affected primitives, communication functions, system layers, and possible CPS consequences.Source: §I · p. 4
  • The article examines adaptive challenge complexity and verification cost separately from proof soundness, consensus behavior, and quantum-security assumptions.It presents PARS and ESL as reusable references under constrained computation, communication, and trust conditions, not universally required solutions.Source: §I · p. 5

II. BACKGROUND OVERVIEW

The background section frames its technical review around the requirements that shape authenticated key establishment, key evolution, and distributed security-state coordination. It describes embodied edge agents’ operational and resource constraints, their expanded attack surface, and the role and limits of layered cryptographic protections.

II. BACKGROUND OVERVIEW

  • The background reviews edge-agent requirements, distributed ledgers, quantum threats, and PQC to inform key establishment, key-epoch evolution, packet size, resource use, and security-state coordination.It is a targeted technical bridge, not an exhaustive treatment of those fields in isolation.Source: §II · p. 5

A. Edge AI Agents and Intelligent Cyber-Physical Systems

  • Embodied edge agents combine sensing, local inference, communication, decisions, and actuation, so security must protect information authenticity, freshness, and consistency as well as confidentiality.Their operation near physical sources supports local decision-making without high-latency cloud dependence, including under intermittent or degraded connectivity.Source: A. Edge AI Agents and Intelligent… · p. 5
  • Edge deployments face interception, replay, impersonation, network disruption, sensor manipulation, and physical threats, while quantum computers add risk to classical public-key mechanisms.PARS addresses authenticated key-epoch evolution and an optional ESL coordinates shared security state; neither guarantees absolute operational security, and effectiveness depends on trusted provisioning, implementation, randomness, erasure, bounded validator compromise, endpoint integrity, and complementary protections.Source: A. Edge AI Agents and Intelligent… · pp. 7–8

B. Blockchain

Blockchains provide replicated, auditable state, but do not inherently provide confidentiality or access control. The section introduces quantum-computing principles and post-quantum cryptography, emphasizing how their roles and costs shape edge-system design.

B. Blockchain

  • A blockchain records ordered state transitions in a replicated ledger; hashes and Merkle commitments provide tamper evidence, signatures authenticate submissions, and consensus validates and finalizes updates.Ledger integrity depends on cryptographic assumptions, consensus rules, validator thresholds, network models and implementation; auditability alone does not provide confidentiality or access control.Source: B. Blockchain · pp. 8–9
  • The optional ESL maintains low-rate security state for key epochs, membership, revocation, cryptographic policies and validation, while operational traffic remains outside ledger consensus.Replication is justified when authorized participants need consistent, auditable state without continuous dependence on one online authority; centralized management or a replicated directory may be preferable otherwise.Source: B. Blockchain · pp. 8–9

C. Quantum Computers

  • Quantum computers use qubits, superposition, entanglement, gates and interference; the section describes these principles as enabling specialized computations with implications for cybersecurity.The text notes that large-scale, fault-tolerant quantum computers remain under development.Source: C. Quantum Computers · pp. 9–10

D. Post-Quantum Cryptography

  • PQC families include lattice-, code-, hash-, multivariate- and historically isogeny-based constructions, while KEMs, signatures, KDFs and symmetric AEAD serve distinct security functions.Hybrid systems can increase handshake and certificate size, computation, memory use and fragmentation; the design question is which standardized maturity, object size, execution time, memory, assurance and hardware support suit a platform.Source: D. Post-Quantum Cryptography · pp. 10, 12

III. RELATED WORK: QUANTUM-SAFE EDGE SYSTEMS

The review identifies gaps between quantum-safe proposals and the operational needs of physical Edge AI and CPS, then positions PARS and an optional ESL as a response. Its prototype results support feasibility only for the evaluated configuration and stated deployment scope.

III. RELATED WORK: QUANTUM-SAFE EDGE SYSTEMS

  • Quantum-mechanics-based proposals range from GHZ-state ledger designs to experimentally investigated QKD-supported consensus, but specialized infrastructure limits near-term edge applicability.The review also notes concerns including infrastructure cost, denial-of-service exposure and incomplete end-to-end attack analysis.Source: §III · pp. 12–13
  • Software-oriented work includes PQC ledgers and lightweight IoT systems, but cited designs leave gaps in edge resource analysis, current standardized parameters, quantum protection or key rotation.The review also describes swarm systems that rely on quantum-vulnerable signatures and lack dynamic key rotation.Source: §III · pp. 13–14
  • Prior work often isolates individual components, omits operational costs, lacks authenticated cryptographic-epoch management, or leaves adversary and compromise assumptions underspecified.Reported omissions include fragmentation, retransmission, energy, persistent storage, failure classification and confidence intervals.Source: §III · pp. 14–15
  • The study responds with rotational secrecy, PARS for time-, action- and group-based key evolution, and an optional ESL for distributed, auditable security-state coordination.The ESL covers key epochs, membership, revocation, algorithm policy, validation outcomes and recovery state.Source: §III · p. 16
  • PARS-ESL is intended for authorized multi-agent deployments needing shared security-state agreement under intermittent connectivity without continuous reliance on one online authority.It is not presented as a universal replacement for centralized key management, PKI, replicated directories or conventional BFT state machines.Source: §III · p. 16
  • The prototype was evaluated with up to 16 nodes, including an 18-hour run with reported peak memory below 3.5 MB.The authors caution that these measurements do not establish suitability for all resource-constrained CPS platforms.Source: §III · p. 16

IV. PROPOSED METHOD

The proposed method frames quantum-resistant key management as a systems problem for space and defense CPS, using a blockchain-based rotational approach. It also outlines attack scenarios in which quantum-capable attackers manipulate swarm communications and control flows.

IV. PROPOSED METHOD

  • PARS-ESL is presented as an optional ledger for distributed key and policy state, with constrained agents able to participate without executing full consensus.The method positions space and defense systems as a demanding representative setting, not as the only application surface.Source: §IV · pp. 16–17
  • The architecture separates high-rate operational traffic from security-state management, avoiding direct storage of extensive data streams on a replicated ledger.It assigns traffic to a symmetric-encryption data plane and identity, key rotation, membership, revocation, and security-state agreement to the security-control plane.Source: §IV · p. 17
  • The method motivates a blockchain-based rotational approach as a feasible, cost-effective response to quantum-vulnerable key exchange and the equipment and network costs of QKD.The passage also notes QKD denial-of-service risks, citing NSA.Source: §IV · p. 17
  • The proposed framework identifies these scenarios as operational risks to autonomous space and defense systems, while noting that published information does not yet show quantum computers producing suspicious deterministic outputs.The passage calls for security-level updates to begin for these critical systems and says the scenarios can affect strategic operational flow.Source: §IV · p. 17

A. Attack Mitigation Cases and Cyber Attack Surface

  • The attack scenarios describe quantum-capable attackers altering swarm cooperation, telemetry, command-and-control, system identity, or line-of-sight exchanges.Examples include steering a missile toward allied forces, falsifying apparent command status, spoofing node identity, and diverting controller actions.Source: A. Attack Mitigation Cases and Cyber… · pp. 18–19

B. Private Authentication for Rotational Secrecy (PARS)

PARS renews peer key chains through crypto-agile setup and authenticated agreement on key state, adapting renewal thresholds to security and link conditions. It activates reconciled keys after peer approval and match signals, supporting distributed operation with Light Nodes and encrypted, ordered executive data while keeping operational message exchange outside consensus progress.

B. Private Authentication for Rotational Secrecy (PARS)

  • PARS triggers rotation using time-, anomaly/action-, or group-based thresholds, while PSTC adapts thresholds to threat assessments and communication-link conditions.The latency penalty is intended to curb fragmentation and network collapse on bandwidth-constrained links during post-quantum key exchanges.Source: B. Private Authentication for… · pp. 20–21
  • PARS offers two interoperable CNRME modes that share consensus and ledger flow but differ in key-setup material and hardness assumptions.The protocol selects between them through crypto-agility parameters, and both broadcast a public one-way value for swarm verification without disclosing the key.Source: B. Private Authentication for… · pp. 20–21
  • Mode A protects and conveys a public key using a Module-LWE-based layer, allowing authorized peers to recover it while outsiders face layered security assumptions.Its construction combines a masked key and one-way tag with Module-LWE diffusion; the stated quantum key-recovery bound is Ω(2^128) Grover queries at λ = 256.Source: B. Private Authentication for… · pp. 21, 23
  • Mode B locally derives the shared key chain from end-to-end seeds and broadcasts only a one-way commitment, placing no key material on the wire.Its HNDL resistance relies on the pre-shared protector bundle remaining secret; peers independently reconstruct matching keys and commitments.Source: B. Private Authentication for… · pp. 23–24
  • The modes target the same 128-bit post-quantum level but trade off interoperability and explicit lattice hardness against lower bandwidth and computation.Mode A transports larger encapsulated public-key objects; Mode B is the lightweight default for resource-constrained nodes but relies more heavily on pre-shared roots and their lifecycle management.Source: B. Private Authentication for… · p. 24
  • Proof-of-Edge consensus uses pattern-based peer validation to confirm participants hold a consistent renewed key, without interrupting ongoing message exchange.After consensus, peers build local blocks anchored by the commitment or encapsulation, hash, generation and consensus times, and owner; blocks move from RAM to flash after synchronization.Source: B. Private Authentication for… · p. 24

Key reconciliation

  • After approval and match signals reach the master, peers check local-chain integrity before a new key becomes operational.Key updates renew peers’ security levels against asymmetric quantum threats.Source: B. Private Authentication for… · p. 25

Node and data architecture

  • PARS prefers Light Nodes over Full Nodes for space and defense systems and stores executive data in order, with sensitive data encrypted for rotational reconciliation.The stated aim is a distributed architecture.Source: B. Private Authentication for… · p. 25

C. PARS-Enabled ESL

PARS-ESL is presented as a security-focused ledger for space and defense systems, supporting node authentication and authorization while keeping operational data off-chain. Its scope includes key management and security-state recording rather than transaction-by-transaction data retention.

C. PARS-Enabled ESL

  • The ledger supports joining and leaving nodes, identity checks, authentication, authorization changes, security-score evaluation, and communication security.The passage describes zero-knowledge proof agreement between validator and prover without revealing confidential information.Source: C. PARS-Enabled ESL · p. 25
  • PARS-ESL records key-epoch commitments, membership and revocation events, algorithm-suite identifiers, validation outcomes, and recovery checkpoints, while operational data remain off-chain.Operational data are protected by the active traffic keys.Source: C. PARS-Enabled ESL · p. 25
  • PARS-ESL is described as a reference implementation whose security and performance claims are bounded by protocol, cryptographic, provisioning, validator, network, and platform assumptions.The text says subsequent evaluation examines costs and limitations empirically.Source: C. PARS-Enabled ESL · p. 25

D. Quantum Hyperlattice Tensor Key Agreement

Quantum Hyperlattice Tensor Key Agreement (QHT-KA) uses two QCKA phases to establish a shared application key without carrying the seed or final key in a classical message. The tensor contributes scheduling and representation services, while the construction still requires authenticated public exchanges and an initial authentication basis.

D. Quantum Hyperlattice Tensor Key Agreement

  • QHT-KA prepares a shared private seed and commits to path representations, then uses fresh QCKA entropy to select a path and complete the application key.Neither the seed nor final key is carried in a classical message.Source: D. Quantum Hyperlattice Tensor Key… · p. 25
  • The keyed virtual tensor adds scheduling and representation services, but no entropy or lattice-hardness assumption; direct KDF(ENC(S, R), context) would provide the same key-secrecy level.The described services include a multidimensional namespace, nested paths, synchronized touch updates, and commit-before-selection.Source: D. Quantum Hyperlattice Tensor Key… · p. 26
  • QHT-KA still exchanges authenticated public data for estimation, reconciliation, privacy amplification, certification, and confirmation, and cannot authenticate previously unknown nodes by itself.The core profile uses direct multipartite key distillation rather than a dealer distributing a group key over pairwise keys.Source: D. Quantum Hyperlattice Tensor Key… · p. 26

1) Two independently distilled quantum secrets:

The protocol derives two group secrets through separate QCKA executions, with authenticated discussion and explicit abort conditions. Its practical group size remains constrained by link quality and finite-key costs.

1) Two independently distilled quantum secrets:

  • Each QCKA execution assigns test and key rounds, authenticates discussion, and aborts unless phase-error, bit-error, loss, source, and finite-size conditions hold.Measurement-device-independent QCKA can reduce detector trust.Source: 1) Two independently distilled quantum… · p. 26
  • State-generation rate, loss, finite-key cost, and the worst participant link bound practical group size.The cited discussion concerns measurement-device-independent QCKA for a finite number of parties.Source: 1) Two independently distilled quantum… · p. 26
  • The protocol uses independently nonced QCKA executions to set S = Zprep and, after authorized reveal, R = Zrev.The executions use independent quantum rounds, extractor seeds, and transcript domains; otherwise a joint entropy proof is required.Source: 1) Two independently distilled quantum… · p. 27
  • Privacy amplification makes the reconciled output close to uniform conditioned on the public transcript and adversary’s quantum side information.The public extractor seed must be uniform and independent of the reconciled string.Source: 1) Two independently distilled quantum… · p. 27

2) Virtual module tensor and fixed-budget sampling:

The section defines a finite tensor-indexed module over an underlying module lattice, then evaluates requested cells online using fixed-budget sampling. This avoids materializing the full tensor.

2) Virtual module tensor and fixed-budget sampling:

  • The tensor-indexed finite module is derived from an underlying module lattice and is explicitly not an LWE instance.“Hyper Lattice Space” names the author-defined product XG = HG×Tq×DΓ, not a recognized hard problem.Source: 2) Virtual module tensor and… · p. 27
  • Requested tensor cells are evaluated online, while fixed-budget XOF sampling selects the first in-range value by constant-time masking or aborts.Conditional on acceptance, rejection sampling is unbiased; direct modulo reduction is forbidden when biased.Source: 2) Virtual module tensor and… · p. 27
  • Online evaluation uses O(1) tensor-cell storage, whereas materializing the tensor costs Nkd⌈log2 q⌉ bits before software overhead.The fixed budget converts variable timing into a parameterized, negligible abort probability.Source: 2) Virtual module tensor and… · p. 27

3) Certified touches and explicit tensor views:

Certified touches authorize state transitions, while an explicit view transform changes virtual tensor coordinates. A touch is not itself a rekey and adds no entropy or recovery property.

3) Certified touches and explicit tensor views:

  • Nodes accept a touch only with a certificate over the previous state, rejecting gaps, duplicates, stale predecessors, invalid certificates, and equivocation evidence.For n ≥ 3f + 1, 2f + 1 signers suffice if each honest signer signs at most one command for a given state and sequence.Source: 3) Certified touches and explicit… · p. 27
  • A touch is a public state-transition instruction, not a rekey, and adds no entropy, forward secrecy, or compromise recovery.Any accepted touch after bank commitment invalidates that commitment and requires rebuilding it.Source: 3) Certified touches and explicit… · p. 27
  • The view transform derives a bijective coordinate map and invertible component transformation from an XOF using the preparation state, touch, and certificate.The construction samples a unit multiplier, offset, component permutation, signs, and negacyclic rotations.Source: 3) Certified touches and explicit… · p. 27

4) Streaming nested paths and bank commitment:

The protocol streams candidate paths and commits to the bank before selection, then delays key derivation until an authorized reveal. Its guarantees are bounded: the opening proves bank membership, not correct secret computation, and the release is neither simultaneous nor fair.

4) Streaming nested paths and bank commitment:

  • Candidate paths use fixed-budget samplers and rolling digests, and their representation maps the final digest to a display symbol without adding entropy.Failure to find a requested display symbol within the fixed-budget counter search causes a synchronized abort.Source: 4) Streaming nested paths and bank… · p. 27
  • Preparation commits to the entire bank before the selected index exists, and every node requires a complete, identity-signed set of identical commitment votes.The Merkle profile specifies arity, index order, tags, padding, and digest suite.Source: 4) Streaming nested paths and bank… · p. 27
  • The later selected opening proves membership in the precommitted bank, but does not prove that the secret tensor computation was performed correctly.That stronger property requires a separately analyzed zero-knowledge proof or attestation mechanism.Source: 4) Streaming nested paths and bank… · p. 28

5) Authorized just-in-time reveal and confirmation:

  • The delayed profile waits for a reveal certificate binding the commitment, membership, nonce, QCKA profile, one-shot sequence, and expiry before an independent QCKA derives R.The selected index and final key do not exist at an endpoint before the fresh reveal execution.Source: 5) Authorized just-in-time reveal and… · p. 28
  • After every expected identity supplies one consistent signed value, the protocol publishes the Merkle opening, erases secret and path state, and marks the bank consumed.This one-shot default prevents later seed exposure from revealing unused candidates.Source: 5) Authorized just-in-time reveal and… · p. 28
  • Transcript-consistent group agreement is not atomic release: an honest node may finish before another, and a malicious member may learn the key then abort.Exact simultaneity or fairness requires additional synchrony, threshold-release, or trusted-hardware assumptions; delayed ignorance against admitted endpoints needs an enforced release scheme.Source: 5) Authorized just-in-time reveal and… · p. 28

6) Two-phase Qiskit-style pseudocode:

The section gives a two-phase QHT-KA implementation structure, states its conditional security boundaries, and evaluates PARS/PARS-ESL resource costs on a physical Edge AI testbed. It distinguishes measured testbed results from platform projections.

6) Two-phase Qiskit-style pseudocode:

  • The two-phase listing prepares a committed bank from a QCKA-derived secret, then reveals a selected path and derives an application key after quorum authorization and confirmation.Preparation binds a Merkle root; reveal verifies an opening and derives the application key from the pre-commit and reveal contributions.Source: 6) Two-phase Qiskit-style pseudocode · pp. 28–29
  • Release profiles trade timing and premature-disclosure resistance against quantum-storage requirements; the threshold profile requires at least t authorized shares under the selected QSS/MPC guarantees.The immediate-release profile excludes outsiders but does not prevent already authorized members from accessing the key.Source: 6) Two-phase Qiskit-style pseudocode · p. 29
  • A display target changes visible representation, not key entropy; password admission needs a separately analyzed post-quantum PAKE, and action vectors need measured conditional min-entropy and spoofing bounds.The paper also cautions that a two-party Module-LWE PAKE does not establish an n-party proof.Source: 6) Two-phase Qiskit-style pseudocode · p. 29

7) Conditional security claim and residual limitations:

  • The security bound is indicative, not a composable theorem; proving it requires modeling adaptive corruption, selective abort, insider leakage, reveal scheduling, erasure, and both QCKA interfaces.Tensor size, path count, and representation count do not increase the entropy available from (S, R).Source: 7) Conditional security claim and… · p. 29
  • The just-in-time profile is not an unconditional time lock: authorized or sufficiently large colluding sets can reveal early, every admitted member learns the released key, and one-shot state requires fresh preparation after specified failures or changes.The concrete profile must also freeze suites, parameter and sampler budgets, quorum rules, and storage, erasure, and recovery behavior.Source: 7) Conditional security claim and… · p. 29

V. RESOURCE-CONSTRAINED EDGE AI IMPLEMENTATION

  • The authors report the first end-to-end validation of a quantum-safe decentralized key-management architecture on an operational physical Edge AI CPS testbed, bounded by the reported hardware, topology, algorithms, and workloads.They caution that these results are not evidence of universal Edge AI suitability.Source: §V · p. 30
  • The evaluation separates primitive and protocol measurements from derived overhead and platform projections, with cryptographic microbenchmarks run on an Intel i5-12450H gateway-class platform.Measured runtimes represent a capable validator or ground-control-class node, not a deeply embedded endpoint; projections are not physical measurements.Source: §V · p. 30
  • 0.8-10 KB is the reported range for post-quantum public keys, ciphertexts, and signatures, shifting deployment pressure toward fragmentation, buffering, transmission, and replicated-state costs.The analysis identifies object size, rather than computation alone, as the primary operational bottleneck.Source: §V · p. 30

A. Average Consensus Time

Consensus time captures agreement on a key-epoch update, not operational traffic latency; the reported sweep and size measurements show that post-quantum reconciliation has coordination and communication costs that vary with configuration. Ledger traffic remains distinct from bulk application traffic, but its bursts and memory needs must be assessed against the target link and measurement scope.

A. Average Consensus Time

  • The four legacy “ZKP hardness” settings are challenge-complexity levels, not demonstrated zero-knowledge security levels.Formal completeness, soundness, and zero-knowledge properties have not been established; the measurements also lack phase-level decomposition and a classical control configuration.Source: A. Average Consensus Time · pp. 31–32
  • 8.53 s to 91.05 s: completion times span the reported sweep, from the lowest-cost ML-KEM configuration to a 16-node SIKE stress case at the highest challenge level.The upper value uses an excluded, cryptanalytically broken scheme; lattice configurations generally complete in the single-digit to low-tens-of-seconds range, but uncertainty and trial counts are needed for a service bound.Source: A. Average Consensus Time · p. 31, Figure 35 · p. 32
  • Consensus latency rises with node count and challenge complexity as participants add messages, validations, synchronization points, and potential retries.Consensus time measures key-epoch proposal agreement, not telemetry protected by an already active symmetric key.Source: A. Average Consensus Time · pp. 30, 32

B. Size Metrics (Message-Transaction-Memory-Bandwidth)

  • About 10 KB is the maximum reported message size, while crypto-agile options include approximately 100 B and 800 B; PQC objects can raise reconciliation link budgets by one to two orders of magnitude over classical primitives.Fig. 36(a) tracks underlying public-key, ciphertext, and signature sizes; low-MTU links may require fragmentation, and the exact burden depends on link profile.Source: B. Size Metrics… · p. 32, Figure 36 · p. 32
  • Transactions range from 0.2 KB to 9.53 KB; after key reconciliation, synced per-node memory remains below 256 KB across tested node counts for low-rate ledger records.The memory figure is interpreted according to the implemented retention policy and measured memory category.Source: B. Size Metrics… · p. 32
  • Reconciliation demands 1.67–78 kbit/s in instantaneous bandwidth across multiple iteration cycles, while operational traffic proceeds under standard conditions after reconciliation.The security-control burst must be absorbed by the network layer; the reported range does not establish that every configuration is sustainable.Source: Figure 36 · p. 32, B. Size Metrics… · p. 33

C. Cumulative Memory Footprint

In the tested 16-node, 18-hour configuration, peak memory remained below 3.5 MB, while longer rotation intervals reduced accumulated state. Denial and failure observations vary with operating conditions, but single-series results and self-induced failures limit reliability and adversarial conclusions.

C. Cumulative Memory Footprint

  • Peak memory stayed below 3.5 MB across 16 nodes during an 18-hour run.The node count was chosen as a realistic tactical-swarm ceiling; the result demonstrates bounded growth only for the tested duration and configuration.Source: C. Cumulative Memory Footprint · p. 33
  • Longer PSTC intervals reduce accumulated state because fewer key-epoch transitions produce fewer security-state blocks over a fixed period.This is a rotation-frequency trade-off, not an independent memory optimization; the memory result does not establish weeks-to-years behavior or MCU feasibility.Source: C. Cumulative Memory Footprint · p. 33

D. Denial of Consensus and Failure Analysis

  • The reported denial count decreases as the reconciliation period lengthens, but that does not show that individual rounds become more reliable.The observed count reflects fewer attempted rotations and fewer opportunities for timeout or synchronization failure; reporting both attempted and failed rounds is needed to assess rates.Source: D. Denial of Consensus and Failure… · p. 34
  • Failure rates span roughly 10^-4 to 10^-1, rising with challenge level and node count; the upper end occurs only for the largest swarm, highest levels, and slowest excluded schemes.The passages attribute observed events to protocol/network causes such as proof timeouts, rejected proofs, and desynchronization, rather than expected ML-KEM decapsulation failure.Source: D. Denial of Consensus and Failure… · p. 34, Figure 39 · p. 34
  • Larger node counts, greater challenge complexity, and slower historical schemes show higher unsuccessful-round rates, but the single series per condition has no confidence intervals.The reported trends are prototype observations, not statistically bounded reliability estimates.Source: D. Denial of Consensus and Failure… · p. 34
  • The present failures are self-induced by load, timeout, and synchronization behavior; active-adversary behavior remains untested.The authors call for repeated runs and adversarial tests involving malformed transactions, replay, conflicting commitments, Byzantine validators, and partition-and-merge behavior.Source: D. Denial of Consensus and Failure… · pp. 34–35

E. Cross-Platform Projection for Edge AI Hardware

The cross-platform analysis projects PQC costs across heterogeneous edge hardware and describes how programmable logic can accelerate cryptographic operations. It also identifies cryptographic-object size—not computation—as a binding constraint on bandwidth-limited links, motivating controlled key-transition windows and symmetric data channels.

E. Cross-Platform Projection for Edge AI Hardware

  • Platform-cost projections use published per-platform cycle counts where available and otherwise scale the GCS-class Intel i5-12450H baseline by clock frequency and issue width, crediting ISA extensions.The projections cover twelve representative platforms; projected entries include a measured-versus-projected error column.Source: E. Cross-Platform Projection for Edge… · p. 35, Table V · p. 36
  • 8.7–12.5× (KeyGen), 6.3–7.3× (Sign), and 9.1–12.2× (Verify) Dilithium speedups are reported over the Cortex-A9 software baseline for Zynq co-designs.Comparable RISC-V/FPGA co-designs accelerate Kyber and Dilithium by 3–5× while using under 5% of the fabric.Source: E. Cross-Platform Projection for Edge… · p. 36
  • 3,768/5,079/6,668 cycles for the lightest Kyber parameter set and 9,380/11,321/13,908 for the heaviest are reported for key generation, encapsulation, and decapsulation on Artix-7 hardware at 204 MHz.The implementation uses roughly 7,400 LUTs, two DSPs, and three BRAMs.Source: E. Cross-Platform Projection for Edge… · p. 36
  • Tens-of-microseconds pure-PL encapsulation latency is reported, while constant-time spatial datapaths can support redundancy and avoid software cache and branch timing leakage.Power and electromagnetic side channels remain and require datapath-level countermeasures.Source: E. Cross-Platform Projection for Edge… · p. 36
  • 9.53 KB for a heavy-scheme reconciliation transaction and 1.67–78 kbit/s tactical links make PQC object size, rather than accelerated computation, the binding link constraint.The cited discussion warns that naive per-message re-keying can trigger fragmentation and congestion.Source: E. Cross-Platform Projection for Edge… · pp. 36–37
  • PARS schedules large key-establishment and authentication objects within controlled transition windows, then uses symmetric AEAD for high-rate data and amortizes object overhead across application packets.The ESL stores commitments and security-control metadata rather than the application stream.Source: E. Cross-Platform Projection for Edge… · p. 37
  • Published embedded profiles place NIST lattice-scheme energy per operation on the same order of magnitude as elliptic-curve cryptography, while SLH-DSA is one to two orders of magnitude higher.Direct energy-per-consensus-round measurement is included in the planned on-device campaign.Source: E. Cross-Platform Projection for Edge… · p. 37

F. PQC Support Across the Edge Platforms

The paper describes software and accelerated PQC paths across edge platforms, with PARS crypto-agility accommodating evolving algorithm options. For heterogeneous Edge AI traffic, it assigns different cryptographic patterns and hardware roles, while treating platform mappings as guidance requiring on-device validation.

F. PQC Support Across the Edge Platforms

  • PARS crypto-agility can accommodate HQC or compact-signature FN-DSA as standards mature, while FN-DSA's floating-point signing path complicates constant-time embedded implementation.The passage identifies SIKE as permanently excluded.Source: F. PQC Support Across the Edge Platforms · p. 37
  • Every target platform has at least a software path to ML-KEM and ML-DSA, and most also offer an accelerated or hardware-hardened path.The described paths include OpenSSL on Linux-capable processors, GPU implementations, programmable-logic cores, and optimized microcontroller software.Source: F. PQC Support Across the Edge Platforms · pp. 37–38

G. Securing Heterogeneous Edge AI Data Flows

  • 16 B is the steady-state per-record AES-256-GCM tag cost; kilobyte-scale PQC objects are paid per session, rotation, signing, or update rather than per packet.The object-size values are derived from cited standards and should be recalculated for the implemented parameter sets and protocol encoding.Source: G. Securing Heterogeneous Edge AI Data… · pp. 38–39
  • Federated-learning updates use ML-DSA signatures with key rotation at PSTC boundaries, while firmware signing follows CNSA 2.0's LMS/XMSS or ML-DSA-87 requirements for NSS.Stateful LMS/XMSS key management requires never reusing a leaf index; the passage confines these schemes to controlled, low-volume firmware signing.Source: G. Securing Heterogeneous Edge AI Data… · p. 39
  • Classified NSS deployments must use Category-5 ML-KEM-1024 and ML-DSA-87, with LMS/XMSS or ML-DSA-87 for firmware signing; lighter parameter sets and specified alternatives are disallowed.The passage states an exclusive-use procurement gate from 1 January 2027 and full NSS transition by 2035.Source: G. Securing Heterogeneous Edge AI Data… · p. 40
  • Structured lattice schemes are recommended for real-time re-keying, while SLH-DSA is reserved for offline uses and SIKE is excluded from configurations.The passage gives an approximately 75 ms SLH-DSA processing bottleneck and recommends smaller-signature SLH-DSA-128s for offline signing outside NSS.Source: G. Securing Heterogeneous Edge AI Data… · p. 40
  • PSTC renewal can be lengthened in degraded links while action- or group-triggered re-keying remains available; large swarms can be partitioned into independent sub-clusters.The passage frames this as a trade-off between re-key frequency and assured delivery in DDIL environments.Source: G. Securing Heterogeneous Edge AI Data… · p. 40
  • Platform-specific mappings and projections are architectural extrapolations, not physical on-device measurements, and require validation in active edge deployments.The stated future work includes broader trials, a classical control run, energy and latency profiling, and validation on embedded targets.Source: G. Securing Heterogeneous Edge AI Data… · pp. 40–41

H. Hardware Co-Design Analysis Across 12 Target Substrates

Across 12 target substrates, the analysis compares PQC hardware support and PARS-ESL deployment considerations. The conclusion reports bounded memory and consensus-failure results while identifying conditions and future evaluation needs.

H. Hardware Co-Design Analysis Across 12 Target Substrates

  • Pure-software execution on constrained embedded cores is projected to run 12× to 300× slower than the GCS baseline because of Module-LWE NTT loops.The analysis spans 12 substrates across five deployment tiers.Source: H. Hardware Co-Design Analysis Across… · p. 41
  • Adaptive-SoC programmable-logic offload is reported to recover desktop-equivalent cryptographic latencies at single-digit-Watt budgets; 512-bit RVV systems reach 1.2× to 2.2× baseline latency.The RVV result is reported without discrete FPGA coprocessors.Source: H. Hardware Co-Design Analysis Across… · p. 41
  • Hardware acceleration reduces runtime but does not shrink PQC public-key, ciphertext, or signature byte footprints.PARS confines expensive lattice handshakes to controlled PSTC windows and keeps steady-state telemetry on AES-256-GCM; peak volatile memory is reported below 3.5 MB across the 12 platform classes.Source: H. Hardware Co-Design Analysis Across… · p. 41
  • Table X is identified as a cross-platform overview of hardware co-design, PQC support tiers, and PARS-ESL across representative deployment substrates.The listed tiers include gateways, space systems, defense avionics, robotic and UAV swarms, and microcontroller endpoints.Source: H. Hardware Co-Design Analysis Across… · p. 41, Table X · p. 42

VI. CONCLUSION

  • The conclusion describes consensus-based key renewal and predetermined key-control intervals, with future work planned on heterogeneous inter-ESLs and expanded empirical metrics.The conclusion frames feasibility within the proposed approach and identifies operational inter-ESLs as future work.Source: §VI · pp. 42–43
  • Under 256 KBytes of local-chain memory, peak memory stayed below 3.5 MB over 18 hours, while consensus failure rates ranged roughly from 0.01 to 0.001.Consensus completed in under 91.05 seconds across the reported PQC-algorithm, node-set-size, and ZKP-hardness combinations.Source: §VI · pp. 42–43

Related papers

Loading 2610.33862v1…