from 01.01.1921 until now
This paper presents OmniGuard V2X, a security prototype for connected and au-tonomous vehicles. The system integrates blockchain auditing, zero-knowledge privacy, decentral-ized identity, post-quantum key establishment, federated learning, and an encrypted forensic black box into a unified architecture. Rather than making unrealistic performance claims, the study openly discusses its current limitations, including the lack of protection against Sybil attacks and majority-controlled validators.
V2X, blockchain, zero-knowledge proofs, post-quantum cryptography, decen-tralized identity, federated learning, vehicle security
Connected and autonomous vehicles depend on constant V2X communication to exchange telemetry, traffic, and safety data, and that dependence creates real problems for authentication, integrity, privacy, and secure transport of information. The long service life of a modern vehicle adds another constraint: any security architecture adopted today has to survive the transition to post-quantum cryptography and still produce forensic evidence that can be trusted after an incident [1].
This paper presents OmniGuard V2X, a prototype that combines blockchain-based audit logging [2], privacy-preserving data sharing, decentralized identity, hybrid post-quantum key establishment, federated learning for anomaly detection, and an encrypted forensic black box into one six-layer architecture. None of the cryptographic primitives are new; the contribution is in how established techniques are combined into a working, modular pipeline for connected vehicles. The paper reports a reproducible end-to-end latency of 5.34 ms and is deliberately open about what the system does not yet solve — Sybil resistance and majority-controlled validators chief among them — because a realistic account of the boundaries is more useful to other researchers than an inflated one.
Blockchain has repeatedly been shown to strengthen V2X security through trusted authentication, secure data sharing, and decentralized access control [3], but the literature is consistent on what is still missing: low-latency operation, lightweight security for constrained vehicle hardware, scalability, and real-world validation beyond simulation. Most published work isolates a single mechanism rather than evaluating a complete pipeline end to end.
OmniGuard V2X does not introduce new cryptography. It combines Pedersen commitments [4] with Schnorr-style zero-knowledge proofs for privacy, Lamport one-time signatures for identity, Merkle trees for compact verification, and SHA-3 with AES-GCM for integrity and storage, on top of a hybrid post-quantum key exchange [3] and a FedAvg-based federated learning model [5] with trimmed-mean aggregation for resilience against bad participants. What distinguishes this work from many conceptual security frameworks is that the prototype has actually been run, and its limitations — no defense against Sybil identities, majority-controlled validators, or large-scale Byzantine adversaries — are stated rather than glossed over.
Figure 1 shows the six-layer design. Sensor data is validated first, then passed through an authenticated key exchange, a privacy and identity layer, and an intermediate federated-learning stage where vehicles can collaborate on anomaly detection without sharing raw data. Verified events are written to a blockchain audit layer, and the top layer exposes APIs and a dashboard for monitoring. Two points about this layering matter for everything that follows. First, federated-learning updates are verified and logged exactly like any other event, rather than being treated as a special case. Second, the layers separate functions, not trust boundaries — a valid signature proves where data came from, not that the data is correct if the originating sensor was already compromised.
Figure 1– OmniGuard V2X six-layer prototype architecture
(L1–L5, with the L3.5 federated-learning layer).
The proposed architecture combines various existing security techniques to create a unified framework for smart vehicles communication. Rather than trying to come up with a solution from scratch, the approach combines decentralized identities, private authentication, blockchain audit, quantum-resilient communications, federated learning, and forensic log encryption.
Every smart vehicle gets assigned a decentralized identity based on Lamport one-time signature scheme that only uses one-way hash functions and, therefore, is believed to be adequate for post-quantum world. Each key pair can be used only once so a new one has to be created for every authentication attempt. The current implementation of the framework uses open registration and thus anybody can create an identity in the system. However, this doesn't prevent Sybil attack and thus a real-life implementation should include some external entity responsible for enrollment such as vehicle manufacturer or transportation authority.
To ensure privacy in the system the framework uses combination of Pedersen commitment and Schnorr-style zero-knowledge proof [4]. In particular, this enables vehicles to prove their compliance to certain rules, like remaining under certain speed limit, without disclosing their exact speed. Thus, it provides good privacy but relies on classical assumption of discrete logarithm problem and thus is not post-quantum yet.
Integrity of the blockchain is guaranteed through usage of dual-hash structure based on SHA-2-256 and SHA-3-256 hashes. Such structure makes detecting the modifications very easy. Proof-of-authority consensus guarantees proper validation of new blocks and thus provides robust auditing and traceability although still relying on trusted validators. Security of communications between vehicles starts from capability negotiation and then session key establishment based on ML-KEM (Kyber) [3]. Crypto-agility layer is implemented to select proper signing algorithm, while post-quantum security applies only to the key exchange.
Finally, the architecture uses federated learning for joint anomaly detection without exchanging raw vehicle data. Update of local models is secured with differential privacy, filtered to decrease influence of malicious contribution, and aggregated through trimmed mean strategy. First tests show good resistance to abnormal updates although larger scale test are necessary yet. Other features of the architecture include encryption of forensic logs, biometric safe-mode activation, proof of proximity verification, and AES-256-GCM encrypted blockchain storage.
The computational complexity of the proposed framework varies according to the function being executed. Most local operations, including telemetry validation, ML-KEM key establishment, and blockchain insertion, are performed in constant time, 
. As a result, the processing cost of an individual vehicle remains stable even as the number of connected vehicles increases. Communication-intensive operations behave differently. When every vehicle exchanges information with every other vehicle or when multiple validators participate in the same consensus round, the total communication cost grows to 
. Blockchain verification depends primarily on the number of stored blocks, whereas the cost of federated-learning aggregation increases with both the number of participating clients and the size of the learning model. These complexity values describe expected theoretical behavior rather than measurements from large-scale deployment. Committee-based validation, hierarchical gossip, batching, and sharding are planned to improve scalability in future versions of the framework.
Performance was evaluated using the prototype implementation in a controlled local environment. The measured end-to-end latency of the complete processing pipeline was 5.34 ms, recorded after the system had reached a stable operating state with initialized cryptographic contexts and cached resources. This result is presented as a reproducible prototype measurement rather than as a benchmark for production-scale V2X networks. An earlier latency value of 3.57 ms was removed from the final evaluation because repeated experiments could not reproduce the same result with sufficient consistency.
Only results that could be reproduced reliably are included in this study; protocol overhead, large-scale adversarial detection accuracy, and comprehensive federated-learning robustness are therefore not reported as quantitative metrics. The current implementation demonstrates feasibility, while a more extensive evaluation involving larger vehicle fleets, diverse network conditions, and repeated trials remains necessary to validate performance under real-world operating conditions.
Four limitations recur often enough across the preceding sections to be worth restating together, summarized in Table 1. Identity and consensus share one root cause — open registration — since nothing under the current policy limits how many DIDs a single actor can create, so neither the majority-vote chain nor the proximity check is Sybil-resistant by itself; both assume an admission step that has not been built. Dual-hash integrity and majority-controlled validation answer different questions — whether a record was altered after the fact, versus whether enough validators agreed to accept it when it was written — and a quorum-controlling coalition can satisfy the second while still recording something false, which the hash chain will then faithfully preserve. Post-quantum coverage is real but partial: the ML-KEM handshake is genuinely protected against a future quantum adversary, while the Pedersen/Schnorr privacy layer is not, and describing the whole system as post-quantum without that qualification would be inaccurate. Finally, the federated-learning robustness result is the most provisional claim in the paper — a three-peer, single-trial sanity check demonstrates that one large, obvious outlier gets caught, not that the pipeline withstands an adversary operating near the detection boundary.
Table 1 – Documented limitations and what would close each one.
|
Mechanism |
Documented limitation |
What would close it |
|
DID identity (4.1) |
No Sybil resistance under open registration. |
External admission: certificate, enrollment, or stake. |
|
Majority-vote chain (4.2) |
Quorum control can sign false but valid-looking blocks. |
Bounded-committee BFT / certificate-gated validators. |
|
PQC handshake (4.2) |
Only the key-establishment path is post-quantum. |
Lattice- or hash-based replacement for the privacy layer. |
|
FL aggregation (4.3) |
Validated on 3 peers, 1 trial, 24 samples only. |
Larger peer counts, multi-trial, confidence intervals. |
|
ZKP privacy (4.1) |
Classical discrete-log security; commit-only mode. |
PQC-safe proof system; secure-aggregation protocol. |
|
Network layer (§5) |
O(n²) volume under full mesh / naive gossip. |
Committee selection, hierarchical gossip, batching. |
The project's own label for these corrections — corrected but requiring new experiments — is, in the authors' view, the right level of confidence for a system at this stage: the architecture is coherent and each component is individually defensible, but Sybil resistance, Byzantine-robust learning, full post-quantum coverage, and large-scale performance all need experiments beyond what a single-developer prototype can currently provide.
The clearest next steps follow directly from the limitations above. Closing the Sybil gap means integrating an external admission mechanism, most plausibly a manufacturer certificate chain or transport-authority enrollment, ahead of DID issuance. Replacing simple-majority validation with a bounded-committee Byzantine-fault-tolerant protocol [6] would address majority control directly, provided committee size stays small enough to avoid reintroducing O(n²) communication cost. Extending post-quantum coverage past the key-establishment path means replacing the Pedersen/Schnorr construction with a lattice- or hash-based proof system, since post-quantum range proofs remain an active research problem rather than an off-the-shelf component. On the learning side, the immediate need is simply more evidence — a larger peer count, multiple trials, a published confidence interval — since the trimmed-mean technique already in use is adequate; what is missing is statistical power. Network-layer work includes committee-selection, hierarchical-gossip, and sharding mitigations to keep full-mesh and naive-gossip patterns from dominating at fleet scale, and the forensic black box would benefit from hardware secure-enclave key wrapping and a standardized chain-of-custody format.
This paper presented OmniGuard V2X, a prototype framework that integrates blockchain auditing, privacy-preserving mechanisms, decentralized identity, post-quantum key establishment, federated learning, and secure forensic logging into a unified security architecture for connected smart vehicles, demonstrating how established security techniques can be combined into a practical, modular, and reproducible V2X security pipeline.
The prototype confirms that these components can operate together while maintaining a reproducible end-to-end latency and a clearly defined computational profile, and the study deliberately distinguishes validated results from future objectives by documenting current security boundaries and experimental limitations instead of making unsupported performance claims.
Although several challenges remain, including stronger identity admission, Byzantine-resilient consensus, broader post-quantum protection, and large-scale experimental validation, the proposed architecture provides a solid foundation for future research on secure intelligent transportation systems. The authors believe that openly presenting both the capabilities and the present limitations of the prototype makes OmniGuard V2X a useful reference platform for the continued development and evaluation of next-generation V2X security solutions.
1. Novikova, T. P. Matematicheskaya model' intermodal'nyh gruzovyh perevozok / T. P. Novikova, S. A. Evdokimova, A. I. Novikov // Modelirovanie sistem i processov. – 2024. – T. 17, № 2. – S. 62-70. – DOI:https://doi.org/10.12737/2219-0767 2024-17-2-62-70. – EDN ERRGGK. DOI: https://doi.org/10.12737/2219-0767-2024-17-2-62-70
2. Nakamoto, S. Bitkoyn: odnorangovaya elektronnaya platezhnaya sistema / S. Nakamoto. – 2008. – URL: https://bitcoin.org/bitcoin.pdf (data obrascheniya: 23.06.2026).
3. Standart mehanizma inkapsulyacii klyuchey na osnove modul'noy reshetki: Federal'naya publikaciya standartov obrabotki informacii FIPS 203 / Nacional'nyy institut standartov i tehnologiy. – Gaytersberg, Merilend: Nacional'nyy institut standartov i tehnologiy, 2024. – DOI:https://doi.org/10.6028/NIST.FIPS.203 (data obrascheniya: 23.06.2026).
4. Pedersen, T. P. Neinteraktivnoe i informacionno-teoreticheski bezopasnoe proveryaemoe razdelenie sekreta / T. P. Pedersen // Dostizheniya v kriptologii — CRYPTO’91. Lekcionnye zametki po informatike. – Berlin: Springer. – 1992. – T. 576. – S. 129-140.
5. Makmahan, H. B., Mur, E., Ramadzh, D., Hempson, S., Arkas, B. A. Kommunikacionno-effektivnoe obuchenie glubokih setey na osnove decentralizovannyh dannyh // Trudy 20-y Mezhdunarodnoy konferencii po iskusstvennomu intellektu i statistike (AISTATS). – PMLR. – 2017. – T. 54. – S. 1273-1282.
6. Kastro, M. Prakticheskaya vizantiyskaya otkazoustoychivost' / M. Kastro, B. Liskov // Trudy 3-go simpoziuma USENIX po proektirovaniyu i realizacii operacionnyh sistem (OSDI). – Novyy Orlean, Luiziana. – 1999. – S. 173-186.
