Toward Efficient Primal Attacks on Learning With Errors in IoT Environments
IEEE Internet of Things Journal · 2026-08 · 13 (16) · 36440-36454
DOI: 10.1109/jiot.2026.3701746 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Internet of Things Journal, 2026-08. Verfasst von Jingguo Bi und 6 weiteren Autor:innen. Die Arbeit nennt 44 Literaturangaben. Gefördert von: National Key Research and Development Program of China, Beijing Natural Science Foundation, Beijing University of Posts and Telecommunications-China Mobile Communications Group Company Ltd. Joint Institute.
Abstract (Originalsprache)
With the quantum threat looming, Internet of Things (IoT) deployments urgently require postquantum cryptographic (PQC) solutions that deliver strong security within tight resource limits. Lattice-based schemes, including NTRU-type encryption (IEEE Std 1363.1) and fully homomorphic encryption (FHE), are attractive in this setting because they can combine quantum resistance with relatively low-computational overhead. However, the accurate security assessment of these schemes is important for parameter selection in resource-constrained IoT deployments. This work addresses the ternary learning with errors (LWE) problem that underpins these cryptosystems by adapting cryptographic puncturing to lattice attacks. In this article, we revisit the randomized dimension reduction (RDR) technique originally proposed by May to assess the security of the original NTRU cryptosystem. More specifically, we study how the LWE sample dimension can be reduced while preserving the effective secret-error search space through an explicit puncturing threshold, replacing ad hoc sample-selection rules with a single-distribution-aware criterion. We also study LWE with side-channel hints, using elimination-based linear equation solving to construct the reduced hint lattice more transparently before the final embedding step, which is relevant in settings, where devices may be physically exposed to leakage. Experimental results demonstrate 41% threshold embedding-dimension reduction for NTRU-like schemes $(n,\log _{2}~q,h) = (100,12,50)$ and FHE parameters $(n,\log _{2}~q,h,\sigma _{e}) = (128,14,12,3.2)$ . Relative to standard lattice-estimator evaluations, our refined analysis tightens representative security estimates by 1–3 bits: a CKKS/HEAAN-style parameter set $(n,q,w) = (1024,2^{16},64)$ with sparse ternary secret is reduced by 3 bits, while NTRU-Prime $(n,q,w) = (653,4621,288)$ and LAC $(n,q,w) = (512,251,128)$ parameters are reduced by 1 bit. Additional coefficient-hint experiments on Kyber/ML-KEM parameter settings and representative hint counts from May–Nowakowski show that applying RDR after hint elimination can further reduce the estimated BKZ block size in hint-assisted regimes. These refined estimates provide a more concrete basis for parameter assessment in IoT-style settings.
Förderung
- National Key Research and Development Program of China
- Beijing Natural Science Foundation
- Beijing University of Posts and Telecommunications-China Mobile Communications Group Company Ltd. Joint Institute
Zitieren
APA
Bi, J., Luo, S., Lai, C., Zhou, B., Zhang, R., Li, L., & Peng, H. (2026). Toward Efficient Primal Attacks on Learning With Errors in IoT Environments. IEEE Internet of Things Journal, 13(16), 36440-36454. https://doi.org/10.1109/jiot.2026.3701746
Vancouver
Bi J, Luo S, Lai C, Zhou B, Zhang R, Li L, et al. Toward Efficient Primal Attacks on Learning With Errors in IoT Environments. IEEE Internet of Things Journal. 2026;13(16):36440-36454. doi:10.1109/jiot.2026.3701746
BibTeX
@article{Bi2026,
author = {Bi, Jingguo and Luo, Shuwen and Lai, Chunjiang and Zhou, Boyang and Zhang, Ran and Li, Lixiang and Peng, Haipeng},
title = {Toward Efficient Primal Attacks on Learning With Errors in IoT Environments},
journal = {IEEE Internet of Things Journal},
year = {2026},
volume = {13},
number = {16},
pages = {36440--36454},
doi = {10.1109/jiot.2026.3701746}
}
EUVIMED