Experimental Evaluation of Post-Quantum Homomorphic Encryption for Privacy-Preserving I2I Communication in ITS

TL;DR

This paper experimentally evaluates post-quantum homomorphic encryption schemes for privacy-preserving data exchange in intelligent transportation systems, demonstrating their practical latency performance and proposing optimization strategies for real-world deployment.

Contribution

It implements and benchmarks three lattice-based HE schemes in a real ITS environment, providing practical latency insights and optimization pathways for privacy-preserving analytics.

Findings

BFV achieves sub-5-second latency for intersection analytics

BGV supports regional data aggregation with 10-30 second updates

CKKS is suitable for applications with minute-scale latency

Abstract

This study experimentally evaluates the feasibility of post-quantum secure Homomorphic Encryption (HE) for privacy-preserving Infrastructure-to-Infrastructure (I2I) communication in Intelligent Transportation Systems (ITS). Unlike prior simulation-based efforts, this work implements three lattice-based HE schemes: Brakerski-Fan-Vercauteren (BFV), Brakerski-Gentry-Vaikuntanathan (BGV), and Cheon-Kim-Kim-Song (CKKS), within a real experimental pipeline representing roadside unit (RSU)-Cloud data exchange over Wi-Fi and Ethernet networks. The experiments benchmark encrypted addition and addition-plus-multiplication operations representing key analytical tasks, such as vehicle queue assessment and regional speed computation. Results show that while BFV achieves sub-5-second latency suitable for intersection-level analytics, BGV supports regional aggregation with 10 to 30-second updates. CKKS, though exhibiting higher latency (21-32 seconds), remains practical for minute-scale applications like eco-driving. These findings demonstrate that post-quantum HE can enable privacy-preserving ITS backhaul analytics when latency requirements align with application needs. The study also presents optimization pathways, including algorithmic tuning, network adaptation, and hardware acceleration, to reduce end-to-end delay.