Title: A cross-domain adaptive deep learning framework for encrypted-decrypted traffic correlation and IoT terminal traceability

Authors: Libin Li; Tingting Yang; Kai Ma; Yueming Wang; Yongjiao Cao

Addresses: Information and Communication Company, State Grid Jilin Electric Power Co., Ltd., Changchun, 130012, China ' Information and Communication Company, State Grid Jilin Electric Power Co., Ltd., Changchun, 130012, China ' Information and Communication Company, State Grid Jilin Electric Power Co., Ltd., Changchun, 130012, China ' Information and Communication Company, State Grid Jilin Electric Power Co., Ltd., Changchun, 130012, China ' Information and Communication Company, State Grid Jilin Electric Power Co., Ltd., Changchun, 130012, China

Abstract: Encrypted traffic analysis poses a persistent and formidable challenge for IoT network security, especially in tracing terminal devices concealed behind Network Address Translation (NAT) gateways. This study introduces TCG-Net, a novel framework that enables robust encrypted-decrypted traffic correlation and precise IoT terminal traceability without payload inspection. The framework systematically integrates three innovative modules: a Temporal-Spectral Analysis Module (TSAM) that distils stable periodic signatures from noisy traffic data; a Cross-Modal Attention Fusion Network (CMAF-Net) that performs principled alignment of encrypted and decrypted representations in a shared latent space; and a Graph Attention-based Mapping (GAM) module that constructs an interpretable address-translation graph for terminal identification. Comprehensive experiments on three public datasets (CIC-IoT-2023, CICIDS-2017, and TON_IoT-2022) demonstrate that TCG-Net decisively outperforms state-of-the-art baselines, achieving 98.2% matching accuracy and 95.4% traceability accuracy. The results validate our central hypothesis that stable spectral-temporal patterns are the key to enabling secure, scalable, and privacy-preserving traceability in complex IoT ecosystems.

Keywords: encrypted traffic analysis; IoT traceability; cross-domain learning; time-series analysis; graph attention network.

DOI: 10.1504/IJWMC.2026.155713

International Journal of Wireless and Mobile Computing, 2026 Vol.31 No.2, pp.187 - 199

Received: 27 Oct 2025
Accepted: 12 Feb 2026
Published online: 11 Aug 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article