@article{Kim2026, 
author = {Hyunghoon Kim and Wonsuk Choi and Kyungho Joo and Hyojin Jo},
title = {Analytical Modeling of Transcoding Artifacts for Detecting SIMBox-Routed Calls},
year = {2026},
journal = {Computer Modeling in Engineering & Sciences},
volume = {148},
number = {1},
pages = {53},
keywords = {SIMBox-routed call detection, VoIP-to-VoLTE transcoding, client-side call analysis},
url = {https://www.sciopen.com/article/10.32604/cmes.2026.083948},
doi = {10.32604/cmes.2026.083948},
abstract = {As mobile networks evolve toward next-generation architectures in which cellular and IP-based voice services are increasingly converged, SIMBox-based call routing has emerged as an important issue in modern telecommunication networks. By converting Voice over IP (VoIP) traffic into local cellular calls, SIMBox appliances allow IP-originated calls to appear as domestic cellular calls. Although SIMBox usage is not inherently fraudulent, detecting SIMBox-routed calls is important for identifying abnormal call-routing behavior and supporting network-side and client-side security applications. In this paper, we propose a client-side framework for SIMBox-routed call detection. Calls routed through SIMBox infrastructure are identified by exploiting acoustic artifacts introduced by VoIP-to-VoLTE codec transcoding. Variable-length call recordings are segmented into fixed-duration windows and analyzed using supervised classifiers to capture spectral patterns associated with transcoding operations. Unlike network-side SIMBox detection methods that rely on carrier-controlled metadata such as call detail records, subscriber identifiers, or cell-location patterns, our approach focuses on acoustic evidence observable from the call audio itself. We evaluate the proposed framework using two datasets: 25,706 codec-processed voice samples generated from 12,853 source recordings with the ITU-T G.191 Software Tool, and 200 real-world call samples collected through a commercial SIMBox platform (DINSTAR UC2000-VE) under realistic call-routing conditions. Among five evaluated models, the CNN-based classifiers achieve the best performance, reaching an F1-score of 1.00. These results show that audio-level codec artifacts can serve as reliable indicators for detecting SIMBox-routed voice calls at the client side, and offer a promising direction for securing converged voice services in 5G and beyond.}
}