Perception-Aware RLNC with In-Network Recoding for Reliable Multi-Hop Data Transmission in Cloud-Native 5G
Description
Artefacts for "Perception-Aware RLNC with In-Network Recoding for Reliable Multi-Hop Data Transmission in Cloud-Native 5G" (Journal of Network and Computer Applications, JNCA-D-26-01586). Contains the encoder, in-network recoder and decoder code, the run configurations, the region-of-interest labels, the two ROI-labeled source PCAP traces, and the analysis scripts. Not included: the KODO coding library, which is closed source and requires a licence, although all code that drives it is here; and a standalone loss-injection script, because loss was applied directly with iptables rules whose per-configuration rule sets were not preserved. Section 4.4 of the article gives the injection points and the probabilities used. The two ROI cases are separate ROI-labeled traces of the same source sequence, not two labelings of a single trace: Case 1 has 54,220 packets (5,204 ROI) and Case 2 has 55,103 (31,970 ROI). The data, traces and labels in this deposit are released under CC BY 4.0; the source code is additionally made available under the MIT License.
Files
Institutions
- Chungnam National UniversityDaejeon, Daejeon
- Kongju National UniversityChungcheongnam-do, Gongju
- Technische Universität DresdenSaxony, Dresden
Categories
Funders
- National Research Foundation of KoreaMinistry of Science and ICTDaejeon, DaejeonGrant ID: RS-2023-00278925
- Bundesministerium für Forschung, Technologie und RaumfahrtNorth Rhine-Westphalia, BonnGrant ID: 16KISS012K
- Kongju National UniversityChungcheongnam-do, GongjuGrant ID: The research grant of Kongju National University in 2026