Simulation-Driven Link-K Data Generation Framework to Optimize Resource Allocation

Sungbeen Kim, Ga San Jhun, Hyun Jun Yook, Su Yeon Kim, Daeyoung Song, Jae-Won Kim, Dohoon Kim, Youn Kyu Lee

Abstract: The ability to autonomously learn patterns from datasets has made artificial intelligence (AI) models increasingly attractive for use in optimizing timeslot allocation in Link-K time-division multiple access (TDMA) networks. Despite their potential, they require large-scale training datasets, and acquiring such data from tactical data link TDMA systems is challenging because of security constraints. This study focuses on the optimization of timeslot resource allocation in the Korean Joint Tactical Data Link System (JTDLS), which operates on a Link-K time-division multiple access network. Currently, network operators allocate timeslot resources mainly based on empirical experience. However, this study employs AI models trained on simulation data to enable predictive resource allocation. A simulation environment, the Link-K simulation data generator (LK-SDG), was developed to generate simulation data representing timeslot allocation outcomes across various operational scenarios. The generated dataset was validated for consistency using the Link-K timeslot allocation formula (LK-TAF) to ensure its reliability while requiring continuous enhancement through feature vector incorporation and weight adjustment for complementary improvement. Feature vectors comprising operational scenario parameters, node characteristics, and message transmission patterns, were utilized to train AI models capable of predicting the per-node timeslot allocation. The proposed approach provides proactive support for JTDLS network administration by enabling faster and more adaptive timeslot allocation compared with experience-based allocation, thereby minimizing overallocation and reducing transmission delays for mission-critical tactical messages.

External IDs:doi:10.1109/access.2025.3642558