Generating Datasets for Real-Time Scheduling on 5G New Radio

Area of Science:

• Computer Science
• Electrical Engineering
• Wireless Communication

Background:

• 5G new radio (NR) systems offer advanced industrial wireless motion control.
• Existing real-time scheduling algorithms are incompatible with complex 5G NR scheduling models, limiting resource availability.
• Supervised learning shows promise for complex scheduling but lacks adequate training datasets for 5G NR.

Purpose of the Study:

• To develop methods for generating training datasets for 5G NR scheduling.
• To enable the application of supervised learning to optimize 5G NR resource allocation for industrial systems.

Main Methods:

• Proposed two methods: one based on optimization modulo theories (OMT) and another, more efficient, based on satisfiability modulo theories (SMT).
• The OMT method was optimized for fewer variables to speed up solver performance.
• The SMT method further reduced solution time by tightening the search space using specific theorems and an algorithm.

Main Results:

• The SMT-based method achieved a 74.7% reduction in solution time compared to existing approaches.
• A supervised learning model trained on the generated dataset demonstrated superior scheduling performance over traditional polynomial-time algorithms.

Conclusions:

• The developed OMT and SMT methods effectively generate training datasets for 5G NR scheduling.
• Supervised learning, powered by these datasets, offers a viable and high-performing solution for complex 5G NR scheduling challenges in industrial applications.