Dynamic User Grouping based on Location and Heading in 5G NR Systems

TL;DR

This work demonstrates how Sounding Reference Signals channel fingerprints can be used for dynamic user grouping in a 5G NR commercial deployment based on outdoor positions and heading direction employing machine learning methods such as neural networks combined with clustering methods.

Abstract

User grouping based on geographic location in fifth generation (5G) New Radio (NR) systems has several applications that can significantly improve network performance, user experience, and service delivery. We demonstrate how Sounding Reference Signals channel fingerprints can be used for dynamic user grouping in a 5G NR commercial deployment based on outdoor positions and heading direction employing machine learning methods such as neural networks combined with clustering methods.

Figures (8)

• Figure 1: The 5G NR base station was installed on top of a 20-meter-high building. During this measurement campaign, a test vehicle traversed two predefined routes: A 10-meter-high garage path for LoS measurements and a ground-level path for NLoS measurements below the base station building. Each route features four different movement patterns. Thinner lines depict random trajectories.
• Figure 2: SRS data stream collection and pre-processed CTF dataset.
• Figure 3: The architecture of the ML-driven grouping framework along with the input pipeline and intended output. The output of the positioning model is averaged with the previous prediction as a post-processing step and normalized before forwarding it to the clustering algorithms for user grouping.
• Figure 4: Positioning and COG errors.
• Figure 5: LoS clustering results based on position