User Equipment Grouping in 5G TDD System using Machine Learning
(2024) EITM02 20241Department of Electrical and Information Technology
- Abstract
- User Equipment (UE) grouping entails categorizing multiple UEs, including mobile phones or smart devices, according to defined criteria. It provides valuable insights for applications to optimize network resources, enhance handover procedures, and improve user experience. This study investigates UE grouping in a Fifth Generation (5G) Time Division Duplex (TDD) system based on Uplink (UL) Sounding Reference Signal (SRS) channel fingerprints using Machine Learning (ML) techniques. Specifically, the study comprises two main blocks: UE position and direction estimation, and UE grouping.
In the first block, the estimation model for UE position and Course Over Ground (COG), the actual direction of motion, is developed. This model utilized UL... (More) - User Equipment (UE) grouping entails categorizing multiple UEs, including mobile phones or smart devices, according to defined criteria. It provides valuable insights for applications to optimize network resources, enhance handover procedures, and improve user experience. This study investigates UE grouping in a Fifth Generation (5G) Time Division Duplex (TDD) system based on Uplink (UL) Sounding Reference Signal (SRS) channel fingerprints using Machine Learning (ML) techniques. Specifically, the study comprises two main blocks: UE position and direction estimation, and UE grouping.
In the first block, the estimation model for UE position and Course Over Ground (COG), the actual direction of motion, is developed. This model utilized UL SRS channel estimation results collected from 5G TDD Base Station (BS) for diverse routes, including both Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) scenarios. Channel Transfer Function (CTF) snapshots are created according to system specifications using SRS data. To ensure the accuracy of the CTF when encountering missing SRS, we applied forward-filling to maintain data integrity. CTF snapshots are synchronized in time with Global Navigation Satellite System (GNSS) data. Supervised ML techniques, culminating in an ensemble model refined with post-processing methods, achieved highly precise estimation results. The positioning Root Mean Square Error (RMSE) is less than 0.93 meters and the direction RMSE is less than 9 degree across all routes.
In the second block, clustering algorithms are used to group UEs based on their estimated positions and directions. Firstly, the timestamps of the UEs' data are synchronized to assume they move simultaneously despite being measured at different times. Moreover, to ensure a fair assessment of clustering results, more UEs are needed than measured ones. Each route is then split into two halves, treating the first and second halves as distinct virtual UEs derived from real data measurements. Various clustering methods are applied based on the estimated features to see how models group UEs differently. It has been demonstrated that the sensitivity of the clustering models can be adjusted by modifying the initialization parameters to align with the specific criteria of various network functionalities. Additionally, the best beam prediction models are presented in the Appendix as an application area of SRS signals. (Less) - Popular Abstract
- Have you ever wondered how your smartphone knows precisely where you are? Imagine using your phone to stream a movie or video call a friend on a busy street. Why does the signal sometimes weaken, or does the video start buffering? Since many people use the same network simultaneously, it can get crowded up in the digital world! What if the network could adapt to the dynamic radio environment and provide optimal performance in various situations? This would ensure that your favorite shows do not pause at the most exciting moments or that you do not encounter connection issues during important online job interviews or meetings. We have an idea! We are using advanced computer techniques to figure out exactly where each user is and in which... (More)
- Have you ever wondered how your smartphone knows precisely where you are? Imagine using your phone to stream a movie or video call a friend on a busy street. Why does the signal sometimes weaken, or does the video start buffering? Since many people use the same network simultaneously, it can get crowded up in the digital world! What if the network could adapt to the dynamic radio environment and provide optimal performance in various situations? This would ensure that your favorite shows do not pause at the most exciting moments or that you do not encounter connection issues during important online job interviews or meetings. We have an idea! We are using advanced computer techniques to figure out exactly where each user is and in which direction they are facing. This way, we can ensure that everyone's signal stays strong and their videos keep streaming without pauses.
How does it work? Think of it as having a bunch of invisible helpers in the sky who can see where everyone is and where they are headed. These helpers, powered by machine learning, analyze all the data from base stations to create a digital map of the world around us. Here is the exciting part: Once we know where everyone is, we can start organizing things better. We can group users based on their locations and directions, similar to organizing cars on a busy highway to keep traffic flowing smoothly. This approach allows for strategic capacity planning and load balancing based on user population, enhancing overall user experience. Service providers can strategically plan the placement of new base stations based on user distribution to maintain service continuity effectively. With machine learning and sophisticated data analysis, 5G networks are getting smarter every day! This technology opens up a world of possibilities, and we invite you to be a part of it by delving into our thesis report, "User Equipment Grouping in 5G TDD System using Machine Learning", to understand our methodology for utilizing data received at the 5G base station and see how accurately we could estimate user positions and directions. Furthermore, we have explored various options for grouping users to optimize network performance. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9171753
- author
- Arslantürk, Korkut LU
- supervisor
-
- Xuesong Cai LU
- Dino Pjanić LU
- organization
- course
- EITM02 20241
- year
- 2024
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- MIMO, 5G New Radio, Sounding Reference Signal, Positioning, Course Over Ground, Machine Learning, User Equipment Grouping, Clustering
- report number
- LU/LTH-EIT 2024-1011
- language
- English
- id
- 9171753
- date added to LUP
- 2024-08-19 14:35:47
- date last changed
- 2024-08-19 14:35:47
@misc{9171753, abstract = {{User Equipment (UE) grouping entails categorizing multiple UEs, including mobile phones or smart devices, according to defined criteria. It provides valuable insights for applications to optimize network resources, enhance handover procedures, and improve user experience. This study investigates UE grouping in a Fifth Generation (5G) Time Division Duplex (TDD) system based on Uplink (UL) Sounding Reference Signal (SRS) channel fingerprints using Machine Learning (ML) techniques. Specifically, the study comprises two main blocks: UE position and direction estimation, and UE grouping. In the first block, the estimation model for UE position and Course Over Ground (COG), the actual direction of motion, is developed. This model utilized UL SRS channel estimation results collected from 5G TDD Base Station (BS) for diverse routes, including both Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) scenarios. Channel Transfer Function (CTF) snapshots are created according to system specifications using SRS data. To ensure the accuracy of the CTF when encountering missing SRS, we applied forward-filling to maintain data integrity. CTF snapshots are synchronized in time with Global Navigation Satellite System (GNSS) data. Supervised ML techniques, culminating in an ensemble model refined with post-processing methods, achieved highly precise estimation results. The positioning Root Mean Square Error (RMSE) is less than 0.93 meters and the direction RMSE is less than 9 degree across all routes. In the second block, clustering algorithms are used to group UEs based on their estimated positions and directions. Firstly, the timestamps of the UEs' data are synchronized to assume they move simultaneously despite being measured at different times. Moreover, to ensure a fair assessment of clustering results, more UEs are needed than measured ones. Each route is then split into two halves, treating the first and second halves as distinct virtual UEs derived from real data measurements. Various clustering methods are applied based on the estimated features to see how models group UEs differently. It has been demonstrated that the sensitivity of the clustering models can be adjusted by modifying the initialization parameters to align with the specific criteria of various network functionalities. Additionally, the best beam prediction models are presented in the Appendix as an application area of SRS signals.}}, author = {{Arslantürk, Korkut}}, language = {{eng}}, note = {{Student Paper}}, title = {{User Equipment Grouping in 5G TDD System using Machine Learning}}, year = {{2024}}, }