- 96 views
- 101 downloads
Application of Machine Learning in RAN Evolution for new Generation of mobile networks
-
- Author(s) / Creator(s)
-
The telecommunication industry has experienced considerable improvement and changes during the past years. Many standards and protocols have been introduced and implemented. This revolution in Radio Access Networks (RAN) is known as GSM, UMTS, LTE, 5G, and now B5G networks. Satisfying the user demands and keeping the level of QoS and QoE within the acceptable range have always been challenges for internet service providers and telecommunication operators. The researchers and studies are ongoing to address these massive requests and users' tendency to achieve reliable, low latency, and high throughput services.
oftware Define Networking (SDN), Network Virtualization Function (NVF), Self-Organizing Networks (SON), and increasing capacity solutions (mmWave communication, Massive MIMO, Network Slicing, Beamforming, and RAN Evolutions) are the main proposed and implemented solutions during the past decade in 5G networks. However, whenever we talk about the data, we will see the brilliant role of machine learning.
In this study, we have researched and implemented machine-learning algorithms in new evolutions of RAN. We can mention RAN evolution as Distributed-RAN, Cloud-RAN, Virtual-RAN, and now Open-RAN. Open RAN is a novel method of setting up and running wireless networks Using standardized, interoperable hardware and software components. Instead of being dependent on the proprietary technology of a single vendor, an open RAN architecture separates the radio access network into interchangeable, functional components.
The proposed scenario uses supervised-learning algorithms to make predation (classification) of services and slices in Open-RAN 5G networks. This AI/ML scenario is implemented in the RIC (Radio Intelligent Controller) block of O-RAN, and we have evaluated and compared the performance of five different supervised-learning algorithms. A novel method based on hyperparameters tuning and K-fold cross-validating is proposed for Random Forest Algorithm. This technique will improve the classified results compared to the introduced baselines. The algorithms' training phase utilizes the KPI and KQI data of a 5G network. Moreover, simulation results prove that considering both KPI and KQI will improve the results compared to only KPI scenarios. -
- Date created
- 2023-04-01
-
- Subjects / Keywords
-
- Type of Item
- Report