Optimizing Beamforming for Efficient User Equipment Association with Base Stations within an Open RAN Compliant Network

Abstract

In the rapidly evolving field of fifth-generation (5G) and beyond wireless communication, optimizing signal reception and beamforming is crucial for enhancing network efficiency. Despite advancements in this domain, there remains a gap in effectively predicting and managing signal strength across different user locations and frequencies. This project aims to develop an efficient Open Radio Access Network (O-RAN) xApp using machine learning (ML) models to connect users to the strongest beam. Specifically, the models recommend the four strongest beams between the two closest towers, dynamically adjusting to the movements of users. Utilizing a dataset with 50 original user locations, 11 frequency bands, and two beam stations, the models were trained to predict the optimal signal pathways. The K-Means clustering combined with Random Forest (RF) model achieved a high silhouette score of 172.34 and a cluster accuracy of 94.55%, with a mean squared error (MSE) of 0.00134. In comparison, the K-Nearest Neighbors (K-NN) model yielded a silhouette score of 0.368 and demonstrated rapid clustering and search times. These results highlight the superior accuracy of the K-Means + RF model, while the K-NN model excels in computational efficiency. The findings suggest that incorporating advanced ML techniques can significantly enhance signal optimization processes in 5G and beyond wireless communication networks, offering a practical tool for real-time network management and improved user experiences within the O-RAN architecture.