Data-driven Approach for Optimising Resource Allocation of O-RAN Networks

Radio Access Network (RAN) deployments are evolving quickly owing to the innovative approaches of the Open Radio Access Network (O-RAN) Alliance. Specifically, they are moving away from closed, customized hardware implementations and toward virtualized instances operating on shared platforms. Future successful and affordable RAN deployments are made possible by this paradigm change, which is characterised by the separation of radio software components from hardware. Real-time network parameter configuration, sufficient computing resources for virtualized RAN (vRAN) deployment, and dependable processing unit sharing among numerous vRAN instances are some of the obstacles still standing in the way of successful O-RAN network implementations. Thus, this paper explored and compared the effectiveness of diverse optimization algorithms for minimising the number of resource blocks (nRBS), including machine learning (RandomForestRegressor), heuristic, and mathematical methods. Moreover, it investigates the lessons learned and the limitations of the proposed system. It demonstrates the practical success of a heuristic approach in O-RAN optimization, achieving significant reductions in resource blocks based on the Throughput-to-Bandwidth Ratio. It also provided insights into challenges with the RandomForestRegressor model, highlighted the importance of considering real-world network dynamics, and offered valuable lessons for future research, emphasizing the need for adaptive solutions and exploring hybrid optimization approaches, ultimately contributing to an enhanced understanding of O-RAN optimization.