A Predictive Load Balancing Optimization Framework for Open Radio Access Networks via Xn Protocol

Open radio access networks (ORANs) implement a new paradigm based on the Xn protocol (high-level C/U split) that improves interoperability between open distributed units (O-DUs), open central units (O-CUs) and open radio units (O-RUs). In this paper, some of the most critical issues such as heterogeneous user (UE) requirements and preferences, limited resources, dynamic variation in the conditions of the network, and inherent interference in the ORAN are solved by optimizing load balancing and resource allocation. We focus in this paper on using Xn interface for load balancing without relying on external algorithms and techniques for efficient load sharing and coordination among next generation node base stations (gNBs) and to promote multi-vendor intercommunication, not to mention reducing the overhead costs. The predictive load balancing was represented in the form of a weighted metric via several scenarios of multi-objectives optimization problems to maximize network efficiency while improving energy efficiency (EE), minimizing delay, and maximize bandwidth utilization. Numerical simulations based on UEs and gNBs validate the efficiency of these models, demonstrating the enhanced quality of service (QoS). Lagrange multiplier method with standard constrained numerical optimization solvers including sequential quadratic programming (SQP), active set and interior point methods has been used to solve the formulated optimization problems. These solvers are used to validate the effectiveness of the proposed frameworks on scenarios with different allocation approaches (i.e., static, proportional, priority-based, and priority-weighted proportional). The analysis of these allocations highlights the balance between resource fairness and user satisfaction by achieving load distribution among O-DUs and virtual machines (VMs), resulting in reduced load variance and enhanced fairness. Subsequently, applicatio-level QoS measurements, such as throughput, latency, and EE shows that UEs are satisfied with fluctuating networks. Last, the proposed Xn-based models are compared with a classical load balancing model, namely autoregressive integrated moving average (ARIMA) model.