SON-RRM Intelligent Optimization Based on Cross-Layer Deep Reinforcement Learning

Self-Organizing Network (SON) is a promising and enabling technology for network automation that is an important component of 6G native-intelligence. Coverage and Capacity Optimization (CCO) is one of key use cases to self-optimize users’ experience. However, the temporal granularity of CCO as a network-level optimization is comparatively larger when contrasted with that of Radio Resource Management (RRM). Moreover, the Key Performance Indicators (KPIs) and Network Configuration Parameters (NCPs) to be adjusted may overlap between both optimization techniques. To further mitigate Operation, Administration, and Maintenance (OAM) cost and address conflicts arising from different optimization techniques, this study initially presents a framework for SON-RRM intelligent optimization. Subsequently, a cross-layer Deep Reinforcement Learning approach (cl-DQN) is proposed to address the challenge of delineating the optimization tasks of SON-RRM with different temporal granularities, taking CCO as an use case. The simulation results demonstrate that the proposed cl-DQN based CCO-RRM intelligent optimization approach outperforms traditional CCO, with 6% capacity as well as 4% coverage & capacity improvements respectively. Additionally, it reduces the OAM cost stemming from antenna parameter adjustments, with a maximum reduction of 93%.