Radio Configuration Control with UE Clustering and Dual Optimization

The patent segments the radio configuration parameter optimization into two distinct parts: (1) UE clustering using classification algorithms with predefined input parameters, and (2) radio configuration parameter optimization using Reinforcement Learning. This segmentation allows each algorithm to focus on specific tasks, reducing the overall complexity of input parameter selection while maintaining performance improvements.

The patent performs preliminary action by pre-defining the input parameters for the classification algorithm (such as path loss, interference level, and UE characteristics) before the Reinforcement Learning process begins. This preliminary preparation of structured input data simplifies the subsequent RL algorithm's task, reducing its complexity while preserving its ability to optimize performance.

The patent implements dynamics by using a Reinforcement Learning algorithm that continuously adapts radio configuration parameters based on real-time network conditions and UE behavior. The algorithm learns optimal parameter adjustments through interaction with the dynamic environment, maintaining robustness while managing computational complexity through efficient learning mechanisms.

The patent employs feedback mechanisms where the Reinforcement Learning algorithm receives performance metrics and network state information, adjusts parameters accordingly, and observes the outcomes. This closed-loop feedback enables adaptive parameter adjustment that responds to dynamic conditions while maintaining computational efficiency through learned policies rather than exhaustive optimization.

The patent segments the optimization process into two stages with different complexity requirements: (1) UE clustering using computationally efficient classification algorithms that group UEs based on key characteristics, and (2) parameter optimization using Reinforcement Learning that achieves high precision within each cluster. This segmentation allows low-complexity processing for clustering while preserving optimization precision through the subsequent RL stage.

The patent applies local quality by using simple classification algorithms for global UE clustering while deploying sophisticated Reinforcement Learning for local parameter optimization within each cluster. This approach matches the computational complexity to the required precision at each stage, achieving overall high optimization precision without requiring low-complexity algorithms throughout the entire process.