Self-Organizing Network Optimization via Monte Carlo Tree Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge in mobile communications networks is to efficiently optimize coverage and capacity in Self-Organizing Networks (SON) without systematically exploring the vast and exponentially growing number of possible configurations, which is an NP-hard combinatorial optimization problem, especially when using model-free Reinforcement Learning approaches that require fast but inaccurate simulators.
Innovation Solution
A method and system that classify network cells into modifiable and non-modifiable classes, using a Monte Carlo Tree Search algorithm with an accurate network simulator to explore and evaluate different configurations, predicting network performance indicators with a machine learning model, and deploying optimized configurations while maintaining network operation, leveraging an Upper Confidence Bound criterion and a regressor for parameter adjustments like antenna tilts and power transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If model-free Reinforcement Learning approaches are used for optimization, then the speed of optimization is improved, but the accuracy of simulation deteriorates
Solution Approach 1:
The patent introduces a simulator as an intermediary component between the reinforcement learning agent and the environment. This simulator acts as a mediator that provides accurate environmental dynamics and performance evaluation, enabling the RL algorithm to learn from high-fidelity simulations without requiring real-time interactions with the actual network, thus resolving the contradiction between optimization speed and simulation accuracy
Solution Approach 2:
The patent performs optimization simulations in advance using accurate network simulators before deploying configurations to the live network. By conducting preliminary optimization iterations in a simulated environment with high accuracy, the system can identify optimal configurations without compromising real-time network operations, thereby achieving both speed and accuracy
2Measurement precision
If a vast number of configurations are systematically explored, then the optimization accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent implements feedback mechanisms where the accurate simulator evaluates configuration performance and provides reward signals to the reinforcement learning algorithm. This feedback loop enables the system to iteratively improve optimization accuracy by learning from simulation results, while the RL algorithm's efficient exploration strategies prevent exhaustive search of all possible configurations, managing computational complexity
Solution Approach 2:
The patent optimizes network parameters such as antenna tilts and power transmission by making targeted changes based on reinforcement learning decisions rather than systematically exploring all parameter combinations. The accurate simulator evaluates only the most promising parameter changes, achieving high optimization accuracy without the computational burden of exhaustive configuration exploration
3Measurement precision
If real-time interactions with accurate simulators are performed, then the simulation accuracy is improved, but the time consumption increases
Solution Approach 1:
The patent performs all optimization simulations and evaluations in advance using accurate network simulators before deploying configurations to the live network. By conducting comprehensive simulations preliminarily, the system achieves high simulation accuracy without requiring real-time interactions during network operations, thus avoiding time consumption delays in production environments
Solution Approach 2:
The simulator serves as an intermediary that enables offline optimization by providing accurate environmental modeling and performance evaluation. This intermediary allows the reinforcement learning algorithm to learn optimal policies through extensive simulations without time constraints, and the learned policy can then be deployed to the actual network without requiring continued real-time simulation interactions
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method, implemented by a data processing system (125), of adjusting modifiable parameters of network cells (110) of a self-organizing cellular mobile communications network, comprising: retrieving (605) a current configuration (205) of network cells (110, 115) currently deployed on field, including modifiable parameters; exploring (610, 620) different configurations of the network cells, each differing from the retrieved current configuration and from other different configurations by a change in the value of at least one of the modifiable parameters of at least one network cell; evaluating (625) the explored different configurations of the network cells. When a suitable goodness is assessed for a certain configuration of network cells, said new configuration of network cells is automatically deployed on field by modifying one or more of the modifiable parameters of network cells.