Machine-Trained Radio Access Network Configuration Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wireless network planning in production environments is complex and inefficient, often relying on heuristics due to the complexity of mixed-integer linear programming, which limits the number of parameters considered and does not scale well with the number of configuration parameters, leading to suboptimal radio access network configurations.
Innovation Solution
A method utilizing a machine-trained radio access network configuration model, specifically a convolutional neural network, to determine optimal configurations for radio access nodes by propagating environmental representations and performance indicators, constraining deployment locations and improving communication characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If mixed-integer linear programming is used for network optimization, then configuration accuracy is improved, but computational complexity increases and scalability deteriorates
Solution Approach 1:
The patent replaces traditional mixed-integer linear programming (mathematical optimization) with a machine learning model (convolutional neural network). This substitution transforms the complex computational optimization problem into a pattern recognition task, where the pre-trained model directly predicts optimal configurations from environment representations, avoiding the need for real-time complex calculations.
Solution Approach 2:
The patent performs optimization work in advance by pre-training the machine learning model on extensive simulation data covering various production environments. This preliminary action allows the model to learn optimal configuration patterns beforehand, enabling fast deployment without performing complex optimization during actual network setup.
2Reliability
If the number of configuration parameters is increased, then network performance is improved, but problem complexity increases and heuristic limitations arise
Solution Approach 1:
The patent creates a universal machine learning model that handles multiple configuration parameters (antenna patterns, frequency bands, power settings, etc.) simultaneously through a single integrated framework. The convolutional neural network processes all parameters together, eliminating the need for sequential heuristic optimization and enabling comprehensive parameter optimization without exponential complexity increases.
3Ease of manufacture
If iterative heuristic approaches are used for network planning, then implementation simplicity is maintained, but configuration optimality deteriorates and manual overhead increases
Solution Approach 1:
The patent implements an automated system where the machine learning model independently generates optimal network configurations without requiring manual iterative adjustments. The system takes environment specifications as input and autonomously produces complete configuration sets, eliminating the need for repeated manual optimization cycles and expert intervention.
4Power
If sequential parameter optimization is performed, then computational load per iteration is reduced, but total configuration time increases and productivity decreases
Solution Approach 1:
The patent merges all configuration parameter optimizations into a single unified prediction step performed by the machine learning model. Instead of sequentially optimizing parameters one by one or in small groups, the model processes all parameters simultaneously in one forward pass, dramatically reducing total configuration time while maintaining comprehensive optimization.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
There is provided a method comprising: determining (102) a representation of a production environment, wherein the representation comprises at least a plurality of spatial positions associated with a presence of a respective production equipment; determining (104) at least one performance indicator that indicates a performance requirement for reception or transmission of radio signals by at least one of a plurality of radio access nodes or by at least one of a plurality of radio terminals that is associated with the respective production equipment; and propagating (106) the representation and the at least one performance indicator through a machine-trained radio access network configuration model, wherein the representation and the at least one performance indicator are provided at an input section of the machine-trained radio access network configuration model; wherein at least one configuration for a radio access network is provided at an output section of the machine-trained radio access network configuration model; and wherein the at least one configuration of the radio access network comprises at least one or a set of configurations of radio access nodes.