Augmenting Channel Data for Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for training artificial neural networks in mobile communication systems require independent channel data for each antenna configuration, leading to increased costs due to the need for multiple measurements and data collection processes.
Innovation Solution
A method where a base station delivers augmentation assistance information to a terminal, allowing the terminal to generate augmented channel data based on a correspondence relationship between antenna configurations, which is then used for training artificial neural networks, reducing the need for multiple measurements and data collection processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If independent channel data is collected for each antenna configuration, then training data accuracy is improved, but measurement costs and data collection complexity increase
Solution Approach 1:
The patent creates virtual channel data copies for different antenna configurations by applying correspondence relationships to a single measured channel data set. Instead of physically measuring each antenna configuration separately, the system generates synthetic training data through mathematical transformations, reducing measurement complexity while maintaining training data quality
Solution Approach 2:
The system pre-establishes correspondence relationships between different antenna configurations in advance. These predefined transformation rules allow the terminal to generate augmented channel data without performing repeated measurements, thereby reducing real-time data collection complexity while ensuring accurate training data generation
2Reliability
If multiple independent artificial neural networks are constructed for different antenna configurations, then model accuracy is improved, but training costs and system complexity increase
Solution Approach 1:
The patent enables a single artificial neural network to handle multiple antenna configurations by using augmented channel data from correspondence relationships as training input. The network learns universal patterns that generalize across different antenna setups, eliminating the need to construct and maintain multiple separate networks while preserving accuracy
Solution Approach 2:
The system merges training data from multiple antenna configurations into a unified training set through data augmentation. By combining correspondence-based generated data with actual measurements, the system creates a comprehensive training dataset that allows one neural network to learn from diverse antenna scenarios, reducing system complexity
3Quantity of substance
If multiple measurements are performed for different antenna configurations, then data completeness is improved, but time consumption and cost increase
Solution Approach 1:
The system generates complete training datasets by creating virtual copies of measured channel data through correspondence relationships. This approach ensures data completeness for all antenna configurations without requiring proportional increases in measurement time, as synthetic data generation is computationally efficient
Solution Approach 2:
The system changes the parameter representation of channel data by applying transformation rules that map between different antenna configurations. By modifying data parameters through mathematical relationships rather than physical remeasurement, the system achieves complete training data coverage with minimal time investment
Data Source
AI summary
A method of a terminal may comprise: receiving augmentation assistance information from a base station; generating augmented channel data by applying the augmentation assistance information to original channel data; and training an artificial neural network using the augmented channel data.