Wireless Quality Estimation via Transfer Learning With Sparse Terminal Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting wireless quality in wireless communication systems, such as those used in autonomous driving, face challenges in accurately estimating quality without sufficient measurement data under desired conditions, leading to delayed responses and inaccurate predictions.
Innovation Solution
A learning method utilizing transfer learning to create a pre-trained model from a large dataset of various terminals, followed by retraining with data from the specific terminal to be used, allowing for accurate wireless quality estimation with minimal measurement data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If measurement data is collected from multiple terminals to improve model accuracy, then estimation accuracy is improved, but the complexity of data collection and processing increases
Solution Approach 1:
The patent creates a virtual copy of the terminal's measurement environment by collecting data from multiple terminals with similar characteristics. This allows the system to train a comprehensive model without physically deploying measurements from every possible terminal configuration, thereby improving accuracy while managing complexity through virtual representation.
Solution Approach 2:
The patent segments the data collection process by categorizing terminals into different groups based on their characteristics (e.g., terminal type, location, usage pattern). This segmentation allows the system to collect and process data from a manageable subset of terminals that represent different scenarios, reducing overall complexity while maintaining estimation accuracy across diverse conditions.
2Measurement precision
If measurement data is collected under conditions close to actual usage to improve accuracy, then estimation accuracy is improved, but the quantity of available data decreases
Solution Approach 1:
The patent makes the measurement data collection system universal by designing a framework that can gather data from any terminal under various conditions and still maintain relevance for the target terminal. The model is trained to handle data from different terminal types and usage scenarios, allowing the same data collection infrastructure to serve multiple purposes and accumulate larger datasets without sacrificing accuracy.
Solution Approach 2:
The patent applies parameter changes by transforming and normalizing measurement data from different terminals and conditions to make them comparable and applicable to the target terminal. Through feature engineering and data transformation, the system can utilize a broader range of data sources while maintaining estimation accuracy, effectively increasing the quantity of usable data.
3Measurement precision
If a pre-trained model is created from diverse terminal data and then retrained with specific terminal data, then estimation accuracy for the target terminal is improved, but the time required for training increases
Solution Approach 1:
The patent performs preliminary action by creating a pre-trained model using data from multiple terminals before final training with the specific terminal's data. This pre-training phase establishes a solid foundation with general knowledge from diverse conditions, and the subsequent fine-tuning with specific terminal data builds upon this foundation rather than starting from scratch, significantly reducing the overall training time while maintaining high accuracy.
Solution Approach 2:
The patent applies partial action by focusing the final training phase only on the specific terminal's data rather than retraining the entire model from scratch. This selective retraining approach updates only the necessary parameters for the target terminal while preserving the beneficial general knowledge from pre-training, achieving high accuracy with reduced computational time and resources.
Data Source
AI summary
A learning method in which a learning device performs learning on a model for estimating wireless quality by machine learning, and the learning method includes a learning step of generating a pre-trained model by performing the learning on the model by using measurement data of wireless quality obtained by a plurality of terminals, and a retraining step of retraining the pre-trained model by using measurement data satisfying a predetermined condition.