Channel Data Validation for AI Model Training in Wireless Communication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Wireless communication systems, particularly at the physical layer, are significantly affected by external environments, leading to low-quality channel data with high noise levels, which can adversely affect the performance of AI models trained using such data.
Innovation Solution
A communication method and device that determine the validity of channel data based on metrics like RSRP, RSRQ, and power distribution in the delay domain, filtering out low-quality data to ensure only high-quality data is used for model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If channel data is collected directly from the physical layer, then data quantity is sufficient, but data quality is low with high noise levels
Solution Approach 1:
The patent applies preliminary action by performing validity judgment on channel data before using it for model training. The network device or terminal device evaluates whether channel data meets validity conditions (such as RSRP thresholds, RSRQ thresholds, or power distribution characteristics) before incorporating it into training datasets. This preliminary filtering ensures that only high-quality data proceeds to the training stage, resolving the contradiction between having sufficient data quantity and ensuring high data quality.
2Ease of manufacture
If low-quality channel data is used for model training, then training process is simple, but model performance is affected
Solution Approach 1:
The patent applies the extraction principle by separating valid channel data from invalid channel data through validity judgment. The system extracts only the portion of channel data that meets predefined validity conditions (such as power distribution characteristics, RSRP/RSRQ thresholds) and uses only this extracted valid data for model training. This extraction process maintains training simplicity while significantly improving model performance by eliminating noisy, low-quality data points.
3Productivity
If all channel data is used for training, then data utilization is high, but training effect is reduced due to noise
Solution Approach 1:
The patent applies local quality by treating different portions of channel data differently based on their validity. Instead of uniformly using all channel data, the system performs local quality assessment through validity judgment on individual channel data samples or groups. Valid data samples (those meeting power distribution, RSRP, or RSRQ conditions) are selected for training, while invalid samples are excluded. This localized quality control approach maximizes the utilization of high-quality data while filtering out noise, thereby improving training effect.
Data Source
Figure 1~2
Figure 3A~3B
Figure 4
AI summary
A communication method and a communication device are provided. The method includes the following. A first communication device determines whether first channel data is valid or invalid, where the first channel data is usable for training a first model when the first channel data is valid, and the first channel data is unusable for training the first model when the first channel data is invalid. Data used to train the first model is filtered data. The first channel data can be used to train the first model only when the first channel data is valid. It can be understood that for the training of the first model, the quality of valid channel data is higher than that of invalid channel data. Therefore, based on the method provided in this disclosure, low-quality parts in the channel data can be effectively filtered out, and the first model can be trained using higher-quality channel data, thereby improving the training effect and performance of the first model.