AI Channel Information Processing Across Variable Sub-Band Quantities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI network models for channel information processing require separate training and configuration for different quantities of sub-bands, leading to increased complexity and overheads in training and transmission.
Innovation Solution
Group frequency domain resources and process channel information of each group using corresponding AI network models, allowing a low quantity model to handle high quantity information, reducing the number of models and model size.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate AI network models are trained for different quantities of sub-bands, then channel information processing accuracy is improved, but device complexity and training overhead increase
Solution Approach 1:
The patent applies universality by designing a single AI network model that can process channel information for different quantities of sub-bands. The model uses dynamic input dimension adjustment where the number of input channels corresponds to the quantity of sub-bands, allowing one model to serve multiple functions across different configuration scenarios without requiring separate specialized models for each sub-band quantity.
Solution Approach 2:
The patent employs parameter changes by dynamically adjusting the input dimension parameter of the AI network model based on the quantity of sub-bands. Instead of training separate models, the system modifies the model's input channel dimension to match the actual number of sub-bands, enabling a single model to adapt to different processing requirements while maintaining accuracy.
2Measurement precision
If multiple AI network models are trained and transferred for different sub-band quantities, then processing accuracy is improved, but transmission overhead increases
Solution Approach 1:
The patent eliminates the need to transmit multiple specialized models by creating a universal model that can handle all sub-band quantities. This single model reduces transmission overhead significantly compared to transferring multiple different models, while still maintaining the ability to process channel information accurately for any given sub-band configuration.
Solution Approach 2:
The patent extracts the variable dimensionality requirement from the model structure itself and handles it through dynamic input processing. By separating the fixed neural network architecture from the variable input dimension, the system transmits only one model structure while adapting the input size at runtime, thereby reducing the quantity of model data that needs to be transmitted.
3Ease of manufacture
If AI network models are trained based on fixed input dimensions, then model training is simplified, but adaptability to different sub-band quantities is reduced
Solution Approach 1:
The patent resolves this contradiction by making the input dimension parameter variable rather than fixed. The model is trained with the capability to accept different numbers of input channels corresponding to different sub-band quantities. This allows the model to maintain training simplicity while achieving high adaptability, as the same trained model can process inputs of varying dimensions without requiring retraining.
Data Source
AI summary
This application discloses an information transmission method and apparatus, and a communication device. The information transmission method includes: a terminal determines K groups of second channel information from first channel information based on first information, where the first information includes group information of K groups of frequency domain resources, the K groups of second channel information are in a one-to-one correspondence with the K groups of frequency domain resources, each of the K groups of frequency domain resources includes at least one frequency domain resource; the terminal performs first processing on M groups of second channel information based on first AI network models respectively corresponding to the M groups of second channel information, to obtain M pieces of channel characteristic information; and the terminal sends second information to a network side device, where the second information includes the M pieces of channel characteristic information.