CSI Channel Expansion for Fixed-Length AI Feature Reporting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models for channel state information (CSI) in communication systems face challenges due to varying input data lengths, leading to performance issues and excessive storage and training overheads when adapting to different channel information lengths.
Innovation Solution
A terminal expands first channel information into second channel information of a fixed length, matching the input requirements of AI models, thereby reducing the need for multiple AI model training and minimizing storage and training overheads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If AI models are trained to handle varying input data lengths for channel state information, then the system can adapt to different channel conditions, but the storage and training overheads become excessive
Solution Approach 1:
The patent changes the parameter of input data length from variable to fixed by expanding channel information to a standard length of 64 sub-bands. This parameter change allows the AI model to be trained once with a fixed input format, eliminating the need for multiple models with different input lengths and significantly reducing storage and training overheads while maintaining adaptability through the expansion mechanism.
Solution Approach 2:
The patent creates a universal AI model that can handle all channel information lengths by expanding any input to the standard format of 64 sub-bands. This single universal model replaces the need for multiple specialized models, achieving multi-functionality across different channel conditions without requiring separate training for each scenario.
2Manufacturing precision
If multiple AI models are trained for different channel information lengths, then each model can be optimized for its specific length, but the training time and computational resources increase significantly
Solution Approach 1:
The patent standardizes the input parameter length to 64 sub-bands for all channel information, transforming the problem from training multiple models with different lengths to training a single model with a fixed input format. This dramatically reduces training time and computational resources while maintaining optimization effectiveness through the expansion process.
Solution Approach 2:
The patent performs preliminary expansion of channel information to the standard length of 64 sub-bands before feeding it into the AI model. This preliminary action ensures that the model only needs to be trained once with the standard format, avoiding the need for repeated training for different input lengths and significantly reducing overall training time.
3Quantity of substance
If channel information is expanded to fixed length to match AI model requirements, then model training overhead is reduced, but the processing complexity at the terminal increases
Solution Approach 1:
The patent applies partial expansion to channel information, transforming it from variable length to a fixed standard length of 64 sub-bands. This partial action (expanding only when necessary) reduces training overhead by creating a universal model format while the complexity is managed by implementing the expansion only at the terminal side rather than throughout the entire system.
Data Source
AI summary
This application relates to the field of communication technologies, and discloses an information processing method and apparatus, a terminal, and a network side device. The information processing method according to embodiments of this application includes: expanding, by a terminal, first channel information into second channel information, where the first channel information includes M1 sub-bands or N1 ports corresponding to each sub-band, the second channel information includes M2 sub-bands or N2 ports corresponding to each sub-band, M1≤M2, N1≤N2, and M1, M2, N1, and N2 are all positive integers; and processing, by the terminal, the second channel information into channel feature information, and reporting the channel feature information.