Terminal Information Compression for AI Training Overhead Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In wireless communication systems, the use of artificial intelligence (AI) or machine learning (ML) methods requires a large amount of data for training, leading to significant air interface overheads due to the need for extensive data collection and reporting.
Innovation Solution
A method and apparatus for compressing and reporting information related to training data, including target information, measurement information, and neural network use information, using a preset compression manner to reduce air interface overheads, with different compression methods and conditions applied based on specific application scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If AI or ML methods are used to replace modules in wireless communication systems, then system performance is improved, but air interface overhead increases due to large amount of training data collection and reporting requirements
Solution Approach 1:
The patent extracts and reports only the most critical and changed parameters from the full measurement dataset. Instead of reporting all training data, the system identifies key parameters that significantly impact AI model performance and reports only those, thereby reducing air interface overhead while maintaining system performance benefits.
Solution Approach 2:
The patent applies different reporting strategies to different parameters based on their importance and characteristics. Critical parameters are reported with higher granularity and frequency, while less important parameters are reported with lower granularity. This localized quality approach optimizes the balance between performance and overhead by tailoring reporting to specific parameter needs.
2Measurement precision
If comprehensive measurement information is collected for AI training, then model accuracy is improved, but signaling overhead and transmission resources increase
Solution Approach 1:
The patent implements partial action by reporting only a subset of measurement parameters that are most critical for AI model training. Instead of transmitting complete measurement datasets, the system selectively reports key parameters such as channel state information, interference measurements, and quality metrics, achieving sufficient model accuracy with reduced signaling overhead.
Solution Approach 2:
The patent segments the measurement information into different categories and prioritizes reporting of high-importance segments. Measurement data is divided into critical parameters (reported with high precision and frequency), important parameters (reported with moderate precision), and auxiliary parameters (reported with low precision or aggregated), thereby optimizing the information-transmission efficiency.
3Reliability
If detailed target information and measurement data are reported, then AI model training quality is improved, but air interface resource consumption increases
Solution Approach 1:
The patent dynamically changes reporting parameters based on system conditions, parameter importance, and model training requirements. Reporting frequency, precision, and granularity are adjusted according to the criticality of each parameter and current network conditions. This parameter adaptation ensures high training quality while optimizing air interface resource utilization and reducing transmission complexity.
Data Source
AI summary
This application discloses an information processing method and apparatus, and a terminal. The method in embodiments of this application includes: A terminal compresses first information in a preset compression manner, and reports compressed first information, where the first information includes at least one of the following: first target information; representation information of the first target information; measurement information corresponding to the first target information; neural network use information corresponding to the first target information; information indicating whether the first target information meets a first condition; value information corresponding to the first target information; and first target information that meets the first condition, where the first target information includes at least one of the following: position information of the terminal; first measurement quantity information; first event information; first identifier information; first transmission parameter information; application layer configuration information; and first configuration information.