UE Ground Truth Grouping for Low-Overhead AI/ML Reporting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current communication systems face high resource and signaling overhead due to frequent reporting of ground truth from UEs to base stations, which is inefficient and burdens the system, and there is a need for improved quantization methods to reduce this burden.
Innovation Solution
Implementing a method for group reporting of ground truth using channel state information reference signal (CSI-RS) measurement occasions, allowing periodic, semi-persistent, or aperiodic reporting of grouped ground truths on an uplink channel, along with innovative quantization techniques to reduce reporting content size and overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If ground truth is reported frequently from UE to gNB, then model training accuracy is improved, but system overhead and resource consumption increase
Solution Approach 1:
Multiple ground truth measurements are grouped together and reported as a single batch rather than individually. The UE collects ground truth data from multiple CSI measurement occasions and transmits them collectively, reducing the number of separate reporting events and associated signaling overhead while maintaining sufficient data for model training.
Solution Approach 2:
Ground truth reporting is performed periodically at configured intervals rather than continuously or on every measurement occasion. The gNB configures periodic reporting parameters, and the UE reports ground truth data at these periodic intervals, balancing the need for accurate model training with reduced system overhead and resource consumption.
2Loss of information
If detailed ground truth data is reported, then model training quality is improved, but signaling overhead increases
Solution Approach 1:
Only the essential ground truth information needed for model training is extracted and reported, rather than transmitting all raw measurement data. The UE processes the measured data and extracts key ground truth parameters that are sufficient for AI/ML model training, reducing the volume of signaling while maintaining training quality.
Solution Approach 2:
The ground truth data undergoes parameter transformation including quantization and encoding before reporting. The UE applies quantization to convert continuous measurement values into discrete levels, and applies encoding schemes to compress the data representation, reducing signaling overhead while preserving the essential information content for model training.
3Measurement precision
If ground truth reporting is performed individually for each measurement, then data accuracy is maintained, but scheduling overhead increases
Solution Approach 1:
Multiple individual ground truth measurements are merged into a single grouped report. Instead of scheduling separate transmission resources for each measurement, the UE aggregates multiple measurements and transmits them together in one scheduled event, significantly reducing the scheduling overhead and resource allocation complexity.
Solution Approach 2:
The UE performs preliminary data collection and grouping before the actual reporting event. Ground truth measurements are accumulated in a buffer over multiple measurement occasions, and the grouping is prepared in advance, allowing the scheduling system to handle fewer, larger reporting events rather than numerous small individual reports.
Data Source
AI summary
Communication devices and methods for reporting of grouped ground truths for artificial intelligence (AI)/machine learning (ML) are disclosed. The method for reporting of grouped ground truths for AI/ML performed by a user equipment (UE) includes determining, by the UE, channel state information reference signal (CSI-RS) measurement occasions to collect ground truths for generating grouped ground truth report for an AI/ML and reporting, to a base station, the grouped ground truths for the AI/ML in a periodic manner, a semi-persistent manner, and/or an aperiodic manner on an uplink channel.