Beam Management Model Monitoring Using Similarity Reporting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The traditional approach for AI/ML model monitoring in beam management requires terminal devices to measure and report a large amount of beam measurement reference signals, leading to significant overhead, and the model's performance deteriorates when field data differs significantly from training data.
Innovation Solution
The terminal device calculates and reports similarity information between field and training beam data, reducing the need for extensive beam measurements and reporting, and allowing the network device to assess AI/ML model performance indirectly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the traditional approach is used to monitor AI/ML model performance by comparing predicted and actual beam information, then the model performance can be assessed, but the terminal device must measure and report a large amount of beam measurement reference signals, causing huge overhead
Solution Approach 1:
The patent extracts only the essential similarity information from the full beam measurement data. Instead of measuring and reporting all beam measurement reference signals, the terminal device calculates a compact similarity metric between training data and field data, extracting only the necessary information for model performance assessment while discarding redundant data.
Solution Approach 2:
The patent inverts the traditional monitoring approach by not directly comparing predicted and actual beam information, but instead comparing the similarity between training data distribution and field data distribution. This indirect approach through similarity calculation reduces the measurement and reporting burden while still enabling model performance assessment.
2Measurement precision
If the terminal device measures a large amount of beam measurement reference signals to ensure accurate model monitoring, then the model performance assessment becomes reliable, but the complexity and overhead of the measurement and reporting process increases significantly
Solution Approach 1:
The patent creates a simplified copy of the beam measurement process by using similarity calculation between training data and field data. Instead of performing full beam measurements and detailed comparisons, the system uses a compact similarity metric that copies the essential information needed for model monitoring while avoiding the complexity of complete measurement and reporting procedures.
3Reliability
If the AI/ML model is trained on historical data, then the model can provide accurate predictions, but the model performance deteriorates when field data differs significantly from training data
Solution Approach 1:
The patent implements a feedback mechanism by continuously calculating the similarity between training data distribution and field data distribution. When the similarity falls below a threshold, indicating significant distribution shift, the system triggers model retraining or updating, providing feedback that adapts the model to current field conditions and maintains prediction accuracy.
Solution Approach 2:
The patent introduces dynamics to the model monitoring process by making it adaptive to changing field conditions. Instead of static model deployment, the system dynamically assesses data distribution similarity and triggers model updates when necessary, allowing the system to adapt to evolving environmental conditions and maintain reliability.
Data Source
AI summary
Example embodiments of the present disclosure relate to methods, devices, and computer storage medium for communication. The method comprises: receiving, at a terminal device, at least one first set of Reference Signals (RSs) from a network device, the network device being deployed with at least one Artificial Intelligence/Machine Learning (AI/ML) model, and each of the at least one first set of RSs corresponding to one of the at least one AI/ML model; calculating, at the terminal device, at least one similarity based on the at least one first set of RSs; determining, at the terminal device, at least one similarity information based on the at least one calculated similarity, and at least one model information corresponding to the at least one similarity information, the at least one model information indicating an index of at least one AI/ML model; and transmitting, to the network device, at least one of: the at least one determined similarity information or the at least one model information.