AI Beam Selection With Reduced Measurement Reporting Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing beam management (BM) systems face high overhead in beam measurements and reporting due to the use of AI/ML model inference, particularly when predicting beams that are not part of the configured set, leading to inefficient communication processes.
Innovation Solution
The solution involves determining a subset of beams for measurements based on an applied beam, reducing the number of beams to be measured and reported, and enhancing beam reporting by indicating whether predicted beams are within the configured set using AI/ML model inference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI/ML model inference is used for beam prediction on a set of configured beams, then beam selection accuracy is improved, but measurement and reporting overhead increases
Solution Approach 1:
The patent segments the beam management process into two distinct phases: (1) a training phase where the network device trains an AI/ML model using measurement results from a first set of beams, and (2) an inference phase where the terminal device uses the trained model to predict beams in a second set. This segmentation allows the network to leverage AI/ML for accurate predictions while the terminal performs only lightweight inference, significantly reducing measurement and reporting overhead at the terminal side.
2Adaptability or versatility
If the terminal device performs model inference on all configured beams, then comprehensive beam selection is achieved, but processing complexity and time increase
Solution Approach 1:
The patent applies preliminary action by having the network device pre-train the AI/ML model using comprehensive measurement data from multiple beams before deployment. The trained model encapsulates complex patterns and relationships, allowing the terminal device to perform simple inference operations without needing to process all the original training data. This preliminary training action transfers the computational burden from the terminal to the network, reducing terminal complexity while maintaining comprehensive beam selection capability.
3Measurement precision
If beam measurements are performed on a large set of beams, then beam prediction accuracy is improved, but measurement time and energy consumption increase
Solution Approach 1:
The patent uses copying by creating a trained AI/ML model that replicates the complex relationships between beams observed during the training phase. Instead of repeatedly performing comprehensive measurements, the system creates a virtual copy of the beam environment through the trained model that can be efficiently queried by the terminal. This model copy allows accurate predictions without requiring the terminal to perform time-consuming measurements on all beams during operation.
Data Source
AI summary
Embodiments of the present disclosure relate to methods, devices and computer readable media of communication. A terminal device determines an applied beam and determines, based on the applied beam, a subset of beams in a first set of beams for measurements, the subset of beams being used for model inference for beam selection from a second set of beams. In this way, the number of beams to be measured and reported may be reduced and overhead for beam measurements and reporting may be reduced accordingly.