Neural Network CSI Measurement Overhead Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional beam measurement and reporting in NR CSI frameworks incur a significant time-frequency-resource overhead, reducing system efficiency by consuming resources needed for data transmission.
Innovation Solution
Implementing machine learning-based methods where a terminal device uses neural networks to measure and report CSI-RS resources with reduced overhead, selecting suitable beams and reporting results efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional beam measurement and reporting methods are used in NR CSI frameworks, then beam quality can be measured and reported, but significant time-frequency-resource overhead is incurred, reducing system efficiency
Solution Approach 1:
The patent extracts only the essential measurement information from full beam measurements. Instead of measuring and reporting all N CSI-RS resources, the system measures M resources (where M < N) and uses neural network inference to estimate the remaining measurements, thereby extracting only the critical measurement data needed for beam selection while discarding redundant measurements.
Solution Approach 2:
The patent applies preliminary action by using neural network models that are pre-trained to predict beam measurement results. The neural network is configured with pre-established weight parameters that enable it to infer measurement results of unmeasured CSI-RS resources based on measured resources, allowing the system to prepare accurate beam selection data without performing exhaustive measurements.
2Loss of information
If conventional beam measurement methods measure all N CSI-RS resources, then complete beam quality information is obtained, but resource overhead increases and system efficiency decreases
Solution Approach 1:
The patent introduces a neural network as an intermediary between partial measurements and complete beam selection decisions. The neural network receives measurements from M CSI-RS resources and produces inferred measurements for all N resources, acting as a mediator that reconstructs complete measurement information from partial data without requiring exhaustive measurements.
Solution Approach 2:
The patent changes the measurement parameter from measuring all N resources to measuring only M resources (where M < N). By changing the measurement sampling rate and using neural network inference to compensate, the system maintains measurement information quality while reducing the number of actual measurements performed, thereby improving system efficiency.
3Loss of time
If machine learning-based methods are used for beam measurement, then resource overhead is reduced, but the complexity of the measurement process increases due to neural network configuration
Solution Approach 1:
The patent implements self-service by enabling the terminal device to autonomously perform neural network-based beam measurement and reporting. The terminal device is configured with the neural network model and automatically performs measurements on M resources, infers results for all N resources using the neural network, and generates reports without requiring complex network-side processing or manual intervention, thereby reducing overhead while managing complexity at the device level.
Data Source
AI summary
Provided is a method for machine learning based CSI measurement and reporting. The method includes: receiving, by a terminal device, configuration information of a set of “N” CSI-RS resources; receiving, by the terminal device, “M” CSI RS resources out of the “N” CSI RS resources; performing, by the terminal device, a measurement on the “M” CSI-RS resources; and generating, by the terminal device, a beam measurement result for the “N” CSI-RS resources by applying a first neural network on a result of the measurement on the “M” CSI-RS resources.