AI-Based CSI Compression for Wireless Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing number of antennas in wireless communication systems leads to higher Channel State Information (CSI) feedback overhead, and existing methods struggle to optimize CSI feedback performance due to varying requirements for reference signals in AI training processes.
Innovation Solution
A method is provided where a wireless communication node receives information blocks indicating functions and reference signal associations, allowing for flexible configuration of AI algorithms and parameters to optimize CSI feedback by selecting optimal algorithms based on reference signals, thereby balancing complexity and performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of antennas is increased to improve communication quality through beamforming and multi-antenna processing, then spatial degrees of freedom and communication quality are improved, but CSI feedback overhead increases
Solution Approach 1:
The patent extracts only the essential CSI parameters needed for beamforming and multi-antenna processing, rather than feeding back complete channel state information. By identifying and transmitting only the critical components (such as beam indices, precoding matrix indicators, and rank indicators), the system maintains communication quality while significantly reducing feedback overhead.
Solution Approach 2:
The patent changes the parameter representation of CSI by using compressed indicators (such as CRI, PMI, RI) instead of full channel matrices. This parameter transformation reduces the dimensionality of feedback data from potentially hundreds of complex values to a small set of discrete indices, thereby reducing overhead while preserving the information needed for multi-antenna processing.
2Reliability
If enhanced multi-antenna technologies such as multi-user MIMO are applied to improve communication performance, then feedback accuracy requirements increase, but feedback overhead increases further
Solution Approach 1:
The patent segments the CSI feedback process into multiple stages: channel measurement, channel quantization, and feedback transmission. By dividing the feedback into hierarchical levels (such as wideband CSI and subband CSI, or beam-level CSI and resource-level CSI), the system can provide enhanced accuracy for multi-user MIMO while controlling overall overhead through selective feedback of different segments.
Solution Approach 2:
The patent implements partial feedback by transmitting only the most relevant CSI components for enhanced multi-antenna operations. Instead of providing complete channel state information for all users and all resources, the system selectively feeds back CSI for active users and relevant resource blocks, achieving sufficient accuracy for multi-user MIMO with reduced overhead.
3Measurement precision
If different reference signals with different beamforming requirements are used for AI training, then training performance varies, but system complexity increases due to needing different AI parameters for each reference signal
Solution Approach 1:
The patent develops a universal AI-based CSI compression framework that can handle multiple types of reference signals (such as CSI-RS, SSB, and SRS) with different beamforming requirements using a single set of trained parameters. The universal model learns to adapt to different reference signal characteristics through its architecture design, eliminating the need for separate AI models for each reference signal type and reducing system complexity.
Solution Approach 2:
The patent implements dynamic adaptation within the AI compression framework, where the system can dynamically adjust its processing based on the type of reference signal being used. The AI model incorporates conditional processing that automatically adapts its behavior according to the reference signal characteristics, maintaining optimal compression performance across different beamforming scenarios without requiring separate static models for each case.
Data Source
AI summary
The present application provides a method and device in a node for wireless communications. A first node receives a first information block and a second information block, and transmits a third information block. The first information block indicates a first function; the second information block indicates whether a target reference signal resource is associated with the first function; the third information block indicates a first compressed CSI, and a first pre-compressed CSI is used as an input to the first function to generate the first compressed CSI. The above method can flexibly configure a relation between reference signals and AI algorithms/parameters, and select the optimal AI algorithm/parameter to compress/decompress a CSI based on a certain reference signal, thus optimizing the performance of CSI feedback.